From 070c7fdc3e72f96a4a64656ab19f2c0672df928c Mon Sep 17 00:00:00 2001 From: josch Date: Mon, 3 Nov 2014 20:35:23 +0100 Subject: [PATCH] add 3rdparty/xdelta3-3.0.8 --- 3rdparty/xdelta3-3.0.8/COPYING | 340 ++ 3rdparty/xdelta3-3.0.8/README | 34 + 3rdparty/xdelta3-3.0.8/xdelta3-blkcache.h | 569 +++ 3rdparty/xdelta3-3.0.8/xdelta3-cfgs.h | 173 + 3rdparty/xdelta3-3.0.8/xdelta3-decode.h | 1148 +++++ 3rdparty/xdelta3-3.0.8/xdelta3-djw.h | 1837 +++++++ 3rdparty/xdelta3-3.0.8/xdelta3-fgk.h | 860 ++++ 3rdparty/xdelta3-3.0.8/xdelta3-hash.h | 170 + 3rdparty/xdelta3-3.0.8/xdelta3-internal.h | 145 + 3rdparty/xdelta3-3.0.8/xdelta3-list.h | 130 + 3rdparty/xdelta3-3.0.8/xdelta3-lzma.h | 196 + 3rdparty/xdelta3-3.0.8/xdelta3-main.h | 4048 +++++++++++++++ 3rdparty/xdelta3-3.0.8/xdelta3-merge.h | 579 +++ 3rdparty/xdelta3-3.0.8/xdelta3-second.h | 315 ++ 3rdparty/xdelta3-3.0.8/xdelta3-test.h | 2872 +++++++++++ 3rdparty/xdelta3-3.0.8/xdelta3.c | 5497 +++++++++++++++++++++ 3rdparty/xdelta3-3.0.8/xdelta3.h | 1373 +++++ 17 files changed, 20286 insertions(+) create mode 100644 3rdparty/xdelta3-3.0.8/COPYING create mode 100644 3rdparty/xdelta3-3.0.8/README create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3-blkcache.h create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3-cfgs.h create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3-decode.h create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3-djw.h create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3-fgk.h create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3-hash.h create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3-internal.h create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3-list.h create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3-lzma.h create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3-main.h create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3-merge.h create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3-second.h create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3-test.h create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3.c create mode 100644 3rdparty/xdelta3-3.0.8/xdelta3.h diff --git a/3rdparty/xdelta3-3.0.8/COPYING b/3rdparty/xdelta3-3.0.8/COPYING new file mode 100644 index 0000000..5b6e7c6 --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/COPYING @@ -0,0 +1,340 @@ + GNU GENERAL PUBLIC LICENSE + Version 2, June 1991 + + Copyright (C) 1989, 1991 Free Software Foundation, Inc. + 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + Everyone is permitted to copy and distribute verbatim copies + of this license document, but changing it is not allowed. + + Preamble + + The licenses for most software are designed to take away your +freedom to share and change it. By contrast, the GNU General Public +License is intended to guarantee your freedom to share and change free +software--to make sure the software is free for all its users. This +General Public License applies to most of the Free Software +Foundation's software and to any other program whose authors commit to +using it. (Some other Free Software Foundation software is covered by +the GNU Library General Public License instead.) You can apply it to +your programs, too. + + When we speak of free software, we are referring to freedom, not +price. Our General Public Licenses are designed to make sure that you +have the freedom to distribute copies of free software (and charge for +this service if you wish), that you receive source code or can get it +if you want it, that you can change the software or use pieces of it +in new free programs; and that you know you can do these things. + + To protect your rights, we need to make restrictions that forbid +anyone to deny you these rights or to ask you to surrender the rights. +These restrictions translate to certain responsibilities for you if you +distribute copies of the software, or if you modify it. + + For example, if you distribute copies of such a program, whether +gratis or for a fee, you must give the recipients all the rights that +you have. You must make sure that they, too, receive or can get the +source code. And you must show them these terms so they know their +rights. + + We protect your rights with two steps: (1) copyright the software, and +(2) offer you this license which gives you legal permission to copy, +distribute and/or modify the software. + + Also, for each author's protection and ours, we want to make certain +that everyone understands that there is no warranty for this free +software. If the software is modified by someone else and passed on, we +want its recipients to know that what they have is not the original, so +that any problems introduced by others will not reflect on the original +authors' reputations. + + Finally, any free program is threatened constantly by software +patents. We wish to avoid the danger that redistributors of a free +program will individually obtain patent licenses, in effect making the +program proprietary. To prevent this, we have made it clear that any +patent must be licensed for everyone's free use or not licensed at all. + + The precise terms and conditions for copying, distribution and +modification follow. + + GNU GENERAL PUBLIC LICENSE + TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION + + 0. This License applies to any program or other work which contains +a notice placed by the copyright holder saying it may be distributed +under the terms of this General Public License. The "Program", below, +refers to any such program or work, and a "work based on the Program" +means either the Program or any derivative work under copyright law: +that is to say, a work containing the Program or a portion of it, +either verbatim or with modifications and/or translated into another +language. (Hereinafter, translation is included without limitation in +the term "modification".) Each licensee is addressed as "you". + +Activities other than copying, distribution and modification are not +covered by this License; they are outside its scope. The act of +running the Program is not restricted, and the output from the Program +is covered only if its contents constitute a work based on the +Program (independent of having been made by running the Program). +Whether that is true depends on what the Program does. + + 1. You may copy and distribute verbatim copies of the Program's +source code as you receive it, in any medium, provided that you +conspicuously and appropriately publish on each copy an appropriate +copyright notice and disclaimer of warranty; keep intact all the +notices that refer to this License and to the absence of any warranty; +and give any other recipients of the Program a copy of this License +along with the Program. + +You may charge a fee for the physical act of transferring a copy, and +you may at your option offer warranty protection in exchange for a fee. + + 2. You may modify your copy or copies of the Program or any portion +of it, thus forming a work based on the Program, and copy and +distribute such modifications or work under the terms of Section 1 +above, provided that you also meet all of these conditions: + + a) You must cause the modified files to carry prominent notices + stating that you changed the files and the date of any change. + + b) You must cause any work that you distribute or publish, that in + whole or in part contains or is derived from the Program or any + part thereof, to be licensed as a whole at no charge to all third + parties under the terms of this License. + + c) If the modified program normally reads commands interactively + when run, you must cause it, when started running for such + interactive use in the most ordinary way, to print or display an + announcement including an appropriate copyright notice and a + notice that there is no warranty (or else, saying that you provide + a warranty) and that users may redistribute the program under + these conditions, and telling the user how to view a copy of this + License. (Exception: if the Program itself is interactive but + does not normally print such an announcement, your work based on + the Program is not required to print an announcement.) + +These requirements apply to the modified work as a whole. If +identifiable sections of that work are not derived from the Program, +and can be reasonably considered independent and separate works in +themselves, then this License, and its terms, do not apply to those +sections when you distribute them as separate works. But when you +distribute the same sections as part of a whole which is a work based +on the Program, the distribution of the whole must be on the terms of +this License, whose permissions for other licensees extend to the +entire whole, and thus to each and every part regardless of who wrote it. + +Thus, it is not the intent of this section to claim rights or contest +your rights to work written entirely by you; rather, the intent is to +exercise the right to control the distribution of derivative or +collective works based on the Program. + +In addition, mere aggregation of another work not based on the Program +with the Program (or with a work based on the Program) on a volume of +a storage or distribution medium does not bring the other work under +the scope of this License. + + 3. You may copy and distribute the Program (or a work based on it, +under Section 2) in object code or executable form under the terms of +Sections 1 and 2 above provided that you also do one of the following: + + a) Accompany it with the complete corresponding machine-readable + source code, which must be distributed under the terms of Sections + 1 and 2 above on a medium customarily used for software interchange; or, + + b) Accompany it with a written offer, valid for at least three + years, to give any third party, for a charge no more than your + cost of physically performing source distribution, a complete + machine-readable copy of the corresponding source code, to be + distributed under the terms of Sections 1 and 2 above on a medium + customarily used for software interchange; or, + + c) Accompany it with the information you received as to the offer + to distribute corresponding source code. (This alternative is + allowed only for noncommercial distribution and only if you + received the program in object code or executable form with such + an offer, in accord with Subsection b above.) + +The source code for a work means the preferred form of the work for +making modifications to it. For an executable work, complete source +code means all the source code for all modules it contains, plus any +associated interface definition files, plus the scripts used to +control compilation and installation of the executable. However, as a +special exception, the source code distributed need not include +anything that is normally distributed (in either source or binary +form) with the major components (compiler, kernel, and so on) of the +operating system on which the executable runs, unless that component +itself accompanies the executable. + +If distribution of executable or object code is made by offering +access to copy from a designated place, then offering equivalent +access to copy the source code from the same place counts as +distribution of the source code, even though third parties are not +compelled to copy the source along with the object code. + + 4. You may not copy, modify, sublicense, or distribute the Program +except as expressly provided under this License. Any attempt +otherwise to copy, modify, sublicense or distribute the Program is +void, and will automatically terminate your rights under this License. +However, parties who have received copies, or rights, from you under +this License will not have their licenses terminated so long as such +parties remain in full compliance. + + 5. You are not required to accept this License, since you have not +signed it. However, nothing else grants you permission to modify or +distribute the Program or its derivative works. These actions are +prohibited by law if you do not accept this License. Therefore, by +modifying or distributing the Program (or any work based on the +Program), you indicate your acceptance of this License to do so, and +all its terms and conditions for copying, distributing or modifying +the Program or works based on it. + + 6. Each time you redistribute the Program (or any work based on the +Program), the recipient automatically receives a license from the +original licensor to copy, distribute or modify the Program subject to +these terms and conditions. You may not impose any further +restrictions on the recipients' exercise of the rights granted herein. +You are not responsible for enforcing compliance by third parties to +this License. + + 7. If, as a consequence of a court judgment or allegation of patent +infringement or for any other reason (not limited to patent issues), +conditions are imposed on you (whether by court order, agreement or +otherwise) that contradict the conditions of this License, they do not +excuse you from the conditions of this License. If you cannot +distribute so as to satisfy simultaneously your obligations under this +License and any other pertinent obligations, then as a consequence you +may not distribute the Program at all. For example, if a patent +license would not permit royalty-free redistribution of the Program by +all those who receive copies directly or indirectly through you, then +the only way you could satisfy both it and this License would be to +refrain entirely from distribution of the Program. + +If any portion of this section is held invalid or unenforceable under +any particular circumstance, the balance of the section is intended to +apply and the section as a whole is intended to apply in other +circumstances. + +It is not the purpose of this section to induce you to infringe any +patents or other property right claims or to contest validity of any +such claims; this section has the sole purpose of protecting the +integrity of the free software distribution system, which is +implemented by public license practices. Many people have made +generous contributions to the wide range of software distributed +through that system in reliance on consistent application of that +system; it is up to the author/donor to decide if he or she is willing +to distribute software through any other system and a licensee cannot +impose that choice. + +This section is intended to make thoroughly clear what is believed to +be a consequence of the rest of this License. + + 8. If the distribution and/or use of the Program is restricted in +certain countries either by patents or by copyrighted interfaces, the +original copyright holder who places the Program under this License +may add an explicit geographical distribution limitation excluding +those countries, so that distribution is permitted only in or among +countries not thus excluded. In such case, this License incorporates +the limitation as if written in the body of this License. + + 9. The Free Software Foundation may publish revised and/or new versions +of the General Public License from time to time. Such new versions will +be similar in spirit to the present version, but may differ in detail to +address new problems or concerns. + +Each version is given a distinguishing version number. If the Program +specifies a version number of this License which applies to it and "any +later version", you have the option of following the terms and conditions +either of that version or of any later version published by the Free +Software Foundation. If the Program does not specify a version number of +this License, you may choose any version ever published by the Free Software +Foundation. + + 10. If you wish to incorporate parts of the Program into other free +programs whose distribution conditions are different, write to the author +to ask for permission. For software which is copyrighted by the Free +Software Foundation, write to the Free Software Foundation; we sometimes +make exceptions for this. Our decision will be guided by the two goals +of preserving the free status of all derivatives of our free software and +of promoting the sharing and reuse of software generally. + + NO WARRANTY + + 11. BECAUSE THE PROGRAM IS LICENSED FREE OF CHARGE, THERE IS NO WARRANTY +FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW. EXCEPT WHEN +OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES +PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED +OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF +MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS +TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU. SHOULD THE +PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING, +REPAIR OR CORRECTION. + + 12. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING +WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY AND/OR +REDISTRIBUTE THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, +INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING +OUT OF THE USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED +TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY +YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER +PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE +POSSIBILITY OF SUCH DAMAGES. + + END OF TERMS AND CONDITIONS + + How to Apply These Terms to Your New Programs + + If you develop a new program, and you want it to be of the greatest +possible use to the public, the best way to achieve this is to make it +free software which everyone can redistribute and change under these terms. + + To do so, attach the following notices to the program. It is safest +to attach them to the start of each source file to most effectively +convey the exclusion of warranty; and each file should have at least +the "copyright" line and a pointer to where the full notice is found. + + + Copyright (C) + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + + +Also add information on how to contact you by electronic and paper mail. + +If the program is interactive, make it output a short notice like this +when it starts in an interactive mode: + + Gnomovision version 69, Copyright (C) year name of author + Gnomovision comes with ABSOLUTELY NO WARRANTY; for details type `show w'. + This is free software, and you are welcome to redistribute it + under certain conditions; type `show c' for details. + +The hypothetical commands `show w' and `show c' should show the appropriate +parts of the General Public License. Of course, the commands you use may +be called something other than `show w' and `show c'; they could even be +mouse-clicks or menu items--whatever suits your program. + +You should also get your employer (if you work as a programmer) or your +school, if any, to sign a "copyright disclaimer" for the program, if +necessary. Here is a sample; alter the names: + + Yoyodyne, Inc., hereby disclaims all copyright interest in the program + `Gnomovision' (which makes passes at compilers) written by James Hacker. + + , 1 April 1989 + Ty Coon, President of Vice + +This General Public License does not permit incorporating your program into +proprietary programs. If your program is a subroutine library, you may +consider it more useful to permit linking proprietary applications with the +library. If this is what you want to do, use the GNU Library General +Public License instead of this License. diff --git a/3rdparty/xdelta3-3.0.8/README b/3rdparty/xdelta3-3.0.8/README new file mode 100644 index 0000000..be7c6ce --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/README @@ -0,0 +1,34 @@ +Xdelta 3.x readme.txt +Copyright (C) 2001, 2002, 2003, 2004, 2005, 2006, 2007 + + + +Thanks for downloading Xdelta! + +This directory contains the Xdelta3 command-line interface (CLI) and source +distribution for VCDIFF differential compression, a.k.a. delta +compression. The latest information and downloads are available here: + + http://xdelta.org/ + http://code.google.com/p/xdelta/ + +The command-line syntax: + + http://code.google.com/p/xdelta/wiki/CommandLineSyntax + +Run 'xdelta3 -h' for brief help. Run 'xdelta3 test' for built-in tests. + +Sample commands (like gzip, -e means encode, -d means decode) + + xdelta3 -9 -S djw -e -vfs OLD_FILE NEW_FILE DELTA_FILE + xdelta3 -d -vfs OLD_FILE DELTA_FILE DECODED_FILE + +File bug reports and browse open support issues here: + + http://code.google.com/p/xdelta/issues/list + +The source distribution contains the C/C++/Python APIs, Unix, Microsoft VC++ +and Cygwin builds. Xdelta3 is covered under the terms of the GPL, see +COPYING. + +Commercial inquiries welcome, please contact diff --git a/3rdparty/xdelta3-3.0.8/xdelta3-blkcache.h b/3rdparty/xdelta3-3.0.8/xdelta3-blkcache.h new file mode 100644 index 0000000..82d720e --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3-blkcache.h @@ -0,0 +1,569 @@ +/* xdelta 3 - delta compression tools and library + * Copyright (C) 2001, 2002, 2003, 2004, 2005, 2006, 2007, + * 2008, 2009, 2010 + * Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +/* TODO: This code is heavily revised from 3.0z but still needs major + * refactoring. */ + +#include "xdelta3-internal.h" + +typedef struct _main_blklru main_blklru; +typedef struct _main_blklru_list main_blklru_list; + +struct _main_blklru_list +{ + main_blklru_list *next; + main_blklru_list *prev; +}; + +struct _main_blklru +{ + uint8_t *blk; + xoff_t blkno; + usize_t size; + main_blklru_list link; +}; + +#define MAX_LRU_SIZE 32U +#define XD3_MINSRCWINSZ (XD3_ALLOCSIZE * MAX_LRU_SIZE) +#define XD3_MAXSRCWINSZ (1ULL << 31) + +XD3_MAKELIST(main_blklru_list,main_blklru,link); + +static usize_t lru_size = 0; +static main_blklru *lru = NULL; /* array of lru_size elts */ +static main_blklru_list lru_list; +static int do_src_fifo = 0; /* set to avoid lru */ + +static int lru_hits = 0; +static int lru_misses = 0; +static int lru_filled = 0; + +static void main_lru_reset (void) +{ + lru_size = 0; + lru = NULL; + do_src_fifo = 0; + lru_hits = 0; + lru_misses = 0; + lru_filled = 0; +} + +static void main_lru_cleanup (void) +{ + if (lru != NULL) + { + main_buffree (lru[0].blk); + } + + main_free (lru); + lru = NULL; + + lru_hits = 0; + lru_misses = 0; + lru_filled = 0; +} + +/* This is called at different times for encoding and decoding. The + * encoder calls it immediately, the decoder delays until the + * application header is received. */ +static int +main_set_source (xd3_stream *stream, xd3_cmd cmd, + main_file *sfile, xd3_source *source) +{ + int ret = 0; + usize_t i; + xoff_t source_size = 0; + usize_t blksize; + + XD3_ASSERT (lru == NULL); + XD3_ASSERT (stream->src == NULL); + XD3_ASSERT (option_srcwinsz >= XD3_MINSRCWINSZ); + + /* TODO: this code needs refactoring into FIFO, LRU, FAKE. Yuck! + * This is simplified from 3.0z which had issues with sizing the + * source buffer memory allocation and the source blocksize. */ + + /* LRU-specific */ + main_blklru_list_init (& lru_list); + + if (allow_fake_source) + { + /* TODO: refactor + * TOOLS/recode-specific: Check "allow_fake_source" mode looks + * broken now. */ + sfile->mode = XO_READ; + sfile->realname = sfile->filename; + sfile->nread = 0; + } + else + { + /* Either a regular file (possibly compressed) or a FIFO + * (possibly compressed). */ + if ((ret = main_file_open (sfile, sfile->filename, XO_READ))) + { + return ret; + } + + /* If the file is regular we know it's size. If the file turns + * out to be externally compressed, size_known may change. */ + sfile->size_known = (main_file_stat (sfile, &source_size) == 0); + } + + /* Note: The API requires a power-of-two blocksize and srcwinsz + * (-B). The logic here will use a single block if the entire file + * is known to fit into srcwinsz. */ + option_srcwinsz = xd3_pow2_roundup (option_srcwinsz); + + /* Though called "lru", it is not LRU-specific. We always allocate + * a maximum number of source block buffers. If the entire file + * fits into srcwinsz, this buffer will stay as the only + * (lru_size==1) source block. Otherwise, we know that at least + * option_srcwinsz bytes are available. Split the source window + * into buffers. */ + if ((lru = (main_blklru*) main_malloc (MAX_LRU_SIZE * + sizeof (main_blklru))) == NULL) + { + ret = ENOMEM; + return ret; + } + + memset (lru, 0, sizeof(lru[0]) * MAX_LRU_SIZE); + + /* Allocate the entire buffer. */ + if ((lru[0].blk = (uint8_t*) main_bufalloc (option_srcwinsz)) == NULL) + { + ret = ENOMEM; + return ret; + } + + /* Main calls main_getblk_func() once before xd3_set_source(). This + * is the point at which external decompression may begin. Set the + * system for a single block. */ + lru_size = 1; + lru[0].blkno = (xoff_t) -1; + blksize = option_srcwinsz; + main_blklru_list_push_back (& lru_list, & lru[0]); + XD3_ASSERT (blksize != 0); + + /* Initialize xd3_source. */ + source->blksize = blksize; + source->name = sfile->filename; + source->ioh = sfile; + source->curblkno = (xoff_t) -1; + source->curblk = NULL; + source->max_winsize = option_srcwinsz; + + if ((ret = main_getblk_func (stream, source, 0)) != 0) + { + XPR(NT "error reading source: %s: %s\n", + sfile->filename, + xd3_mainerror (ret)); + return ret; + } + + source->onblk = lru[0].size; /* xd3 sets onblk */ + + /* If the file is smaller than a block, size is known. */ + if (!sfile->size_known && source->onblk < blksize) + { + source_size = source->onblk; + sfile->size_known = 1; + } + + /* If the size is not known or is greater than the buffer size, we + * split the buffer across MAX_LRU_SIZE blocks (already allocated in + * "lru"). */ + if (!sfile->size_known || source_size > option_srcwinsz) + { + /* Modify block 0, change blocksize. */ + blksize = option_srcwinsz / MAX_LRU_SIZE; + source->blksize = blksize; + source->onblk = blksize; /* xd3 sets onblk */ + /* Note: source->max_winsize is unchanged. */ + lru[0].size = blksize; + lru_size = MAX_LRU_SIZE; + + /* Setup rest of blocks. */ + for (i = 1; i < lru_size; i += 1) + { + lru[i].blk = lru[0].blk + (blksize * i); + lru[i].blkno = i; + lru[i].size = blksize; + main_blklru_list_push_back (& lru_list, & lru[i]); + } + } + + if (! sfile->size_known) + { + /* If the size is not know, we must use FIFO discipline. */ + do_src_fifo = 1; + } + + /* Call the appropriate set_source method, handle errors, print + * verbose message, etc. */ + if (sfile->size_known) + { + ret = xd3_set_source_and_size (stream, source, source_size); + } + else + { + ret = xd3_set_source (stream, source); + } + + if (ret) + { + XPR(NT XD3_LIB_ERRMSG (stream, ret)); + return ret; + } + + XD3_ASSERT (stream->src == source); + XD3_ASSERT (source->blksize == blksize); + + if (option_verbose) + { + static shortbuf srcszbuf; + static shortbuf srccntbuf; + static shortbuf winszbuf; + static shortbuf blkszbuf; + static shortbuf nbufs; + + if (sfile->size_known) + { + short_sprintf (srcszbuf, "source size %s [%"Q"u]", + main_format_bcnt (source_size, &srccntbuf), + source_size); + } + else + { + short_sprintf (srcszbuf, "%s", "source size unknown"); + } + + nbufs.buf[0] = 0; + + if (option_verbose > 1) + { + short_sprintf (nbufs, " #bufs %u", lru_size); + } + + XPR(NT "source %s %s blksize %s window %s%s%s\n", + sfile->filename, + srcszbuf.buf, + main_format_bcnt (blksize, &blkszbuf), + main_format_bcnt (option_srcwinsz, &winszbuf), + nbufs.buf, + do_src_fifo ? " (FIFO)" : ""); + } + + return 0; +} + +static int +main_getblk_lru (xd3_source *source, xoff_t blkno, + main_blklru** blrup, int *is_new) +{ + main_blklru *blru = NULL; + usize_t i; + + (*is_new) = 0; + + if (do_src_fifo) + { + /* Direct lookup assumes sequential scan w/o skipping blocks. */ + int idx = blkno % lru_size; + blru = & lru[idx]; + if (blru->blkno == blkno) + { + (*blrup) = blru; + return 0; + } + /* No going backwards in a sequential scan. */ + if (blru->blkno != (xoff_t) -1 && blru->blkno > blkno) + { + return XD3_TOOFARBACK; + } + } + else + { + /* Sequential search through LRU. */ + for (i = 0; i < lru_size; i += 1) + { + blru = & lru[i]; + if (blru->blkno == blkno) + { + main_blklru_list_remove (blru); + main_blklru_list_push_back (& lru_list, blru); + (*blrup) = blru; + return 0; + } + } + } + + if (do_src_fifo) + { + int idx = blkno % lru_size; + blru = & lru[idx]; + } + else + { + XD3_ASSERT (! main_blklru_list_empty (& lru_list)); + blru = main_blklru_list_pop_front (& lru_list); + main_blklru_list_push_back (& lru_list, blru); + } + + lru_filled += 1; + (*is_new) = 1; + (*blrup) = blru; + blru->blkno = -1; + return 0; +} + +static int +main_read_seek_source (xd3_stream *stream, + xd3_source *source, + xoff_t blkno) { + xoff_t pos = blkno * source->blksize; + main_file *sfile = (main_file*) source->ioh; + main_blklru *blru; + int is_new; + size_t nread = 0; + int ret = 0; + + if (!sfile->seek_failed) + { + ret = main_file_seek (sfile, pos); + + if (ret == 0) + { + sfile->source_position = pos; + } + } + + if (sfile->seek_failed || ret != 0) + { + /* For an unseekable file (or other seek error, does it + * matter?) */ + if (sfile->source_position > pos) + { + /* Could assert !IS_ENCODE(), this shouldn't happen + * because of do_src_fifo during encode. */ + if (!option_quiet) + { + XPR(NT "source can't seek backwards; requested block offset " + "%"Q"u source position is %"Q"u\n", + pos, sfile->source_position); + } + + sfile->seek_failed = 1; + stream->msg = "non-seekable source: " + "copy is too far back (try raising -B)"; + return XD3_TOOFARBACK; + } + + /* There's a chance here, that an genuine lseek error will cause + * xdelta3 to shift into non-seekable mode, entering a degraded + * condition. */ + if (!sfile->seek_failed && option_verbose) + { + XPR(NT "source can't seek, will use FIFO for %s\n", + sfile->filename); + + if (option_verbose > 1) + { + XPR(NT "seek error at offset %"Q"u: %s\n", + pos, xd3_mainerror (ret)); + } + } + + sfile->seek_failed = 1; + + if (option_verbose > 1 && pos != sfile->source_position) + { + XPR(NT "non-seekable source skipping %"Q"u bytes @ %"Q"u\n", + pos - sfile->source_position, + sfile->source_position); + } + + while (sfile->source_position < pos) + { + xoff_t skip_blkno; + usize_t skip_offset; + + xd3_blksize_div (sfile->source_position, source, + &skip_blkno, &skip_offset); + + /* Read past unused data */ + XD3_ASSERT (pos - sfile->source_position >= source->blksize); + XD3_ASSERT (skip_offset == 0); + + if ((ret = main_getblk_lru (source, skip_blkno, + & blru, & is_new))) + { + return ret; + } + + XD3_ASSERT (is_new); + blru->blkno = skip_blkno; + + if ((ret = main_read_primary_input (sfile, + (uint8_t*) blru->blk, + source->blksize, + & nread))) + { + return ret; + } + + if (nread != source->blksize) + { + IF_DEBUG1 (DP(RINT "[getblk] short skip block nread = %zu\n", + nread)); + stream->msg = "non-seekable input is short"; + return XD3_INVALID_INPUT; + } + + sfile->source_position += nread; + blru->size = nread; + + IF_DEBUG1 (DP(RINT "[getblk] skip blkno %"Q"u size %u\n", + skip_blkno, blru->size)); + + XD3_ASSERT (sfile->source_position <= pos); + } + } + + return 0; +} + +/* This is the callback for reading a block of source. This function + * is blocking and it implements a small LRU. + * + * Note that it is possible for main_input() to handle getblk requests + * in a non-blocking manner. If the callback is NULL then the caller + * of xd3_*_input() must handle the XD3_GETSRCBLK return value and + * fill the source in the same way. See xd3_getblk for details. To + * see an example of non-blocking getblk, see xdelta-test.h. */ +static int +main_getblk_func (xd3_stream *stream, + xd3_source *source, + xoff_t blkno) +{ + int ret = 0; + xoff_t pos = blkno * source->blksize; + main_file *sfile = (main_file*) source->ioh; + main_blklru *blru; + int is_new; + int did_seek = 0; + size_t nread = 0; + + if (allow_fake_source) + { + source->curblkno = blkno; + source->onblk = 0; + source->curblk = lru[0].blk; + lru[0].size = 0; + return 0; + } + + if ((ret = main_getblk_lru (source, blkno, & blru, & is_new))) + { + return ret; + } + + if (!is_new) + { + source->curblkno = blkno; + source->onblk = blru->size; + source->curblk = blru->blk; + lru_hits++; + return 0; + } + + lru_misses += 1; + + if (pos != sfile->source_position) + { + /* Only try to seek when the position is wrong. This means the + * decoder will fail when the source buffer is too small, but + * only when the input is non-seekable. */ + if ((ret = main_read_seek_source (stream, source, blkno))) + { + return ret; + } + + /* Indicates that another call to main_getblk_lru() may be + * needed */ + did_seek = 1; + } + + XD3_ASSERT (sfile->source_position == pos); + + if (did_seek && + (ret = main_getblk_lru (source, blkno, & blru, & is_new))) + { + return ret; + } + + if ((ret = main_read_primary_input (sfile, + (uint8_t*) blru->blk, + source->blksize, + & nread))) + { + return ret; + } + + /* Save the last block read, used to handle non-seekable files. */ + sfile->source_position = pos + nread; + + if (option_verbose > 3) + { + if (blru->blkno != (xoff_t)-1) + { + if (blru->blkno != blkno) + { + XPR(NT "source block %"Q"u read %zu ejects %"Q"u (lru_hits=%u, " + "lru_misses=%u, lru_filled=%u)\n", + blkno, nread, blru->blkno, lru_hits, lru_misses, lru_filled); + } + else + { + XPR(NT "source block %"Q"u read %zu (lru_hits=%u, " + "lru_misses=%u, lru_filled=%u)\n", + blkno, nread, lru_hits, lru_misses, lru_filled); + } + } + else + { + XPR(NT "source block %"Q"u read %zu (lru_hits=%u, lru_misses=%u, " + "lru_filled=%u)\n", blkno, nread, + lru_hits, lru_misses, lru_filled); + } + } + + source->curblk = blru->blk; + source->curblkno = blkno; + source->onblk = nread; + blru->size = nread; + blru->blkno = blkno; + + IF_DEBUG1 (DP(RINT "[main_getblk] blkno %"Q"u onblk %zu pos %"Q"u " + "srcpos %"Q"u\n", + blkno, nread, pos, sfile->source_position)); + + return 0; +} diff --git a/3rdparty/xdelta3-3.0.8/xdelta3-cfgs.h b/3rdparty/xdelta3-3.0.8/xdelta3-cfgs.h new file mode 100644 index 0000000..b13f7b0 --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3-cfgs.h @@ -0,0 +1,173 @@ +/* xdelta 3 - delta compression tools and library + * Copyright (C) 2001, 2002, 2003, 2004, 2005, 2006, 2007. Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +/****************************************************************** + SOFT string matcher + ******************************************************************/ + +#if XD3_BUILD_SOFT + +#define TEMPLATE soft +#define LLOOK stream->smatcher.large_look +#define LSTEP stream->smatcher.large_step +#define SLOOK stream->smatcher.small_look +#define SCHAIN stream->smatcher.small_chain +#define SLCHAIN stream->smatcher.small_lchain +#define MAXLAZY stream->smatcher.max_lazy +#define LONGENOUGH stream->smatcher.long_enough + +#define SOFTCFG 1 +#include "xdelta3.c" +#undef SOFTCFG + +#undef TEMPLATE +#undef LLOOK +#undef SLOOK +#undef LSTEP +#undef SCHAIN +#undef SLCHAIN +#undef MAXLAZY +#undef LONGENOUGH +#endif + +#define SOFTCFG 0 + +/************************************************************ + FASTEST string matcher + **********************************************************/ +#if XD3_BUILD_FASTEST +#define TEMPLATE fastest +#define LLOOK 9 +#define LSTEP 26 +#define SLOOK 4U +#define SCHAIN 1 +#define SLCHAIN 1 +#define MAXLAZY 6 +#define LONGENOUGH 6 + +#include "xdelta3.c" + +#undef TEMPLATE +#undef LLOOK +#undef SLOOK +#undef LSTEP +#undef SCHAIN +#undef SLCHAIN +#undef MAXLAZY +#undef LONGENOUGH +#endif + +/************************************************************ + FASTER string matcher + **********************************************************/ +#if XD3_BUILD_FASTER +#define TEMPLATE faster +#define LLOOK 9 +#define LSTEP 15 +#define SLOOK 4U +#define SCHAIN 1 +#define SLCHAIN 1 +#define MAXLAZY 18 +#define LONGENOUGH 18 + +#include "xdelta3.c" + +#undef TEMPLATE +#undef LLOOK +#undef SLOOK +#undef LSTEP +#undef SCHAIN +#undef SLCHAIN +#undef MAXLAZY +#undef LONGENOUGH +#endif + +/****************************************************** + FAST string matcher + ********************************************************/ +#if XD3_BUILD_FAST +#define TEMPLATE fast +#define LLOOK 9 +#define LSTEP 8 +#define SLOOK 4U +#define SCHAIN 4 +#define SLCHAIN 1 +#define MAXLAZY 18 +#define LONGENOUGH 35 + +#include "xdelta3.c" + +#undef TEMPLATE +#undef LLOOK +#undef SLOOK +#undef LSTEP +#undef SCHAIN +#undef SLCHAIN +#undef MAXLAZY +#undef LONGENOUGH +#endif + +/************************************************** + SLOW string matcher + **************************************************************/ +#if XD3_BUILD_SLOW +#define TEMPLATE slow +#define LLOOK 9 +#define LSTEP 2 +#define SLOOK 4U +#define SCHAIN 44 +#define SLCHAIN 13 +#define MAXLAZY 90 +#define LONGENOUGH 70 + +#include "xdelta3.c" + +#undef TEMPLATE +#undef LLOOK +#undef SLOOK +#undef LSTEP +#undef SCHAIN +#undef SLCHAIN +#undef MAXLAZY +#undef LONGENOUGH +#endif + +/******************************************************** + DEFAULT string matcher + ************************************************************/ +#if XD3_BUILD_DEFAULT +#define TEMPLATE default +#define LLOOK 9 +#define LSTEP 3 +#define SLOOK 4U +#define SCHAIN 8 +#define SLCHAIN 2 +#define MAXLAZY 36 +#define LONGENOUGH 70 + +#include "xdelta3.c" + +#undef TEMPLATE +#undef LLOOK +#undef SLOOK +#undef LSTEP +#undef SCHAIN +#undef SLCHAIN +#undef MAXLAZY +#undef LONGENOUGH +#endif diff --git a/3rdparty/xdelta3-3.0.8/xdelta3-decode.h b/3rdparty/xdelta3-3.0.8/xdelta3-decode.h new file mode 100644 index 0000000..c0cb578 --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3-decode.h @@ -0,0 +1,1148 @@ +/* xdelta 3 - delta compression tools and library + * Copyright (C) 2002, 2003, 2004, 2005, 2006, 2007. Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +#ifndef _XDELTA3_DECODE_H_ +#define _XDELTA3_DECODE_H_ + +#include "xdelta3-internal.h" + +#define SRCORTGT(x) ((((x) & VCD_SRCORTGT) == VCD_SOURCE) ? \ + VCD_SOURCE : ((((x) & VCD_SRCORTGT) == \ + VCD_TARGET) ? VCD_TARGET : 0)) + +/* Initialize the decoder for a new window. The dec_tgtlen value is + * preserved across successive window decodings, and the update to + * dec_winstart is delayed until a new window actually starts. This + * is to avoid throwing an error due to overflow until the last + * possible moment. This makes it possible to encode exactly 4GB + * through a 32-bit encoder. */ +static int +xd3_decode_init_window (xd3_stream *stream) +{ + stream->dec_cpylen = 0; + stream->dec_cpyoff = 0; + stream->dec_cksumbytes = 0; + + xd3_init_cache (& stream->acache); + + return 0; +} + +/* Allocates buffer space for the target window and possibly the + * VCD_TARGET copy-window. Also sets the base of the two copy + * segments. */ +static int +xd3_decode_setup_buffers (xd3_stream *stream) +{ + /* If VCD_TARGET is set then the previous buffer may be reused. */ + if (stream->dec_win_ind & VCD_TARGET) + { + /* But this implementation only supports copying from the last + * target window. If the offset is outside that range, it can't + * be done. */ + if (stream->dec_cpyoff < stream->dec_laststart) + { + stream->msg = "unsupported VCD_TARGET offset"; + return XD3_INVALID_INPUT; + } + + /* See if the two windows are the same. This indicates the + * first time VCD_TARGET is used. This causes a second buffer + * to be allocated, after that the two are swapped in the + * DEC_FINISH case. */ + if (stream->dec_lastwin == stream->next_out) + { + stream->next_out = NULL; + stream->space_out = 0; + } + + // TODO: VCD_TARGET mode, this is broken + stream->dec_cpyaddrbase = stream->dec_lastwin + + (usize_t) (stream->dec_cpyoff - stream->dec_laststart); + } + + /* See if the current output window is large enough. */ + if (stream->space_out < stream->dec_tgtlen) + { + xd3_free (stream, stream->dec_buffer); + + stream->space_out = + xd3_round_blksize (stream->dec_tgtlen, XD3_ALLOCSIZE); + + if ((stream->dec_buffer = + (uint8_t*) xd3_alloc (stream, stream->space_out, 1)) == NULL) + { + return ENOMEM; + } + + stream->next_out = stream->dec_buffer; + } + + /* dec_tgtaddrbase refers to an invalid base address, but it is + * always used with a sufficiently large instruction offset (i.e., + * beyond the copy window). This condition is enforced by + * xd3_decode_output_halfinst. */ + stream->dec_tgtaddrbase = stream->next_out - stream->dec_cpylen; + + return 0; +} + +static int +xd3_decode_allocate (xd3_stream *stream, + usize_t size, + uint8_t **buf_ptr, + usize_t *buf_alloc) +{ + if (*buf_ptr != NULL && *buf_alloc < size) + { + xd3_free (stream, *buf_ptr); + *buf_ptr = NULL; + } + + if (*buf_ptr == NULL) + { + *buf_alloc = xd3_round_blksize (size, XD3_ALLOCSIZE); + + if ((*buf_ptr = (uint8_t*) xd3_alloc (stream, *buf_alloc, 1)) == NULL) + { + return ENOMEM; + } + } + + return 0; +} + +static int +xd3_decode_section (xd3_stream *stream, + xd3_desect *section, + xd3_decode_state nstate, + int copy) +{ + XD3_ASSERT (section->pos <= section->size); + XD3_ASSERT (stream->dec_state != nstate); + + if (section->pos < section->size) + { + usize_t sect_take; + + if (stream->avail_in == 0) + { + return XD3_INPUT; + } + + if ((copy == 0) && (section->pos == 0)) + { + /* No allocation/copy needed */ + section->buf = stream->next_in; + sect_take = section->size; + } + else + { + usize_t sect_need = section->size - section->pos; + + /* Allocate and copy */ + sect_take = min (sect_need, stream->avail_in); + + if (section->pos == 0) + { + int ret; + + if ((ret = xd3_decode_allocate (stream, + section->size, + & section->copied1, + & section->alloc1))) + { + return ret; + } + + section->buf = section->copied1; + } + + memcpy (section->copied1 + section->pos, + stream->next_in, + sect_take); + } + + section->pos += sect_take; + + stream->dec_winbytes += sect_take; + + DECODE_INPUT (sect_take); + } + + if (section->pos < section->size) + { + stream->msg = "further input required"; + return XD3_INPUT; + } + + XD3_ASSERT (section->pos == section->size); + + stream->dec_state = nstate; + section->buf_max = section->buf + section->size; + section->pos = 0; + return 0; +} + +/* Decode the size and address for half of an instruction (i.e., a + * single opcode). This updates the stream->dec_position, which are + * bytes already output prior to processing this instruction. Perform + * bounds checking for sizes and copy addresses, which uses the + * dec_position (which is why these checks are done here). */ +static int +xd3_decode_parse_halfinst (xd3_stream *stream, xd3_hinst *inst) +{ + int ret; + + /* If the size from the instruction table is zero then read a size value. */ + if ((inst->size == 0) && + (ret = xd3_read_size (stream, + & stream->inst_sect.buf, + stream->inst_sect.buf_max, + & inst->size))) + { + return XD3_INVALID_INPUT; + } + + /* For copy instructions, read address. */ + if (inst->type >= XD3_CPY) + { + IF_DEBUG2 ({ + static int cnt = 0; + XPR(NT "DECODE:%u: COPY at %"Q"u (winoffset %u) size %u winaddr %u\n", + cnt++, + stream->total_out + (stream->dec_position - + stream->dec_cpylen), + (stream->dec_position - stream->dec_cpylen), + inst->size, + inst->addr); + }); + + if ((ret = xd3_decode_address (stream, + stream->dec_position, + inst->type - XD3_CPY, + & stream->addr_sect.buf, + stream->addr_sect.buf_max, + & inst->addr))) + { + return ret; + } + + /* Cannot copy an address before it is filled-in. */ + if (inst->addr >= stream->dec_position) + { + stream->msg = "address too large"; + return XD3_INVALID_INPUT; + } + + /* Check: a VCD_TARGET or VCD_SOURCE copy cannot exceed the remaining + * buffer space in its own segment. */ + if (inst->addr < stream->dec_cpylen && + inst->addr + inst->size > stream->dec_cpylen) + { + stream->msg = "size too large"; + return XD3_INVALID_INPUT; + } + } + else + { + IF_DEBUG2 ({ + if (inst->type == XD3_ADD) + { + static int cnt; + XPR(NT "DECODE:%d: ADD at %"Q"u (winoffset %u) size %u\n", + cnt++, + (stream->total_out + stream->dec_position - stream->dec_cpylen), + stream->dec_position - stream->dec_cpylen, + inst->size); + } + else + { + static int cnt; + XD3_ASSERT (inst->type == XD3_RUN); + XPR(NT "DECODE:%d: RUN at %"Q"u (winoffset %u) size %u\n", + cnt++, + stream->total_out + stream->dec_position - stream->dec_cpylen, + stream->dec_position - stream->dec_cpylen, + inst->size); + } + }); + } + + /* Check: The instruction will not overflow the output buffer. */ + if (stream->dec_position + inst->size > stream->dec_maxpos) + { + stream->msg = "size too large"; + return XD3_INVALID_INPUT; + } + + stream->dec_position += inst->size; + return 0; +} + +/* Decode a single opcode and then decode the two half-instructions. */ +static int +xd3_decode_instruction (xd3_stream *stream) +{ + int ret; + const xd3_dinst *inst; + + if (stream->inst_sect.buf == stream->inst_sect.buf_max) + { + stream->msg = "instruction underflow"; + return XD3_INVALID_INPUT; + } + + inst = &stream->code_table[*stream->inst_sect.buf++]; + + stream->dec_current1.type = inst->type1; + stream->dec_current2.type = inst->type2; + stream->dec_current1.size = inst->size1; + stream->dec_current2.size = inst->size2; + + /* For each instruction with a real operation, decode the + * corresponding size and addresses if necessary. Assume a + * code-table may have NOOP in either position, although this is + * unlikely. */ + if (inst->type1 != XD3_NOOP && + (ret = xd3_decode_parse_halfinst (stream, & stream->dec_current1))) + { + return ret; + } + if (inst->type2 != XD3_NOOP && + (ret = xd3_decode_parse_halfinst (stream, & stream->dec_current2))) + { + return ret; + } + return 0; +} + +/* Output the result of a single half-instruction. OPT: This the + decoder hotspot. Modifies "hinst", see below. */ +static int +xd3_decode_output_halfinst (xd3_stream *stream, xd3_hinst *inst) +{ + /* This method is reentrant for copy instructions which may return + * XD3_GETSRCBLK to the caller. Each time through a copy takes the + * minimum of inst->size and the available space on whichever block + * supplies the data */ + usize_t take = inst->size; + + XD3_ASSERT (inst->type != XD3_NOOP); + + switch (inst->type) + { + case XD3_RUN: + { + /* Only require a single data byte. */ + if (stream->data_sect.buf == stream->data_sect.buf_max) + { + stream->msg = "data underflow"; + return XD3_INVALID_INPUT; + } + + memset (stream->next_out + stream->avail_out, + stream->data_sect.buf[0], + take); + + stream->data_sect.buf += 1; + stream->avail_out += take; + inst->type = XD3_NOOP; + break; + } + case XD3_ADD: + { + /* Require at least TAKE data bytes. */ + if (stream->data_sect.buf + take > stream->data_sect.buf_max) + { + stream->msg = "data underflow"; + return XD3_INVALID_INPUT; + } + + memcpy (stream->next_out + stream->avail_out, + stream->data_sect.buf, + take); + + stream->data_sect.buf += take; + stream->avail_out += take; + inst->type = XD3_NOOP; + break; + } + default: + { + usize_t i; + const uint8_t *src; + uint8_t *dst; + int overlap; + + /* See if it copies from the VCD_TARGET/VCD_SOURCE window or + * the target window. Out-of-bounds checks for the addresses + * and sizes are performed in xd3_decode_parse_halfinst. This + * if/else must set "overlap", "src", and "dst". */ + if (inst->addr < stream->dec_cpylen) + { + /* In both branches we are copying from outside the + * current decoder window, the first (VCD_TARGET) is + * unimplemented. */ + overlap = 0; + + /* This branch sets "src". As a side-effect, we modify + * "inst" so that if we reenter this method after a + * XD3_GETSRCBLK response the state is correct. So if the + * instruction can be fulfilled by a contiguous block of + * memory then we will set: + * + * inst->type = XD3_NOOP; + * inst->size = 0; + */ + if (stream->dec_win_ind & VCD_TARGET) + { + /* TODO: Users have requested long-distance copies of + * similar material within a target (e.g., for dup + * supression in backups). */ + inst->size = 0; + inst->type = XD3_NOOP; + stream->msg = "VCD_TARGET not implemented"; + return XD3_UNIMPLEMENTED; + } + else + { + /* In this case we have to read a source block, which + * could return control to the caller. We need to + * know the first block number needed for this + * copy. */ + xd3_source *source = stream->src; + xoff_t block = source->cpyoff_blocks; + usize_t blkoff = source->cpyoff_blkoff; + const usize_t blksize = source->blksize; + int ret; + + xd3_blksize_add (&block, &blkoff, source, inst->addr); + XD3_ASSERT (blkoff < blksize); + + if ((ret = xd3_getblk (stream, block))) + { + /* could be a XD3_GETSRCBLK failure. */ + if (ret == XD3_TOOFARBACK) + { + stream->msg = "non-seekable source in decode"; + ret = XD3_INTERNAL; + } + return ret; + } + + src = source->curblk + blkoff; + + /* This block is either full, or a partial block that + * must contain enough bytes. */ + if ((source->onblk != blksize) && + (blkoff + take > source->onblk)) + { + IF_DEBUG1 (XPR(NT "[srcfile] short at blkno %"Q"u onblk " + "%u blksize %u blkoff %u take %u\n", + block, + source->onblk, + blksize, + blkoff, + take)); + stream->msg = "source file too short"; + return XD3_INVALID_INPUT; + } + + XD3_ASSERT (blkoff != blksize); + + /* Check if we have enough data on this block to + * finish the instruction. */ + if (blkoff + take <= blksize) + { + inst->type = XD3_NOOP; + inst->size = 0; + } + else + { + take = blksize - blkoff; + inst->size -= take; + inst->addr += take; + + /* because (blkoff + take > blksize), above */ + XD3_ASSERT (inst->size != 0); + } + } + } + else + { + /* TODO: the memcpy/overlap optimization, etc. Overlap + * here could be more specific, it's whether (inst->addr - + * srclen) + inst->size > input_pos ? And is the system + * memcpy really any good? */ + overlap = 1; + + /* For a target-window copy, we know the entire range is + * in-memory. The dec_tgtaddrbase is negatively offset by + * dec_cpylen because the addresses start beyond that + * point. */ + src = stream->dec_tgtaddrbase + inst->addr; + inst->type = XD3_NOOP; + inst->size = 0; + } + + dst = stream->next_out + stream->avail_out; + + stream->avail_out += take; + + if (overlap) + { + /* Can't just memcpy here due to possible overlap. */ + for (i = take; i != 0; i -= 1) + { + *dst++ = *src++; + } + } + else + { + memcpy (dst, src, take); + } + } + } + + return 0; +} + +static int +xd3_decode_finish_window (xd3_stream *stream) +{ + stream->dec_winbytes = 0; + stream->dec_state = DEC_FINISH; + + stream->data_sect.pos = 0; + stream->inst_sect.pos = 0; + stream->addr_sect.pos = 0; + + return XD3_OUTPUT; +} + +static int +xd3_decode_secondary_sections (xd3_stream *secondary_stream) +{ +#if SECONDARY_ANY + int ret; +#define DECODE_SECONDARY_SECTION(UPPER,LOWER) \ + ((secondary_stream->dec_del_ind & VCD_ ## UPPER ## COMP) && \ + (ret = xd3_decode_secondary (secondary_stream, \ + & secondary_stream-> LOWER ## _sect, \ + & xd3_sec_ ## LOWER (secondary_stream)))) + + if (DECODE_SECONDARY_SECTION (DATA, data) || + DECODE_SECONDARY_SECTION (INST, inst) || + DECODE_SECONDARY_SECTION (ADDR, addr)) + { + return ret; + } +#undef DECODE_SECONDARY_SECTION +#endif + return 0; +} + +static int +xd3_decode_sections (xd3_stream *stream) +{ + usize_t need, more, take; + int copy, ret; + + if ((stream->flags & XD3_JUST_HDR) != 0) + { + /* Nothing left to do. */ + return xd3_decode_finish_window (stream); + } + + /* To avoid copying, need this much data available */ + need = (stream->inst_sect.size + + stream->addr_sect.size + + stream->data_sect.size); + + /* The window may be entirely processed. */ + XD3_ASSERT (stream->dec_winbytes <= need); + + /* Compute how much more input is needed. */ + more = (need - stream->dec_winbytes); + + /* How much to consume. */ + take = min (more, stream->avail_in); + + /* See if the input is completely available, to avoid copy. */ + copy = (take != more); + + /* If the window is skipped... */ + if ((stream->flags & XD3_SKIP_WINDOW) != 0) + { + /* Skip the available input. */ + DECODE_INPUT (take); + + stream->dec_winbytes += take; + + if (copy) + { + stream->msg = "further input required"; + return XD3_INPUT; + } + + return xd3_decode_finish_window (stream); + } + + /* Process all but the DATA section. */ + switch (stream->dec_state) + { + default: + stream->msg = "internal error"; + return XD3_INVALID_INPUT; + + case DEC_DATA: + if ((ret = xd3_decode_section (stream, & stream->data_sect, + DEC_INST, copy))) { return ret; } + case DEC_INST: + if ((ret = xd3_decode_section (stream, & stream->inst_sect, + DEC_ADDR, copy))) { return ret; } + case DEC_ADDR: + if ((ret = xd3_decode_section (stream, & stream->addr_sect, + DEC_EMIT, copy))) { return ret; } + } + + XD3_ASSERT (stream->dec_winbytes == need); + + if ((ret = xd3_decode_secondary_sections (stream))) { return ret; } + + if (stream->flags & XD3_SKIP_EMIT) + { + return xd3_decode_finish_window (stream); + } + + /* OPT: A possible optimization is to avoid allocating memory in + * decode_setup_buffers and to avoid a large memcpy when the window + * consists of a single VCD_SOURCE copy instruction. */ + if ((ret = xd3_decode_setup_buffers (stream))) { return ret; } + + return 0; +} + +static int +xd3_decode_emit (xd3_stream *stream) +{ + int ret; + + /* Produce output: originally structured to allow reentrant code + * that fills as much of the output buffer as possible, but VCDIFF + * semantics allows to copy from anywhere from the target window, so + * instead allocate a sufficiently sized buffer after the target + * window length is decoded. + * + * This code still needs to be reentrant to allow XD3_GETSRCBLK to + * return control. This is handled by setting the + * stream->dec_currentN instruction types to XD3_NOOP after they + * have been processed. */ + XD3_ASSERT (! (stream->flags & XD3_SKIP_EMIT)); + XD3_ASSERT (stream->dec_tgtlen <= stream->space_out); + + while (stream->inst_sect.buf != stream->inst_sect.buf_max || + stream->dec_current1.type != XD3_NOOP || + stream->dec_current2.type != XD3_NOOP) + { + /* Decode next instruction pair. */ + if ((stream->dec_current1.type == XD3_NOOP) && + (stream->dec_current2.type == XD3_NOOP) && + (ret = xd3_decode_instruction (stream))) { return ret; } + + /* Output dec_current1 */ + while ((stream->dec_current1.type != XD3_NOOP)) + { + if ((ret = xd3_decode_output_halfinst (stream, & stream->dec_current1))) + { + return ret; + } + } + /* Output dec_current2 */ + while (stream->dec_current2.type != XD3_NOOP) + { + if ((ret = xd3_decode_output_halfinst (stream, & stream->dec_current2))) + { + return ret; + } + } + } + + if (stream->avail_out != stream->dec_tgtlen) + { + IF_DEBUG2 (DP(RINT "AVAIL_OUT(%d) != DEC_TGTLEN(%d)\n", + stream->avail_out, stream->dec_tgtlen)); + stream->msg = "wrong window length"; + return XD3_INVALID_INPUT; + } + + if (stream->data_sect.buf != stream->data_sect.buf_max) + { + stream->msg = "extra data section"; + return XD3_INVALID_INPUT; + } + + if (stream->addr_sect.buf != stream->addr_sect.buf_max) + { + stream->msg = "extra address section"; + return XD3_INVALID_INPUT; + } + + /* OPT: Should cksum computation be combined with the above loop? */ + if ((stream->dec_win_ind & VCD_ADLER32) != 0 && + (stream->flags & XD3_ADLER32_NOVER) == 0) + { + uint32_t a32 = adler32 (1L, stream->next_out, stream->avail_out); + + if (a32 != stream->dec_adler32) + { + stream->msg = "target window checksum mismatch"; + return XD3_INVALID_INPUT; + } + } + + /* Finished with a window. */ + return xd3_decode_finish_window (stream); +} + +int +xd3_decode_input (xd3_stream *stream) +{ + int ret; + + if (stream->enc_state != 0) + { + stream->msg = "encoder/decoder transition"; + return XD3_INVALID_INPUT; + } + +#define BYTE_CASE(expr,x,nstate) \ + do { \ + if ( (expr) && \ + ((ret = xd3_decode_byte (stream, & (x))) != 0) ) { return ret; } \ + stream->dec_state = (nstate); \ + } while (0) + +#define OFFSET_CASE(expr,x,nstate) \ + do { \ + if ( (expr) && \ + ((ret = xd3_decode_offset (stream, & (x))) != 0) ) { return ret; } \ + stream->dec_state = (nstate); \ + } while (0) + +#define SIZE_CASE(expr,x,nstate) \ + do { \ + if ( (expr) && \ + ((ret = xd3_decode_size (stream, & (x))) != 0) ) { return ret; } \ + stream->dec_state = (nstate); \ + } while (0) + + switch (stream->dec_state) + { + case DEC_VCHEAD: + { + if ((ret = xd3_decode_bytes (stream, stream->dec_magic, + & stream->dec_magicbytes, 4))) + { + return ret; + } + + if (stream->dec_magic[0] != VCDIFF_MAGIC1 || + stream->dec_magic[1] != VCDIFF_MAGIC2 || + stream->dec_magic[2] != VCDIFF_MAGIC3) + { + stream->msg = "not a VCDIFF input"; + return XD3_INVALID_INPUT; + } + + if (stream->dec_magic[3] != 0) + { + stream->msg = "VCDIFF input version > 0 is not supported"; + return XD3_INVALID_INPUT; + } + + stream->dec_state = DEC_HDRIND; + } + case DEC_HDRIND: + { + if ((ret = xd3_decode_byte (stream, & stream->dec_hdr_ind))) + { + return ret; + } + + if ((stream->dec_hdr_ind & VCD_INVHDR) != 0) + { + stream->msg = "unrecognized header indicator bits set"; + return XD3_INVALID_INPUT; + } + + stream->dec_state = DEC_SECONDID; + } + + case DEC_SECONDID: + /* Secondary compressor ID: only if VCD_SECONDARY is set */ + if ((stream->dec_hdr_ind & VCD_SECONDARY) != 0) + { + BYTE_CASE (1, stream->dec_secondid, DEC_TABLEN); + + switch (stream->dec_secondid) + { + case VCD_FGK_ID: + FGK_CASE (stream); + case VCD_DJW_ID: + DJW_CASE (stream); + case VCD_LZMA_ID: + LZMA_CASE (stream); + default: + stream->msg = "unknown secondary compressor ID"; + return XD3_INVALID_INPUT; + } + } + + case DEC_TABLEN: + /* Length of code table data: only if VCD_CODETABLE is set */ + SIZE_CASE ((stream->dec_hdr_ind & VCD_CODETABLE) != 0, + stream->dec_codetblsz, DEC_NEAR); + + /* The codetblsz counts the two NEAR/SAME bytes */ + if ((stream->dec_hdr_ind & VCD_CODETABLE) != 0) { + if (stream->dec_codetblsz <= 2) { + stream->msg = "invalid code table size"; + return ENOMEM; + } + stream->dec_codetblsz -= 2; + } + case DEC_NEAR: + /* Near modes: only if VCD_CODETABLE is set */ + BYTE_CASE((stream->dec_hdr_ind & VCD_CODETABLE) != 0, + stream->acache.s_near, DEC_SAME); + case DEC_SAME: + /* Same modes: only if VCD_CODETABLE is set */ + BYTE_CASE((stream->dec_hdr_ind & VCD_CODETABLE) != 0, + stream->acache.s_same, DEC_TABDAT); + case DEC_TABDAT: + /* Compressed code table data */ + + if ((stream->dec_hdr_ind & VCD_CODETABLE) != 0) + { + /* Get the code table data. */ + if ((stream->dec_codetbl == NULL) && + (stream->dec_codetbl = + (uint8_t*) xd3_alloc (stream, + stream->dec_codetblsz, 1)) == NULL) + { + return ENOMEM; + } + + if ((ret = xd3_decode_bytes (stream, stream->dec_codetbl, + & stream->dec_codetblbytes, + stream->dec_codetblsz))) + { + return ret; + } + + if ((ret = xd3_apply_table_encoding (stream, stream->dec_codetbl, + stream->dec_codetblbytes))) + { + return ret; + } + } + else + { + /* Use the default table. */ + stream->acache.s_near = __rfc3284_code_table_desc.near_modes; + stream->acache.s_same = __rfc3284_code_table_desc.same_modes; + stream->code_table = xd3_rfc3284_code_table (); + } + + if ((ret = xd3_alloc_cache (stream))) { return ret; } + + stream->dec_state = DEC_APPLEN; + + case DEC_APPLEN: + /* Length of application data */ + SIZE_CASE((stream->dec_hdr_ind & VCD_APPHEADER) != 0, + stream->dec_appheadsz, DEC_APPDAT); + + case DEC_APPDAT: + /* Application data */ + if (stream->dec_hdr_ind & VCD_APPHEADER) + { + /* Note: we add an additional byte for padding, to allow + 0-termination. */ + if ((stream->dec_appheader == NULL) && + (stream->dec_appheader = + (uint8_t*) xd3_alloc (stream, + stream->dec_appheadsz+1, 1)) == NULL) + { + return ENOMEM; + } + + stream->dec_appheader[stream->dec_appheadsz] = 0; + + if ((ret = xd3_decode_bytes (stream, stream->dec_appheader, + & stream->dec_appheadbytes, + stream->dec_appheadsz))) + { + return ret; + } + } + + /* xoff_t -> usize_t is safe because this is the first block. */ + stream->dec_hdrsize = (usize_t) stream->total_in; + stream->dec_state = DEC_WININD; + + case DEC_WININD: + { + /* Start of a window: the window indicator */ + if ((ret = xd3_decode_byte (stream, & stream->dec_win_ind))) + { + return ret; + } + + stream->current_window = stream->dec_window_count; + + if (XOFF_T_OVERFLOW (stream->dec_winstart, stream->dec_tgtlen)) + { + stream->msg = "decoder file offset overflow"; + return XD3_INVALID_INPUT; + } + + stream->dec_winstart += stream->dec_tgtlen; + + if ((stream->dec_win_ind & VCD_INVWIN) != 0) + { + stream->msg = "unrecognized window indicator bits set"; + return XD3_INVALID_INPUT; + } + + if ((ret = xd3_decode_init_window (stream))) { return ret; } + + stream->dec_state = DEC_CPYLEN; + + IF_DEBUG2 (DP(RINT "--------- TARGET WINDOW %"Q"u -----------\n", + stream->current_window)); + } + + case DEC_CPYLEN: + /* Copy window length: only if VCD_SOURCE or VCD_TARGET is set */ + SIZE_CASE(SRCORTGT (stream->dec_win_ind), stream->dec_cpylen, + DEC_CPYOFF); + + /* Set the initial, logical decoder position (HERE address) in + * dec_position. This is set to just after the source/copy + * window, as we are just about to output the first byte of + * target window. */ + stream->dec_position = stream->dec_cpylen; + + case DEC_CPYOFF: + /* Copy window offset: only if VCD_SOURCE or VCD_TARGET is set */ + OFFSET_CASE(SRCORTGT (stream->dec_win_ind), stream->dec_cpyoff, + DEC_ENCLEN); + + /* Copy offset and copy length may not overflow. */ + if (XOFF_T_OVERFLOW (stream->dec_cpyoff, stream->dec_cpylen)) + { + stream->msg = "decoder copy window overflows a file offset"; + return XD3_INVALID_INPUT; + } + + /* Check copy window bounds: VCD_TARGET window may not exceed + current position. */ + if ((stream->dec_win_ind & VCD_TARGET) && + (stream->dec_cpyoff + (xoff_t) stream->dec_cpylen > + stream->dec_winstart)) + { + stream->msg = "VCD_TARGET window out of bounds"; + return XD3_INVALID_INPUT; + } + + case DEC_ENCLEN: + /* Length of the delta encoding */ + SIZE_CASE(1, stream->dec_enclen, DEC_TGTLEN); + case DEC_TGTLEN: + /* Length of target window */ + SIZE_CASE(1, stream->dec_tgtlen, DEC_DELIND); + + /* Set the maximum decoder position, beyond which we should not + * decode any data. This is the maximum value for dec_position. + * This may not exceed the size of a usize_t. */ + if (USIZE_T_OVERFLOW (stream->dec_cpylen, stream->dec_tgtlen)) + { + stream->msg = "decoder target window overflows a usize_t"; + return XD3_INVALID_INPUT; + } + + /* Check for malicious files. */ + if (stream->dec_tgtlen > XD3_HARDMAXWINSIZE) + { + stream->msg = "hard window size exceeded"; + return XD3_INVALID_INPUT; + } + + stream->dec_maxpos = stream->dec_cpylen + stream->dec_tgtlen; + + case DEC_DELIND: + /* Delta indicator */ + BYTE_CASE(1, stream->dec_del_ind, DEC_DATALEN); + + if ((stream->dec_del_ind & VCD_INVDEL) != 0) + { + stream->msg = "unrecognized delta indicator bits set"; + return XD3_INVALID_INPUT; + } + + /* Delta indicator is only used with secondary compression. */ + if ((stream->dec_del_ind != 0) && (stream->sec_type == NULL)) + { + stream->msg = "invalid delta indicator bits set"; + return XD3_INVALID_INPUT; + } + + /* Section lengths */ + case DEC_DATALEN: + SIZE_CASE(1, stream->data_sect.size, DEC_INSTLEN); + case DEC_INSTLEN: + SIZE_CASE(1, stream->inst_sect.size, DEC_ADDRLEN); + case DEC_ADDRLEN: + SIZE_CASE(1, stream->addr_sect.size, DEC_CKSUM); + + case DEC_CKSUM: + /* Window checksum. */ + if ((stream->dec_win_ind & VCD_ADLER32) != 0) + { + int i; + + if ((ret = xd3_decode_bytes (stream, stream->dec_cksum, + & stream->dec_cksumbytes, 4))) + { + return ret; + } + + for (i = 0; i < 4; i += 1) + { + stream->dec_adler32 = + (stream->dec_adler32 << 8) | stream->dec_cksum[i]; + } + } + + stream->dec_state = DEC_DATA; + + /* Check dec_enclen for redundency, otherwise it is not really used. */ + { + usize_t enclen_check = + (1 + (xd3_sizeof_size (stream->dec_tgtlen) + + xd3_sizeof_size (stream->data_sect.size) + + xd3_sizeof_size (stream->inst_sect.size) + + xd3_sizeof_size (stream->addr_sect.size)) + + stream->data_sect.size + + stream->inst_sect.size + + stream->addr_sect.size + + ((stream->dec_win_ind & VCD_ADLER32) ? 4 : 0)); + + if (stream->dec_enclen != enclen_check) + { + stream->msg = "incorrect encoding length (redundent)"; + return XD3_INVALID_INPUT; + } + } + + /* Returning here gives the application a chance to inspect the + * header, skip the window, etc. */ + if (stream->current_window == 0) { return XD3_GOTHEADER; } + else { return XD3_WINSTART; } + + case DEC_DATA: + case DEC_INST: + case DEC_ADDR: + /* Next read the three sections. */ + if ((ret = xd3_decode_sections (stream))) { return ret; } + + case DEC_EMIT: + + /* To speed VCD_SOURCE block-address calculations, the source + * cpyoff_blocks and cpyoff_blkoff are pre-computed. */ + if (stream->dec_win_ind & VCD_SOURCE) + { + xd3_source *src = stream->src; + + if (src == NULL) + { + stream->msg = "source input required"; + return XD3_INVALID_INPUT; + } + + xd3_blksize_div(stream->dec_cpyoff, src, + &src->cpyoff_blocks, + &src->cpyoff_blkoff); + + IF_DEBUG1(DP(RINT + "decode cpyoff %"Q"u " + "cpyblkno %"Q"u " + "cpyblkoff %u " + "blksize %u\n", + stream->dec_cpyoff, + src->cpyoff_blocks, + src->cpyoff_blkoff, + src->blksize)); + } + + /* xd3_decode_emit returns XD3_OUTPUT on every success. */ + if ((ret = xd3_decode_emit (stream)) == XD3_OUTPUT) + { + stream->total_out += (xoff_t) stream->avail_out; + } + + return ret; + + case DEC_FINISH: + { + if (stream->dec_win_ind & VCD_TARGET) + { + if (stream->dec_lastwin == NULL) + { + stream->dec_lastwin = stream->next_out; + stream->dec_lastspace = stream->space_out; + } + else + { + xd3_swap_uint8p (& stream->dec_lastwin, + & stream->next_out); + xd3_swap_usize_t (& stream->dec_lastspace, + & stream->space_out); + } + } + + stream->dec_lastlen = stream->dec_tgtlen; + stream->dec_laststart = stream->dec_winstart; + stream->dec_window_count += 1; + + /* Note: the updates to dec_winstart & current_window are + * deferred until after the next DEC_WININD byte is read. */ + stream->dec_state = DEC_WININD; + return XD3_WINFINISH; + } + + default: + stream->msg = "invalid state"; + return XD3_INVALID_INPUT; + } +} + +#endif // _XDELTA3_DECODE_H_ diff --git a/3rdparty/xdelta3-3.0.8/xdelta3-djw.h b/3rdparty/xdelta3-3.0.8/xdelta3-djw.h new file mode 100644 index 0000000..d6e2881 --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3-djw.h @@ -0,0 +1,1837 @@ +/* xdelta 3 - delta compression tools and library + * Copyright (C) 2002, 2006, 2007. Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +/* TODO: This code needs a thorough round of commenting. There is + * some slop in the declaration of arrays, which are maybe one element + * larger than they need to be and comments would help clear it up. */ + +#ifndef _XDELTA3_DJW_H_ +#define _XDELTA3_DJW_H_ + +/* The following people deserve much credit for the algorithms and + * techniques contained in this file: + + Julian Seward + Bzip2 sources, implementation of the multi-table Huffman technique. + + Jean-loup Gailly and Mark Adler and L. Peter Deutsch + Zlib source code, RFC 1951 + + Daniel S. Hirschberg and Debra A. LeLewer + "Efficient Decoding of Prefix Codes" + Communications of the ACM, April 1990 33(4). + + David J. Wheeler + Program bred3.c, bexp3 and accompanying documents bred3.ps, huff.ps. + This contains the idea behind the multi-table Huffman and 1-2 coding + techniques. + ftp://ftp.cl.cam.ac.uk/users/djw3/ + +*/ + +/* OPT: during the multi-table iteration, pick the worst-overall + * performing table and replace it with exactly the frequencies of the + * worst-overall performing sector or N-worst performing sectors. */ + +/* REF: See xdfs-0.222 and xdfs-0.226 for some old experiments with + * the Bzip prefix coding strategy. xdfs-0.256 contains the last of + * the other-format tests, including RFC1950 and the RFC1950+MTF + * tests. */ + +#define DJW_MAX_CODELEN 20U /* Maximum length of an alphabet code. */ + +/* Code lengths are themselves code-length encoded, so the total number of + * codes is: [RUN_0, RUN_1, 1-DJW_MAX_CODELEN] */ +#define DJW_TOTAL_CODES (DJW_MAX_CODELEN+2) + +#define RUN_0 0U /* Symbols used in MTF+1/2 coding. */ +#define RUN_1 1U + +/* Number of code lengths always encoded (djw_encode_basic array) */ +#define DJW_BASIC_CODES 5U +#define DJW_RUN_CODES 2U /* Number of run codes */ + +/* Offset of extra codes */ +#define DJW_EXTRA_12OFFSET (DJW_BASIC_CODES + DJW_RUN_CODES) + +/* Number of optionally encoded code lengths (djw_encode_extra array) */ +#define DJW_EXTRA_CODES 15U + +/* Number of bits to code [0-DJW_EXTRA_CODES] */ +#define DJW_EXTRA_CODE_BITS 4U + +#define DJW_MAX_GROUPS 8U /* Max number of group coding tables */ +#define DJW_GROUP_BITS 3U /* Number of bits to code [1-DJW_MAX_GROUPS] */ + +#define DJW_SECTORSZ_MULT 5U /* Multiplier for encoded sectorsz */ +#define DJW_SECTORSZ_BITS 5U /* Number of bits to code group size */ +#define DJW_SECTORSZ_MAX ((1U << DJW_SECTORSZ_BITS) * DJW_SECTORSZ_MULT) + +/* Maximum number of iterations to find group tables. */ +#define DJW_MAX_ITER 6U +/* Minimum number of bits an iteration must reduce coding by. */ +#define DJW_MIN_IMPROVEMENT 20U + +/* Maximum code length of a prefix code length */ +#define DJW_MAX_CLCLEN 15U + +/* Number of bits to code [0-DJW_MAX_CLCLEN] */ +#define DJW_CLCLEN_BITS 4U + +#define DJW_MAX_GBCLEN 7U /* Maximum code length of a group selector */ + +/* Number of bits to code [0-DJW_MAX_GBCLEN] + * TODO: Actually, should never have zero code lengths here, or else a group + * went unused. Write a test for this: if a group goes unused, eliminate + * it? */ +#define DJW_GBCLEN_BITS 3U + +/* It has to save at least this many bits... */ +#define EFFICIENCY_BITS 16U + +typedef struct _djw_stream djw_stream; +typedef struct _djw_heapen djw_heapen; +typedef struct _djw_prefix djw_prefix; +typedef uint32_t djw_weight; + +struct _djw_heapen +{ + uint32_t depth; + uint32_t freq; + uint32_t parent; +}; + +struct _djw_prefix +{ + usize_t scount; + uint8_t *symbol; + usize_t mcount; + uint8_t *mtfsym; + uint8_t *repcnt; +}; + +struct _djw_stream +{ + int unused; +}; + +/* Each Huffman table consists of 256 "code length" (CLEN) codes, + * which are themselves Huffman coded after eliminating repeats and + * move-to-front coding. The prefix consists of all the CLEN codes in + * djw_encode_basic plus a 4-bit value stating how many of the + * djw_encode_extra codes are actually coded (the rest are presumed + * zero, or unused CLEN codes). + * + * These values of these two arrays were arrived at by studying the + * distribution of min and max clen over a collection of DATA, INST, + * and ADDR inputs. The goal is to specify the order of + * djw_extra_codes that is most likely to minimize the number of extra + * codes that must be encoded. + * + * Results: 158896 sections were counted by compressing files (window + * size 512K) listed with: `find / -type f ( -user jmacd -o -perm +444 + * )` + * + * The distribution of CLEN codes for each efficient invocation of the + * secondary compressor (taking the best number of groups/sector size) + * was recorded. Then we look at the distribution of min and max clen + * values, counting the number of times the value C_low is less than + * the min and C_high is greater than the max. Values >= C_high and + * <= C_low will not have their lengths coded. The results are sorted + * and the least likely 15 are placed into the djw_encode_extra[] + * array in order. These values are used as the initial MTF ordering. + + clow[1] = 155119 + clow[2] = 140325 + clow[3] = 84072 + --- + clow[4] = 7225 + clow[5] = 1093 + clow[6] = 215 + --- + chigh[4] = 1 + chigh[5] = 30 + chigh[6] = 218 + chigh[7] = 2060 + chigh[8] = 13271 + --- + chigh[9] = 39463 + chigh[10] = 77360 + chigh[11] = 118298 + chigh[12] = 141360 + chigh[13] = 154086 + chigh[14] = 157967 + chigh[15] = 158603 + chigh[16] = 158864 + chigh[17] = 158893 + chigh[18] = 158895 + chigh[19] = 158896 + chigh[20] = 158896 + +*/ + +static const uint8_t djw_encode_12extra[DJW_EXTRA_CODES] = + { + 9, 10, 3, 11, 2, 12, 13, 1, 14, 15, 16, 17, 18, 19, 20, + }; + +static const uint8_t djw_encode_12basic[DJW_BASIC_CODES] = + { + 4, 5, 6, 7, 8, + }; + +/*********************************************************************/ +/* DECLS */ +/*********************************************************************/ + +static djw_stream* djw_alloc (xd3_stream *stream); +static int djw_init (xd3_stream *stream, + djw_stream *h, + int is_encode); +static void djw_destroy (xd3_stream *stream, + djw_stream *h); + +#if XD3_ENCODER +static int xd3_encode_huff (xd3_stream *stream, + djw_stream *sec_stream, + xd3_output *input, + xd3_output *output, + xd3_sec_cfg *cfg); +#endif + +static int xd3_decode_huff (xd3_stream *stream, + djw_stream *sec_stream, + const uint8_t **input, + const uint8_t *const input_end, + uint8_t **output, + const uint8_t *const output_end); + +/*********************************************************************/ +/* HUFFMAN */ +/*********************************************************************/ + +static djw_stream* +djw_alloc (xd3_stream *stream) +{ + return (djw_stream*) xd3_alloc (stream, sizeof (djw_stream), 1); +} + +static int +djw_init (xd3_stream *stream, djw_stream *h, int is_encode) +{ + /* Fields are initialized prior to use. */ + return 0; +} + +static void +djw_destroy (xd3_stream *stream, + djw_stream *h) +{ + xd3_free (stream, h); +} + + +/*********************************************************************/ +/* HEAP */ +/*********************************************************************/ + +static inline int +heap_less (const djw_heapen *a, const djw_heapen *b) +{ + return a->freq < b->freq || + (a->freq == b->freq && + a->depth < b->depth); +} + +static inline void +heap_insert (usize_t *heap, const djw_heapen *ents, usize_t p, const usize_t e) +{ + /* Insert ents[e] into next slot heap[p] */ + usize_t pp = p/2; /* P's parent */ + + while (heap_less (& ents[e], & ents[heap[pp]])) + { + heap[p] = heap[pp]; + p = pp; + pp = p/2; + } + + heap[p] = e; +} + +static inline djw_heapen* +heap_extract (usize_t *heap, const djw_heapen *ents, usize_t heap_last) +{ + usize_t smallest = heap[1]; + usize_t p, pc, t; + + /* Caller decrements heap_last, so heap_last+1 is the replacement elt. */ + heap[1] = heap[heap_last+1]; + + /* Re-heapify */ + for (p = 1; ; p = pc) + { + pc = p*2; + + /* Reached bottom of heap */ + if (pc > heap_last) { break; } + + /* See if second child is smaller. */ + if (pc < heap_last && heap_less (& ents[heap[pc+1]], & ents[heap[pc]])) + { + pc += 1; + } + + /* If pc is not smaller than p, heap property re-established. */ + if (! heap_less (& ents[heap[pc]], & ents[heap[p]])) { break; } + + t = heap[pc]; + heap[pc] = heap[p]; + heap[p] = t; + } + + return (djw_heapen*) & ents[smallest]; +} + +#if XD3_DEBUG +static void +heap_check (usize_t *heap, djw_heapen *ents, usize_t heap_last) +{ + usize_t i; + for (i = 1; i <= heap_last; i += 1) + { + /* Heap property: child not less than parent */ + XD3_ASSERT (! heap_less (& ents[heap[i]], & ents[heap[i/2]])); + + IF_DEBUG2 (DP(RINT "heap[%d] = %u\n", i, ents[heap[i]].freq)); + } +} +#endif + +/*********************************************************************/ +/* MTF, 1/2 */ +/*********************************************************************/ + +static inline usize_t +djw_update_mtf (uint8_t *mtf, usize_t mtf_i) +{ + int k; + usize_t sym = mtf[mtf_i]; + + for (k = mtf_i; k != 0; k -= 1) { mtf[k] = mtf[k-1]; } + + mtf[0] = sym; + return sym; +} + +static inline void +djw_update_1_2 (int *mtf_run, usize_t *mtf_i, + uint8_t *mtfsym, djw_weight *freq) +{ + int code; + + do + { + /* Offset by 1, since any number of RUN_ symbols implies run>0... */ + *mtf_run -= 1; + + code = (*mtf_run & 1) ? RUN_1 : RUN_0; + + mtfsym[(*mtf_i)++] = code; + freq[code] += 1; + *mtf_run >>= 1; + } + while (*mtf_run >= 1); + + *mtf_run = 0; +} + +static void +djw_init_clen_mtf_1_2 (uint8_t *clmtf) +{ + usize_t i, cl_i = 0; + + clmtf[cl_i++] = 0; + for (i = 0; i < DJW_BASIC_CODES; i += 1) + { + clmtf[cl_i++] = djw_encode_12basic[i]; + } + for (i = 0; i < DJW_EXTRA_CODES; i += 1) + { + clmtf[cl_i++] = djw_encode_12extra[i]; + } +} + +/*********************************************************************/ +/* PREFIX CODES */ +/*********************************************************************/ +#if XD3_ENCODER +static usize_t +djw_build_prefix (const djw_weight *freq, uint8_t *clen, usize_t asize, usize_t maxlen) +{ + /* Heap with 0th entry unused, prefix tree with up to ALPHABET_SIZE-1 + * internal nodes, never more than ALPHABET_SIZE entries actually in the + * heap (minimum weight subtrees during prefix construction). First + * ALPHABET_SIZE entries are the actual symbols, next ALPHABET_SIZE-1 are + * internal nodes. */ + djw_heapen ents[ALPHABET_SIZE * 2]; + usize_t heap[ALPHABET_SIZE + 1]; + + usize_t heap_last; /* Index of the last _valid_ heap entry. */ + usize_t ents_size; /* Number of entries, including 0th fake entry */ + usize_t overflow; /* Number of code lengths that overflow */ + uint32_t total_bits; + usize_t i; + + IF_DEBUG (uint32_t first_bits = 0); + + /* Insert real symbol frequences. */ + for (i = 0; i < asize; i += 1) + { + ents[i+1].freq = freq[i]; + IF_DEBUG2 (DP(RINT "ents[%d] = freq[%d] = %d\n", + i+1, i, freq[i])); + } + + again: + + /* The loop is re-entered each time an overflow occurs. Re-initialize... */ + heap_last = 0; + ents_size = 1; + overflow = 0; + total_bits = 0; + + /* 0th entry terminates the while loop in heap_insert (it's the parent of + * the smallest element, always less-than) */ + heap[0] = 0; + ents[0].depth = 0; + ents[0].freq = 0; + + /* Initial heap. */ + for (i = 0; i < asize; i += 1, ents_size += 1) + { + ents[ents_size].depth = 0; + ents[ents_size].parent = 0; + + if (ents[ents_size].freq != 0) + { + heap_insert (heap, ents, ++heap_last, ents_size); + } + } + + IF_DEBUG (heap_check (heap, ents, heap_last)); + + /* Must be at least one symbol, or else we can't get here. */ + XD3_ASSERT (heap_last != 0); + + /* If there is only one symbol, fake a second to prevent zero-length + * codes. */ + if (heap_last == 1) + { + /* Pick either the first or last symbol. */ + int s = freq[0] ? asize-1 : 0; + ents[s+1].freq = 1; + goto again; + } + + /* Build prefix tree. */ + while (heap_last > 1) + { + djw_heapen *h1 = heap_extract (heap, ents, --heap_last); + djw_heapen *h2 = heap_extract (heap, ents, --heap_last); + + ents[ents_size].freq = h1->freq + h2->freq; + ents[ents_size].depth = 1 + max (h1->depth, h2->depth); + ents[ents_size].parent = 0; + + h1->parent = h2->parent = ents_size; + + heap_insert (heap, ents, ++heap_last, ents_size++); + } + + IF_DEBUG (heap_check (heap, ents, heap_last)); + + /* Now compute prefix code lengths, counting parents. */ + for (i = 1; i < asize+1; i += 1) + { + usize_t b = 0; + + if (ents[i].freq != 0) + { + usize_t p = i; + + while ((p = ents[p].parent) != 0) { b += 1; } + + if (b > maxlen) { overflow = 1; } + + total_bits += b * freq[i-1]; + } + + /* clen is 0-origin, unlike ents. */ + IF_DEBUG2 (DP(RINT "clen[%d] = %d\n", i-1, b)); + clen[i-1] = b; + } + + IF_DEBUG (if (first_bits == 0) first_bits = total_bits); + + if (! overflow) + { + IF_DEBUG2 (if (first_bits != total_bits) + { + DP(RINT "code length overflow changed %u bits\n", + (usize_t)(total_bits - first_bits)); + }); + return total_bits; + } + + /* OPT: There is a non-looping way to fix overflow shown in zlib, but this + * is easier (for now), as done in bzip2. */ + for (i = 1; i < asize+1; i += 1) + { + ents[i].freq = ents[i].freq / 2 + 1; + } + + goto again; +} + +static void +djw_build_codes (usize_t *codes, const uint8_t *clen, usize_t asize, usize_t abs_max) +{ + usize_t i, l; + usize_t min_clen = DJW_MAX_CODELEN; + usize_t max_clen = 0; + usize_t code = 0; + + /* Find the min and max code length */ + for (i = 0; i < asize; i += 1) + { + if (clen[i] > 0 && clen[i] < min_clen) + { + min_clen = clen[i]; + } + + max_clen = max (max_clen, (usize_t) clen[i]); + } + + XD3_ASSERT (max_clen <= abs_max); + + /* Generate a code for each symbol with the appropriate length. */ + for (l = min_clen; l <= max_clen; l += 1) + { + for (i = 0; i < asize; i += 1) + { + if (clen[i] == l) + { + codes[i] = code++; + } + } + + code <<= 1; + } + + IF_DEBUG2 ({ + for (i = 0; i < asize; i += 1) + { + DP(RINT "code[%d] = %u\n", i, codes[i]); + } + }); +} + +/*********************************************************************/ +/* MOVE-TO-FRONT */ +/*********************************************************************/ +static void +djw_compute_mtf_1_2 (djw_prefix *prefix, + uint8_t *mtf, + djw_weight *freq_out, + usize_t nsym) +{ + size_t i, j, k; + usize_t sym; + usize_t size = prefix->scount; + usize_t mtf_i = 0; + int mtf_run = 0; + + /* This +2 is for the RUN_0, RUN_1 codes */ + memset (freq_out, 0, sizeof (freq_out[0]) * (nsym+2)); + + for (i = 0; i < size; ) + { + /* OPT: Bzip optimizes this algorithm a little by effectively checking + * j==0 before the MTF update. */ + sym = prefix->symbol[i++]; + + for (j = 0; mtf[j] != sym; j += 1) { } + + XD3_ASSERT (j <= nsym); + + for (k = j; k >= 1; k -= 1) { mtf[k] = mtf[k-1]; } + + mtf[0] = sym; + + if (j == 0) + { + mtf_run += 1; + continue; + } + + if (mtf_run > 0) + { + djw_update_1_2 (& mtf_run, & mtf_i, prefix->mtfsym, freq_out); + } + + /* Non-zero symbols are offset by RUN_1 */ + prefix->mtfsym[mtf_i++] = (uint8_t)(j+RUN_1); + freq_out[j+RUN_1] += 1; + } + + if (mtf_run > 0) + { + djw_update_1_2 (& mtf_run, & mtf_i, prefix->mtfsym, freq_out); + } + + prefix->mcount = mtf_i; +} + +/* Counts character frequencies of the input buffer, returns the size. */ +static usize_t +djw_count_freqs (djw_weight *freq, xd3_output *input) +{ + xd3_output *in; + usize_t size = 0; + + memset (freq, 0, sizeof (freq[0]) * ALPHABET_SIZE); + + for (in = input; in; in = in->next_page) + { + const uint8_t *p = in->base; + const uint8_t *p_max = p + in->next; + + size += in->next; + + do + { + ++freq[*p]; + } + while (++p < p_max); + } + + IF_DEBUG2 ({int i; + DP(RINT "freqs: "); + for (i = 0; i < ALPHABET_SIZE; i += 1) + { + DP(RINT "%u ", freq[i]); + } + DP(RINT "\n");}); + + return size; +} + +static void +djw_compute_multi_prefix (usize_t groups, + uint8_t clen[DJW_MAX_GROUPS][ALPHABET_SIZE], + djw_prefix *prefix) +{ + usize_t gp, i; + + prefix->scount = ALPHABET_SIZE; + memcpy (prefix->symbol, clen[0], ALPHABET_SIZE); + + for (gp = 1; gp < groups; gp += 1) + { + for (i = 0; i < ALPHABET_SIZE; i += 1) + { + if (clen[gp][i] == 0) + { + continue; + } + + prefix->symbol[prefix->scount++] = clen[gp][i]; + } + } +} + +static void +djw_compute_prefix_1_2 (djw_prefix *prefix, djw_weight *freq) +{ + /* This +1 is for the 0 code-length. */ + uint8_t clmtf[DJW_MAX_CODELEN+1]; + + djw_init_clen_mtf_1_2 (clmtf); + + djw_compute_mtf_1_2 (prefix, clmtf, freq, DJW_MAX_CODELEN); +} + +static int +djw_encode_prefix (xd3_stream *stream, + xd3_output **output, + bit_state *bstate, + djw_prefix *prefix) +{ + int ret; + size_t i; + usize_t num_to_encode; + djw_weight clfreq[DJW_TOTAL_CODES]; + uint8_t clclen[DJW_TOTAL_CODES]; + usize_t clcode[DJW_TOTAL_CODES]; + + /* Move-to-front encode prefix symbols, count frequencies */ + djw_compute_prefix_1_2 (prefix, clfreq); + + /* Compute codes */ + djw_build_prefix (clfreq, clclen, DJW_TOTAL_CODES, DJW_MAX_CLCLEN); + djw_build_codes (clcode, clclen, DJW_TOTAL_CODES, DJW_MAX_CLCLEN); + + /* Compute number of extra codes beyond basic ones for this template. */ + num_to_encode = DJW_TOTAL_CODES; + while (num_to_encode > DJW_EXTRA_12OFFSET && clclen[num_to_encode-1] == 0) + { + num_to_encode -= 1; + } + XD3_ASSERT (num_to_encode - DJW_EXTRA_12OFFSET < (1 << DJW_EXTRA_CODE_BITS)); + + /* Encode: # of extra codes */ + if ((ret = xd3_encode_bits (stream, output, bstate, DJW_EXTRA_CODE_BITS, + num_to_encode - DJW_EXTRA_12OFFSET))) + { + return ret; + } + + /* Encode: MTF code lengths */ + for (i = 0; i < num_to_encode; i += 1) + { + if ((ret = xd3_encode_bits (stream, output, bstate, + DJW_CLCLEN_BITS, clclen[i]))) + { + return ret; + } + } + + /* Encode: CLEN code lengths */ + for (i = 0; i < prefix->mcount; i += 1) + { + usize_t mtf_sym = prefix->mtfsym[i]; + usize_t bits = clclen[mtf_sym]; + usize_t code = clcode[mtf_sym]; + + if ((ret = xd3_encode_bits (stream, output, bstate, bits, code))) + { + return ret; + } + } + + return 0; +} + +static void +djw_compute_selector_1_2 (djw_prefix *prefix, + usize_t groups, + djw_weight *gbest_freq) +{ + uint8_t grmtf[DJW_MAX_GROUPS]; + usize_t i; + + for (i = 0; i < groups; i += 1) { grmtf[i] = i; } + + djw_compute_mtf_1_2 (prefix, grmtf, gbest_freq, groups); +} + +static int +xd3_encode_howmany_groups (xd3_stream *stream, + xd3_sec_cfg *cfg, + usize_t input_size, + usize_t *ret_groups, + usize_t *ret_sector_size) +{ + usize_t cfg_groups = 0; + usize_t cfg_sector_size = 0; + usize_t sugg_groups = 0; + usize_t sugg_sector_size = 0; + + if (cfg->ngroups != 0) + { + if (cfg->ngroups > DJW_MAX_GROUPS) + { + stream->msg = "invalid secondary encoder group number"; + return XD3_INTERNAL; + } + + cfg_groups = cfg->ngroups; + } + + if (cfg->sector_size != 0) + { + if (cfg->sector_size < DJW_SECTORSZ_MULT || + cfg->sector_size > DJW_SECTORSZ_MAX || + (cfg->sector_size % DJW_SECTORSZ_MULT) != 0) + { + stream->msg = "invalid secondary encoder sector size"; + return XD3_INTERNAL; + } + + cfg_sector_size = cfg->sector_size; + } + + if (cfg_groups == 0 || cfg_sector_size == 0) + { + /* These values were found empirically using xdelta3-tune around version + * xdfs-0.256. */ + switch (cfg->data_type) + { + case DATA_SECTION: + if (input_size < 1000) { sugg_groups = 1; sugg_sector_size = 0; } + else if (input_size < 4000) { sugg_groups = 2; sugg_sector_size = 10; } + else if (input_size < 7000) { sugg_groups = 3; sugg_sector_size = 10; } + else if (input_size < 10000) { sugg_groups = 4; sugg_sector_size = 10; } + else if (input_size < 25000) { sugg_groups = 5; sugg_sector_size = 10; } + else if (input_size < 50000) { sugg_groups = 7; sugg_sector_size = 20; } + else if (input_size < 100000) { sugg_groups = 8; sugg_sector_size = 30; } + else { sugg_groups = 8; sugg_sector_size = 70; } + break; + case INST_SECTION: + if (input_size < 7000) { sugg_groups = 1; sugg_sector_size = 0; } + else if (input_size < 10000) { sugg_groups = 2; sugg_sector_size = 50; } + else if (input_size < 25000) { sugg_groups = 3; sugg_sector_size = 50; } + else if (input_size < 50000) { sugg_groups = 6; sugg_sector_size = 40; } + else if (input_size < 100000) { sugg_groups = 8; sugg_sector_size = 40; } + else { sugg_groups = 8; sugg_sector_size = 40; } + break; + case ADDR_SECTION: + if (input_size < 9000) { sugg_groups = 1; sugg_sector_size = 0; } + else if (input_size < 25000) { sugg_groups = 2; sugg_sector_size = 130; } + else if (input_size < 50000) { sugg_groups = 3; sugg_sector_size = 130; } + else if (input_size < 100000) { sugg_groups = 5; sugg_sector_size = 130; } + else { sugg_groups = 7; sugg_sector_size = 130; } + break; + } + + if (cfg_groups == 0) + { + cfg_groups = sugg_groups; + } + + if (cfg_sector_size == 0) + { + cfg_sector_size = sugg_sector_size; + } + } + + if (cfg_groups != 1 && cfg_sector_size == 0) + { + switch (cfg->data_type) + { + case DATA_SECTION: + cfg_sector_size = 20; + break; + case INST_SECTION: + cfg_sector_size = 50; + break; + case ADDR_SECTION: + cfg_sector_size = 130; + break; + } + } + + (*ret_groups) = cfg_groups; + (*ret_sector_size) = cfg_sector_size; + + XD3_ASSERT (cfg_groups > 0 && cfg_groups <= DJW_MAX_GROUPS); + XD3_ASSERT (cfg_groups == 1 || + (cfg_sector_size >= DJW_SECTORSZ_MULT && + cfg_sector_size <= DJW_SECTORSZ_MAX)); + + return 0; +} + +static int +xd3_encode_huff (xd3_stream *stream, + djw_stream *h, + xd3_output *input, + xd3_output *output, + xd3_sec_cfg *cfg) +{ + int ret; + usize_t groups, sector_size; + bit_state bstate = BIT_STATE_ENCODE_INIT; + xd3_output *in; + int output_bits; + usize_t input_bits; + usize_t input_bytes; + usize_t initial_offset = output->next; + djw_weight real_freq[ALPHABET_SIZE]; + uint8_t *gbest = NULL; + uint8_t *gbest_mtf = NULL; + + input_bytes = djw_count_freqs (real_freq, input); + input_bits = input_bytes * 8; + + XD3_ASSERT (input_bytes > 0); + + if ((ret = xd3_encode_howmany_groups (stream, cfg, input_bytes, + & groups, & sector_size))) + { + return ret; + } + + if (0) + { + regroup: + /* Sometimes we dynamically decide there are too many groups. Arrive + * here. */ + output->next = initial_offset; + xd3_bit_state_encode_init (& bstate); + } + + /* Encode: # of groups (3 bits) */ + if ((ret = xd3_encode_bits (stream, & output, & bstate, + DJW_GROUP_BITS, groups-1))) { goto failure; } + + if (groups == 1) + { + /* Single Huffman group. */ + usize_t code[ALPHABET_SIZE]; /* Codes */ + uint8_t clen[ALPHABET_SIZE]; + uint8_t prefix_mtfsym[ALPHABET_SIZE]; + djw_prefix prefix; + + output_bits = + djw_build_prefix (real_freq, clen, ALPHABET_SIZE, DJW_MAX_CODELEN); + djw_build_codes (code, clen, ALPHABET_SIZE, DJW_MAX_CODELEN); + + if (output_bits + EFFICIENCY_BITS >= input_bits && ! cfg->inefficient) + { + goto nosecond; + } + + /* Encode: prefix */ + prefix.mtfsym = prefix_mtfsym; + prefix.symbol = clen; + prefix.scount = ALPHABET_SIZE; + + if ((ret = djw_encode_prefix (stream, & output, & bstate, & prefix))) + { + goto failure; + } + + if (output_bits + (8 * output->next) + EFFICIENCY_BITS >= + input_bits && ! cfg->inefficient) + { + goto nosecond; + } + + /* Encode: data */ + for (in = input; in; in = in->next_page) + { + const uint8_t *p = in->base; + const uint8_t *p_max = p + in->next; + + do + { + usize_t sym = *p++; + usize_t bits = clen[sym]; + + IF_DEBUG (output_bits -= bits); + + if ((ret = xd3_encode_bits (stream, & output, + & bstate, bits, code[sym]))) + { + goto failure; + } + } + while (p < p_max); + } + + XD3_ASSERT (output_bits == 0); + } + else + { + /* DJW Huffman */ + djw_weight evolve_freq[DJW_MAX_GROUPS][ALPHABET_SIZE]; + uint8_t evolve_clen[DJW_MAX_GROUPS][ALPHABET_SIZE]; + djw_weight left = input_bytes; + usize_t gp; + usize_t niter = 0; + usize_t select_bits; + usize_t sym1 = 0, sym2 = 0, s; + usize_t gcost[DJW_MAX_GROUPS]; + usize_t gbest_code[DJW_MAX_GROUPS+2]; + uint8_t gbest_clen[DJW_MAX_GROUPS+2]; + usize_t gbest_max = 1 + (input_bytes - 1) / sector_size; + usize_t best_bits = 0; + usize_t gbest_no; + usize_t gpcnt; + const uint8_t *p; + IF_DEBUG2 (usize_t gcount[DJW_MAX_GROUPS]); + + /* Encode: sector size (5 bits) */ + if ((ret = xd3_encode_bits (stream, & output, & bstate, + DJW_SECTORSZ_BITS, + (sector_size/DJW_SECTORSZ_MULT)-1))) + { + goto failure; + } + + /* Dynamic allocation. */ + if (gbest == NULL) + { + if ((gbest = (uint8_t*) xd3_alloc (stream, gbest_max, 1)) == NULL) + { + ret = ENOMEM; + goto failure; + } + } + + if (gbest_mtf == NULL) + { + if ((gbest_mtf = (uint8_t*) xd3_alloc (stream, gbest_max, 1)) == NULL) + { + ret = ENOMEM; + goto failure; + } + } + + /* OPT: Some of the inner loops can be optimized, as shown in bzip2 */ + + /* Generate initial code length tables. */ + for (gp = 0; gp < groups; gp += 1) + { + djw_weight sum = 0; + djw_weight goal = left / (groups - gp); + + IF_DEBUG2 (usize_t nz = 0); + + /* Due to the single-code granularity of this distribution, it may + * be that we can't generate a distribution for each group. In that + * case subtract one group and try again. If (inefficient), we're + * testing group behavior, so don't mess things up. */ + if (goal == 0 && !cfg->inefficient) + { + IF_DEBUG2 (DP(RINT "too many groups (%u), dropping one\n", + groups)); + groups -= 1; + goto regroup; + } + + /* Sum == goal is possible when (cfg->inefficient)... */ + while (sum < goal) + { + XD3_ASSERT (sym2 < ALPHABET_SIZE); + IF_DEBUG2 (nz += real_freq[sym2] != 0); + sum += real_freq[sym2++]; + } + + IF_DEBUG2(DP(RINT "group %u has symbols %u..%u (%u non-zero) " + "(%u/%u = %.3f)\n", + gp, sym1, sym2, nz, sum, + input_bytes, sum / (double)input_bytes);); + + for (s = 0; s < ALPHABET_SIZE; s += 1) + { + evolve_clen[gp][s] = (s >= sym1 && s <= sym2) ? 1 : 16; + } + + left -= sum; + sym1 = sym2+1; + } + + repeat: + + niter += 1; + gbest_no = 0; + memset (evolve_freq, 0, sizeof (evolve_freq[0]) * groups); + IF_DEBUG2 (memset (gcount, 0, sizeof (gcount[0]) * groups)); + + /* For each input page (loop is irregular to allow non-pow2-size group + * size. */ + in = input; + p = in->base; + + /* For each group-size sector. */ + do + { + const uint8_t *p0 = p; + xd3_output *in0 = in; + usize_t best = 0; + usize_t winner = 0; + + /* Select best group for each sector, update evolve_freq. */ + memset (gcost, 0, sizeof (gcost[0]) * groups); + + /* For each byte in sector. */ + for (gpcnt = 0; gpcnt < sector_size; gpcnt += 1) + { + /* For each group. */ + for (gp = 0; gp < groups; gp += 1) + { + gcost[gp] += evolve_clen[gp][*p]; + } + + /* Check end-of-input-page. */ +# define GP_PAGE() \ + if ((usize_t)(++p - in->base) == in->next) \ + { \ + in = in->next_page; \ + if (in == NULL) { break; } \ + p = in->base; \ + } + + GP_PAGE (); + } + + /* Find min cost group for this sector */ + best = USIZE_T_MAX; + for (gp = 0; gp < groups; gp += 1) + { + if (gcost[gp] < best) + { + best = gcost[gp]; + winner = gp; + } + } + + XD3_ASSERT(gbest_no < gbest_max); + gbest[gbest_no++] = winner; + IF_DEBUG2 (gcount[winner] += 1); + + p = p0; + in = in0; + + /* Update group frequencies. */ + for (gpcnt = 0; gpcnt < sector_size; gpcnt += 1) + { + evolve_freq[winner][*p] += 1; + + GP_PAGE (); + } + } + while (in != NULL); + + XD3_ASSERT (gbest_no == gbest_max); + + /* Recompute code lengths. */ + output_bits = 0; + for (gp = 0; gp < groups; gp += 1) + { + int i; + uint8_t evolve_zero[ALPHABET_SIZE]; + int any_zeros = 0; + + memset (evolve_zero, 0, sizeof (evolve_zero)); + + /* Cannot allow a zero clen when the real frequency is non-zero. + * Note: this means we are going to encode a fairly long code for + * these unused entries. An improvement would be to implement a + * NOTUSED code for when these are actually zero, but this requires + * another data structure (evolve_zero) since we don't know when + * evolve_freq[i] == 0... Briefly tested, looked worse. */ + for (i = 0; i < ALPHABET_SIZE; i += 1) + { + if (evolve_freq[gp][i] == 0 && real_freq[i] != 0) + { + evolve_freq[gp][i] = 1; + evolve_zero[i] = 1; + any_zeros = 1; + } + } + + output_bits += djw_build_prefix (evolve_freq[gp], evolve_clen[gp], + ALPHABET_SIZE, DJW_MAX_CODELEN); + + /* The above faking of frequencies does not matter for the last + * iteration, but we don't know when that is yet. However, it also + * breaks the output_bits computation. Necessary for accuracy, and + * for the (output_bits==0) assert after all bits are output. */ + if (any_zeros) + { + IF_DEBUG2 (usize_t save_total = output_bits); + + for (i = 0; i < ALPHABET_SIZE; i += 1) + { + if (evolve_zero[i]) { output_bits -= evolve_clen[gp][i]; } + } + + IF_DEBUG2 (DP(RINT "evolve_zero reduced %u bits in group %u\n", + save_total - output_bits, gp)); + } + } + + IF_DEBUG2( + DP(RINT "pass %u total bits: %u group uses: ", niter, output_bits); + for (gp = 0; gp < groups; gp += 1) { DP(RINT "%u ", gcount[gp]); } + DP(RINT "\n"); + ); + + /* End iteration. */ + + IF_DEBUG2 (if (niter > 1 && best_bits < output_bits) { + DP(RINT "iteration lost %u bits\n", output_bits - best_bits); }); + + if (niter == 1 || (niter < DJW_MAX_ITER && + (best_bits - output_bits) >= DJW_MIN_IMPROVEMENT)) + { + best_bits = output_bits; + goto repeat; + } + + /* Efficiency check. */ + if (output_bits + EFFICIENCY_BITS >= input_bits && ! cfg->inefficient) + { + goto nosecond; + } + + IF_DEBUG2 (DP(RINT "djw compression: %u -> %0.3f\n", + input_bytes, output_bits / 8.0)); + + /* Encode: prefix */ + { + uint8_t prefix_symbol[DJW_MAX_GROUPS * ALPHABET_SIZE]; + uint8_t prefix_mtfsym[DJW_MAX_GROUPS * ALPHABET_SIZE]; + uint8_t prefix_repcnt[DJW_MAX_GROUPS * ALPHABET_SIZE]; + djw_prefix prefix; + + prefix.symbol = prefix_symbol; + prefix.mtfsym = prefix_mtfsym; + prefix.repcnt = prefix_repcnt; + + djw_compute_multi_prefix (groups, evolve_clen, & prefix); + if ((ret = djw_encode_prefix (stream, & output, & bstate, & prefix))) + { + goto failure; + } + } + + /* Encode: selector frequencies */ + { + /* DJW_MAX_GROUPS +2 is for RUN_0, RUN_1 symbols. */ + djw_weight gbest_freq[DJW_MAX_GROUPS+2]; + djw_prefix gbest_prefix; + usize_t i; + + gbest_prefix.scount = gbest_no; + gbest_prefix.symbol = gbest; + gbest_prefix.mtfsym = gbest_mtf; + + djw_compute_selector_1_2 (& gbest_prefix, groups, gbest_freq); + + select_bits = + djw_build_prefix (gbest_freq, gbest_clen, groups+1, DJW_MAX_GBCLEN); + djw_build_codes (gbest_code, gbest_clen, groups+1, DJW_MAX_GBCLEN); + + for (i = 0; i < groups+1; i += 1) + { + if ((ret = xd3_encode_bits (stream, & output, & bstate, + DJW_GBCLEN_BITS, gbest_clen[i]))) + { + goto failure; + } + } + + for (i = 0; i < gbest_prefix.mcount; i += 1) + { + usize_t gp_mtf = gbest_mtf[i]; + usize_t gp_sel_bits = gbest_clen[gp_mtf]; + usize_t gp_sel_code = gbest_code[gp_mtf]; + + XD3_ASSERT (gp_mtf < groups+1); + + if ((ret = xd3_encode_bits (stream, & output, & bstate, + gp_sel_bits, gp_sel_code))) + { + goto failure; + } + + IF_DEBUG (select_bits -= gp_sel_bits); + } + + XD3_ASSERT (select_bits == 0); + } + + /* Efficiency check. */ + if (output_bits + select_bits + (8 * output->next) + + EFFICIENCY_BITS >= input_bits && ! cfg->inefficient) + { + goto nosecond; + } + + /* Encode: data */ + { + usize_t evolve_code[DJW_MAX_GROUPS][ALPHABET_SIZE]; + usize_t sector = 0; + + /* Build code tables for each group. */ + for (gp = 0; gp < groups; gp += 1) + { + djw_build_codes (evolve_code[gp], evolve_clen[gp], + ALPHABET_SIZE, DJW_MAX_CODELEN); + } + + /* Now loop over the input. */ + in = input; + p = in->base; + + do + { + /* For each sector. */ + usize_t gp_best = gbest[sector]; + usize_t *gp_codes = evolve_code[gp_best]; + uint8_t *gp_clens = evolve_clen[gp_best]; + + XD3_ASSERT (sector < gbest_no); + + sector += 1; + + /* Encode the sector data. */ + for (gpcnt = 0; gpcnt < sector_size; gpcnt += 1) + { + usize_t sym = *p; + usize_t bits = gp_clens[sym]; + usize_t code = gp_codes[sym]; + + IF_DEBUG (output_bits -= bits); + + if ((ret = xd3_encode_bits (stream, & output, & bstate, + bits, code))) + { + goto failure; + } + + GP_PAGE (); + } + } + while (in != NULL); + + XD3_ASSERT (select_bits == 0); + XD3_ASSERT (output_bits == 0); + } + } + + ret = xd3_flush_bits (stream, & output, & bstate); + + if (0) + { + nosecond: + stream->msg = "secondary compression was inefficient"; + ret = XD3_NOSECOND; + } + + failure: + + xd3_free (stream, gbest); + xd3_free (stream, gbest_mtf); + return ret; +} +#endif /* XD3_ENCODER */ + +/*********************************************************************/ +/* DECODE */ +/*********************************************************************/ + +static void +djw_build_decoder (xd3_stream *stream, + usize_t asize, + usize_t abs_max, + const uint8_t *clen, + uint8_t *inorder, + usize_t *base, + usize_t *limit, + usize_t *min_clenp, + usize_t *max_clenp) +{ + usize_t i, l; + const uint8_t *ci; + usize_t nr_clen [DJW_TOTAL_CODES]; + usize_t tmp_base[DJW_TOTAL_CODES]; + usize_t min_clen; + usize_t max_clen; + + /* Assumption: the two temporary arrays are large enough to hold abs_max. */ + XD3_ASSERT (abs_max <= DJW_MAX_CODELEN); + + /* This looks something like the start of zlib's inftrees.c */ + memset (nr_clen, 0, sizeof (nr_clen[0]) * (abs_max+1)); + + /* Count number of each code length */ + i = asize; + ci = clen; + do + { + /* Caller _must_ check that values are in-range. Most of the time the + * caller decodes a specific number of bits, which imply the max value, + * and the other time the caller decodes a huffman value, which must be + * in-range. Therefore, its an assertion and this function cannot + * otherwise fail. */ + XD3_ASSERT (*ci <= abs_max); + + nr_clen[*ci++]++; + } + while (--i != 0); + + /* Compute min, max. */ + for (i = 1; i <= abs_max; i += 1) { if (nr_clen[i]) { break; } } + min_clen = i; + for (i = abs_max; i != 0; i -= 1) { if (nr_clen[i]) { break; } } + max_clen = i; + + /* Fill the BASE, LIMIT table. */ + tmp_base[min_clen] = 0; + base[min_clen] = 0; + limit[min_clen] = nr_clen[min_clen] - 1; + for (i = min_clen + 1; i <= max_clen; i += 1) + { + usize_t last_limit = ((limit[i-1] + 1) << 1); + tmp_base[i] = tmp_base[i-1] + nr_clen[i-1]; + limit[i] = last_limit + nr_clen[i] - 1; + base[i] = last_limit - tmp_base[i]; + } + + /* Fill the inorder array, canonically ordered codes. */ + ci = clen; + for (i = 0; i < asize; i += 1) + { + if ((l = *ci++) != 0) + { + inorder[tmp_base[l]++] = i; + } + } + + *min_clenp = min_clen; + *max_clenp = max_clen; +} + +static inline int +djw_decode_symbol (xd3_stream *stream, + bit_state *bstate, + const uint8_t **input, + const uint8_t *input_end, + const uint8_t *inorder, + const usize_t *base, + const usize_t *limit, + usize_t min_clen, + usize_t max_clen, + usize_t *sym, + usize_t max_sym) +{ + usize_t code = 0; + usize_t bits = 0; + + /* OPT: Supposedly a small lookup table improves speed here... */ + + /* Code outline is similar to xd3_decode_bits... */ + if (bstate->cur_mask == 0x100) { goto next_byte; } + + for (;;) + { + do + { + if (bits == max_clen) { goto corrupt; } + + bits += 1; + code = (code << 1); + + if (bstate->cur_byte & bstate->cur_mask) { code |= 1; } + + bstate->cur_mask <<= 1; + + if (bits >= min_clen && code <= limit[bits]) { goto done; } + } + while (bstate->cur_mask != 0x100); + + next_byte: + + if (*input == input_end) + { + stream->msg = "secondary decoder end of input"; + return XD3_INTERNAL; + } + + bstate->cur_byte = *(*input)++; + bstate->cur_mask = 1; + } + + done: + + if (base[bits] <= code) + { + usize_t offset = code - base[bits]; + + if (offset <= max_sym) + { + IF_DEBUG2 (DP(RINT "(j) %u ", code)); + *sym = inorder[offset]; + return 0; + } + } + + corrupt: + stream->msg = "secondary decoder invalid code"; + return XD3_INTERNAL; +} + +static int +djw_decode_clclen (xd3_stream *stream, + bit_state *bstate, + const uint8_t **input, + const uint8_t *input_end, + uint8_t *cl_inorder, + usize_t *cl_base, + usize_t *cl_limit, + usize_t *cl_minlen, + usize_t *cl_maxlen, + uint8_t *cl_mtf) +{ + int ret; + uint8_t cl_clen[DJW_TOTAL_CODES]; + usize_t num_codes, value; + usize_t i; + + /* How many extra code lengths to encode. */ + if ((ret = xd3_decode_bits (stream, bstate, input, + input_end, DJW_EXTRA_CODE_BITS, & num_codes))) + { + return ret; + } + + num_codes += DJW_EXTRA_12OFFSET; + + /* Read num_codes. */ + for (i = 0; i < num_codes; i += 1) + { + if ((ret = xd3_decode_bits (stream, bstate, input, + input_end, DJW_CLCLEN_BITS, & value))) + { + return ret; + } + + cl_clen[i] = value; + } + + /* Set the rest to zero. */ + for (; i < DJW_TOTAL_CODES; i += 1) { cl_clen[i] = 0; } + + /* No need to check for in-range clen values, because: */ + XD3_ASSERT (1 << DJW_CLCLEN_BITS == DJW_MAX_CLCLEN + 1); + + /* Build the code-length decoder. */ + djw_build_decoder (stream, DJW_TOTAL_CODES, DJW_MAX_CLCLEN, + cl_clen, cl_inorder, cl_base, + cl_limit, cl_minlen, cl_maxlen); + + /* Initialize the MTF state. */ + djw_init_clen_mtf_1_2 (cl_mtf); + + return 0; +} + +static inline int +djw_decode_1_2 (xd3_stream *stream, + bit_state *bstate, + const uint8_t **input, + const uint8_t *input_end, + const uint8_t *inorder, + const usize_t *base, + const usize_t *limit, + const usize_t *minlen, + const usize_t *maxlen, + uint8_t *mtfvals, + usize_t elts, + usize_t skip_offset, + uint8_t *values) +{ + usize_t n = 0, rep = 0, mtf = 0, s = 0; + int ret; + + while (n < elts) + { + /* Special case inside generic code: CLEN only: If not the first group, + * we already know the zero frequencies. */ + if (skip_offset != 0 && n >= skip_offset && values[n-skip_offset] == 0) + { + values[n++] = 0; + continue; + } + + /* Repeat last symbol. */ + if (rep != 0) + { + values[n++] = mtfvals[0]; + rep -= 1; + continue; + } + + /* Symbol following last repeat code. */ + if (mtf != 0) + { + usize_t sym = djw_update_mtf (mtfvals, mtf); + values[n++] = sym; + mtf = 0; + continue; + } + + /* Decode next symbol/repeat code. */ + if ((ret = djw_decode_symbol (stream, bstate, input, input_end, + inorder, base, limit, *minlen, *maxlen, + & mtf, DJW_TOTAL_CODES))) { return ret; } + + if (mtf <= RUN_1) + { + /* Repetition. */ + rep = ((mtf + 1) << s); + mtf = 0; + s += 1; + } + else + { + /* Remove the RUN_1 MTF offset. */ + mtf -= 1; + s = 0; + } + } + + /* If (rep != 0) there were too many codes received. */ + if (rep != 0) + { + stream->msg = "secondary decoder invalid repeat code"; + return XD3_INTERNAL; + } + + return 0; +} + +static inline int +djw_decode_prefix (xd3_stream *stream, + bit_state *bstate, + const uint8_t **input, + const uint8_t *input_end, + const uint8_t *cl_inorder, + const usize_t *cl_base, + const usize_t *cl_limit, + const usize_t *cl_minlen, + const usize_t *cl_maxlen, + uint8_t *cl_mtf, + usize_t groups, + uint8_t *clen) +{ + return djw_decode_1_2 (stream, bstate, input, input_end, + cl_inorder, cl_base, cl_limit, + cl_minlen, cl_maxlen, cl_mtf, + ALPHABET_SIZE * groups, ALPHABET_SIZE, clen); +} + +static int +xd3_decode_huff (xd3_stream *stream, + djw_stream *h, + const uint8_t **input_pos, + const uint8_t *const input_end, + uint8_t **output_pos, + const uint8_t *const output_end) +{ + const uint8_t *input = *input_pos; + uint8_t *output = *output_pos; + bit_state bstate = BIT_STATE_DECODE_INIT; + uint8_t *sel_group = NULL; + usize_t groups, gp; + usize_t output_bytes = (usize_t)(output_end - output); + usize_t sector_size; + usize_t sectors; + int ret; + + /* Invalid input. */ + if (output_bytes == 0) + { + stream->msg = "secondary decoder invalid input"; + return XD3_INTERNAL; + } + + /* Decode: number of groups */ + if ((ret = xd3_decode_bits (stream, & bstate, & input, + input_end, DJW_GROUP_BITS, & groups))) + { + goto fail; + } + + groups += 1; + + if (groups > 1) + { + /* Decode: group size */ + if ((ret = xd3_decode_bits (stream, & bstate, & input, + input_end, DJW_SECTORSZ_BITS, + & sector_size))) { goto fail; } + + sector_size = (sector_size + 1) * DJW_SECTORSZ_MULT; + } + else + { + /* Default for groups == 1 */ + sector_size = output_bytes; + } + + sectors = 1 + (output_bytes - 1) / sector_size; + + /* TODO: In the case of groups==1, lots of extra stack space gets used here. + * Could dynamically allocate this memory, which would help with excess + * parameter passing, too. Passing too many parameters in this file, + * simplify it! */ + + /* Outer scope: per-group symbol decoder tables. */ + { + uint8_t inorder[DJW_MAX_GROUPS][ALPHABET_SIZE]; + usize_t base [DJW_MAX_GROUPS][DJW_TOTAL_CODES]; + usize_t limit [DJW_MAX_GROUPS][DJW_TOTAL_CODES]; + usize_t minlen [DJW_MAX_GROUPS]; + usize_t maxlen [DJW_MAX_GROUPS]; + + /* Nested scope: code length decoder tables. */ + { + uint8_t clen [DJW_MAX_GROUPS][ALPHABET_SIZE]; + uint8_t cl_inorder[DJW_TOTAL_CODES]; + usize_t cl_base [DJW_MAX_CLCLEN+2]; + usize_t cl_limit [DJW_MAX_CLCLEN+2]; + uint8_t cl_mtf [DJW_TOTAL_CODES]; + usize_t cl_minlen; + usize_t cl_maxlen; + + /* Compute the code length decoder. */ + if ((ret = djw_decode_clclen (stream, & bstate, & input, input_end, + cl_inorder, cl_base, cl_limit, & cl_minlen, + & cl_maxlen, cl_mtf))) { goto fail; } + + /* Now decode each group decoder. */ + if ((ret = djw_decode_prefix (stream, & bstate, & input, input_end, + cl_inorder, cl_base, cl_limit, + & cl_minlen, & cl_maxlen, cl_mtf, + groups, clen[0]))) { goto fail; } + + /* Prepare the actual decoding tables. */ + for (gp = 0; gp < groups; gp += 1) + { + djw_build_decoder (stream, ALPHABET_SIZE, DJW_MAX_CODELEN, + clen[gp], inorder[gp], base[gp], limit[gp], + & minlen[gp], & maxlen[gp]); + } + } + + /* Decode: selector clens. */ + { + uint8_t sel_inorder[DJW_MAX_GROUPS+2]; + usize_t sel_base [DJW_MAX_GBCLEN+2]; + usize_t sel_limit [DJW_MAX_GBCLEN+2]; + uint8_t sel_mtf [DJW_MAX_GROUPS+2]; + usize_t sel_minlen; + usize_t sel_maxlen; + + /* Setup group selection. */ + if (groups > 1) + { + uint8_t sel_clen[DJW_MAX_GROUPS+1]; + + for (gp = 0; gp < groups+1; gp += 1) + { + usize_t value; + + if ((ret = xd3_decode_bits (stream, & bstate, & input, + input_end, DJW_GBCLEN_BITS, + & value))) { goto fail; } + + sel_clen[gp] = value; + sel_mtf[gp] = gp; + } + + if ((sel_group = (uint8_t*) xd3_alloc (stream, sectors, 1)) == NULL) + { + ret = ENOMEM; + goto fail; + } + + djw_build_decoder (stream, groups+1, DJW_MAX_GBCLEN, sel_clen, + sel_inorder, sel_base, sel_limit, + & sel_minlen, & sel_maxlen); + + if ((ret = djw_decode_1_2 (stream, & bstate, & input, input_end, + sel_inorder, sel_base, + sel_limit, & sel_minlen, + & sel_maxlen, sel_mtf, + sectors, 0, sel_group))) { goto fail; } + } + + /* Now decode each sector. */ + { + /* Initialize for (groups==1) case. */ + uint8_t *gp_inorder = inorder[0]; + usize_t *gp_base = base[0]; + usize_t *gp_limit = limit[0]; + usize_t gp_minlen = minlen[0]; + usize_t gp_maxlen = maxlen[0]; + usize_t c; + + for (c = 0; c < sectors; c += 1) + { + usize_t n; + + if (groups >= 2) + { + gp = sel_group[c]; + + XD3_ASSERT (gp < groups); + + gp_inorder = inorder[gp]; + gp_base = base[gp]; + gp_limit = limit[gp]; + gp_minlen = minlen[gp]; + gp_maxlen = maxlen[gp]; + } + + XD3_ASSERT (output_end - output > 0); + + /* Decode next sector. */ + n = min (sector_size, (usize_t) (output_end - output)); + + do + { + usize_t sym; + + if ((ret = djw_decode_symbol (stream, & bstate, + & input, input_end, + gp_inorder, gp_base, + gp_limit, gp_minlen, gp_maxlen, + & sym, ALPHABET_SIZE))) + { + goto fail; + } + + *output++ = sym; + } + while (--n); + } + } + } + } + + IF_REGRESSION (if ((ret = xd3_test_clean_bits (stream, & bstate))) + { goto fail; }); + XD3_ASSERT (ret == 0); + + fail: + xd3_free (stream, sel_group); + + (*input_pos) = input; + (*output_pos) = output; + return ret; +} + +#endif diff --git a/3rdparty/xdelta3-3.0.8/xdelta3-fgk.h b/3rdparty/xdelta3-3.0.8/xdelta3-fgk.h new file mode 100644 index 0000000..7011500 --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3-fgk.h @@ -0,0 +1,860 @@ +/* xdelta 3 - delta compression tools and library + * Copyright (C) 2002, 2006, 2007. Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +/* For demonstration purposes only. + */ + +#ifndef _XDELTA3_FGK_h_ +#define _XDELTA3_FGK_h_ + +/* An implementation of the FGK algorithm described by D.E. Knuth in + * "Dynamic Huffman Coding" in Journal of Algorithms 6. */ + +/* A 32bit counter (fgk_weight) is used as the frequency counter for + * nodes in the huffman tree. TODO: Need oto test for overflow and/or + * reset stats. */ + +typedef struct _fgk_stream fgk_stream; +typedef struct _fgk_node fgk_node; +typedef struct _fgk_block fgk_block; +typedef unsigned int fgk_bit; +typedef uint32_t fgk_weight; + +struct _fgk_block { + union { + fgk_node *un_leader; + fgk_block *un_freeptr; + } un; +}; + +#define block_leader un.un_leader +#define block_freeptr un.un_freeptr + +/* The code can also support fixed huffman encoding/decoding. */ +#define IS_ADAPTIVE 1 + +/* weight is a count of the number of times this element has been seen + * in the current encoding/decoding. parent, right_child, and + * left_child are pointers defining the tree structure. right and + * left point to neighbors in an ordered sequence of weights. The + * left child of a node is always guaranteed to have weight not + * greater than its sibling. fgk_blockLeader points to the element + * with the same weight as itself which is closest to the next + * increasing weight block. */ +struct _fgk_node +{ + fgk_weight weight; + fgk_node *parent; + fgk_node *left_child; + fgk_node *right_child; + fgk_node *left; + fgk_node *right; + fgk_block *my_block; +}; + +/* alphabet_size is the a count of the number of possible leaves in + * the huffman tree. The number of total nodes counting internal + * nodes is ((2 * alphabet_size) - 1). zero_freq_count is the number + * of elements remaining which have zero frequency. zero_freq_exp and + * zero_freq_rem satisfy the equation zero_freq_count = + * 2^zero_freq_exp + zero_freq_rem. root_node is the root of the + * tree, which is initialized to a node with zero frequency and + * contains the 0th such element. free_node contains a pointer to the + * next available fgk_node space. alphabet contains all the elements + * and is indexed by N. remaining_zeros points to the head of the + * list of zeros. */ +struct _fgk_stream +{ + usize_t alphabet_size; + usize_t zero_freq_count; + usize_t zero_freq_exp; + usize_t zero_freq_rem; + usize_t coded_depth; + + usize_t total_nodes; + usize_t total_blocks; + + fgk_bit *coded_bits; + + fgk_block *block_array; + fgk_block *free_block; + + fgk_node *decode_ptr; + fgk_node *remaining_zeros; + fgk_node *alphabet; + fgk_node *root_node; + fgk_node *free_node; +}; + +/*********************************************************************/ +/* Encoder */ +/*********************************************************************/ + +static fgk_stream* fgk_alloc (xd3_stream *stream /*, usize_t alphabet_size */); +static int fgk_init (xd3_stream *stream, + fgk_stream *h, + int is_encode); +static int fgk_encode_data (fgk_stream *h, + usize_t n); +static inline fgk_bit fgk_get_encoded_bit (fgk_stream *h); + +static int xd3_encode_fgk (xd3_stream *stream, + fgk_stream *sec_stream, + xd3_output *input, + xd3_output *output, + xd3_sec_cfg *cfg); + +/*********************************************************************/ +/* Decoder */ +/*********************************************************************/ + +static inline int fgk_decode_bit (fgk_stream *h, + fgk_bit b); +static int fgk_decode_data (fgk_stream *h); +static void fgk_destroy (xd3_stream *stream, + fgk_stream *h); + +static int xd3_decode_fgk (xd3_stream *stream, + fgk_stream *sec_stream, + const uint8_t **input, + const uint8_t *const input_end, + uint8_t **output, + const uint8_t *const output_end); + +/*********************************************************************/ +/* Private */ +/*********************************************************************/ + +static unsigned int fgk_find_nth_zero (fgk_stream *h, usize_t n); +static usize_t fgk_nth_zero (fgk_stream *h, usize_t n); +static void fgk_update_tree (fgk_stream *h, usize_t n); +static fgk_node* fgk_increase_zero_weight (fgk_stream *h, usize_t n); +static void fgk_eliminate_zero (fgk_stream* h, fgk_node *node); +static void fgk_move_right (fgk_stream *h, fgk_node *node); +static void fgk_promote (fgk_stream *h, fgk_node *node); +static void fgk_init_node (fgk_node *node, usize_t i, usize_t size); +static fgk_block* fgk_make_block (fgk_stream *h, fgk_node *l); +static void fgk_free_block (fgk_stream *h, fgk_block *b); +static void fgk_factor_remaining (fgk_stream *h); +static inline void fgk_swap_ptrs (fgk_node **one, fgk_node **two); + +/*********************************************************************/ +/* Basic Routines */ +/*********************************************************************/ + +/* returns an initialized huffman encoder for an alphabet with the + * given size. returns NULL if enough memory cannot be allocated */ +static fgk_stream* fgk_alloc (xd3_stream *stream /*, int alphabet_size0 */) +{ + usize_t alphabet_size0 = ALPHABET_SIZE; + fgk_stream *h; + + if ((h = (fgk_stream*) xd3_alloc (stream, 1, sizeof (fgk_stream))) == NULL) + { + return NULL; + } + + h->total_nodes = (2 * alphabet_size0) - 1; + h->total_blocks = (2 * h->total_nodes); + h->alphabet = (fgk_node*) xd3_alloc (stream, h->total_nodes, sizeof (fgk_node)); + h->block_array = (fgk_block*) xd3_alloc (stream, h->total_blocks, sizeof (fgk_block)); + h->coded_bits = (fgk_bit*) xd3_alloc (stream, alphabet_size0, sizeof (fgk_bit)); + + if (h->coded_bits == NULL || + h->alphabet == NULL || + h->block_array == NULL) + { + fgk_destroy (stream, h); + return NULL; + } + + h->alphabet_size = alphabet_size0; + + return h; +} + +static int fgk_init (xd3_stream *stream, fgk_stream *h, int is_encode) +{ + usize_t ui; + ssize_t si; + + h->root_node = h->alphabet; + h->decode_ptr = h->root_node; + h->free_node = h->alphabet + h->alphabet_size; + h->remaining_zeros = h->alphabet; + h->coded_depth = 0; + h->zero_freq_count = h->alphabet_size + 2; + + /* after two calls to factor_remaining, zero_freq_count == alphabet_size */ + fgk_factor_remaining(h); /* set ZFE and ZFR */ + fgk_factor_remaining(h); /* set ZFDB according to prev state */ + + IF_DEBUG (memset (h->alphabet, 0, sizeof (h->alphabet[0]) * h->total_nodes)); + + for (ui = 0; ui < h->total_blocks-1; ui += 1) + { + h->block_array[ui].block_freeptr = &h->block_array[ui + 1]; + } + + h->block_array[h->total_blocks - 1].block_freeptr = NULL; + h->free_block = h->block_array; + + /* Zero frequency nodes are inserted in the first alphabet_size + * positions, with Value, weight, and a pointer to the next zero + * frequency node. */ + for (si = h->alphabet_size - 1; si >= 0; si -= 1) + { + fgk_init_node (h->alphabet + si, (usize_t) si, h->alphabet_size); + } + + return 0; +} + +static void fgk_swap_ptrs(fgk_node **one, fgk_node **two) +{ + fgk_node *tmp = *one; + *one = *two; + *two = tmp; +} + +/* Takes huffman transmitter h and n, the nth elt in the alphabet, and + * returns the number of required to encode n. */ +static int fgk_encode_data (fgk_stream* h, usize_t n) +{ + fgk_node *target_ptr = h->alphabet + n; + + XD3_ASSERT (n < h->alphabet_size); + + h->coded_depth = 0; + + /* First encode the binary representation of the nth remaining + * zero frequency element in reverse such that bit, which will be + * encoded from h->coded_depth down to 0 will arrive in increasing + * order following the tree path. If there is only one left, it + * is not neccesary to encode these bits. */ + if (IS_ADAPTIVE && target_ptr->weight == 0) + { + unsigned int where, shift; + int bits; + + where = fgk_find_nth_zero(h, n); + shift = 1; + + if (h->zero_freq_rem == 0) + { + bits = h->zero_freq_exp; + } + else + { + bits = h->zero_freq_exp + 1; + } + + while (bits > 0) + { + h->coded_bits[h->coded_depth++] = (shift & where) && 1; + + bits -= 1; + shift <<= 1; + }; + + target_ptr = h->remaining_zeros; + } + + /* The path from root to node is filled into coded_bits in reverse so + * that it is encoded in the right order */ + while (target_ptr != h->root_node) + { + h->coded_bits[h->coded_depth++] = (target_ptr->parent->right_child == target_ptr); + + target_ptr = target_ptr->parent; + } + + if (IS_ADAPTIVE) + { + fgk_update_tree(h, n); + } + + return h->coded_depth; +} + +/* Should be called as many times as fgk_encode_data returns. + */ +static inline fgk_bit fgk_get_encoded_bit (fgk_stream *h) +{ + XD3_ASSERT (h->coded_depth > 0); + + return h->coded_bits[--h->coded_depth]; +} + +/* This procedure updates the tree after alphabet[n] has been encoded + * or decoded. + */ +static void fgk_update_tree (fgk_stream *h, usize_t n) +{ + fgk_node *incr_node; + + if (h->alphabet[n].weight == 0) + { + incr_node = fgk_increase_zero_weight (h, n); + } + else + { + incr_node = h->alphabet + n; + } + + while (incr_node != h->root_node) + { + fgk_move_right (h, incr_node); + fgk_promote (h, incr_node); + incr_node->weight += 1; /* incr the parent */ + incr_node = incr_node->parent; /* repeat */ + } + + h->root_node->weight += 1; +} + +static void fgk_move_right (fgk_stream *h, fgk_node *move_fwd) +{ + fgk_node **fwd_par_ptr, **back_par_ptr; + fgk_node *move_back, *tmp; + + move_back = move_fwd->my_block->block_leader; + + if (move_fwd == move_back || + move_fwd->parent == move_back || + move_fwd->weight == 0) + { + return; + } + + move_back->right->left = move_fwd; + + if (move_fwd->left) + { + move_fwd->left->right = move_back; + } + + tmp = move_fwd->right; + move_fwd->right = move_back->right; + + if (tmp == move_back) + { + move_back->right = move_fwd; + } + else + { + tmp->left = move_back; + move_back->right = tmp; + } + + tmp = move_back->left; + move_back->left = move_fwd->left; + + if (tmp == move_fwd) + { + move_fwd->left = move_back; + } + else + { + tmp->right = move_fwd; + move_fwd->left = tmp; + } + + if (move_fwd->parent->right_child == move_fwd) + { + fwd_par_ptr = &move_fwd->parent->right_child; + } + else + { + fwd_par_ptr = &move_fwd->parent->left_child; + } + + if (move_back->parent->right_child == move_back) + { + back_par_ptr = &move_back->parent->right_child; + } + else + { + back_par_ptr = &move_back->parent->left_child; + } + + fgk_swap_ptrs (&move_fwd->parent, &move_back->parent); + fgk_swap_ptrs (fwd_par_ptr, back_par_ptr); + + move_fwd->my_block->block_leader = move_fwd; +} + +/* Shifts node, the leader of its block, into the next block. */ +static void fgk_promote (fgk_stream *h, fgk_node *node) +{ + fgk_node *my_left, *my_right; + fgk_block *cur_block; + + my_right = node->right; + my_left = node->left; + cur_block = node->my_block; + + if (node->weight == 0) + { + return; + } + + /* if left is right child, parent of remaining zeros case (?), means parent + * has same weight as right child. */ + if (my_left == node->right_child && + node->left_child && + node->left_child->weight == 0) + { + XD3_ASSERT (node->left_child == h->remaining_zeros); + XD3_ASSERT (node->right_child->weight == (node->weight+1)); /* child weight was already incremented */ + + if (node->weight == (my_right->weight - 1) && my_right != h->root_node) + { + fgk_free_block (h, cur_block); + node->my_block = my_right->my_block; + my_left->my_block = my_right->my_block; + } + + return; + } + + if (my_left == h->remaining_zeros) + { + return; + } + + /* true if not the leftmost node */ + if (my_left->my_block == cur_block) + { + my_left->my_block->block_leader = my_left; + } + else + { + fgk_free_block (h, cur_block); + } + + /* node->parent != my_right */ + if ((node->weight == (my_right->weight - 1)) && (my_right != h->root_node)) + { + node->my_block = my_right->my_block; + } + else + { + node->my_block = fgk_make_block (h, node); + } +} + +/* When an element is seen the first time this is called to remove it from the list of + * zero weight elements and introduce a new internal node to the tree. */ +static fgk_node* fgk_increase_zero_weight (fgk_stream *h, usize_t n) +{ + fgk_node *this_zero, *new_internal, *zero_ptr; + + this_zero = h->alphabet + n; + + if (h->zero_freq_count == 1) + { + /* this is the last one */ + this_zero->right_child = NULL; + + if (this_zero->right->weight == 1) + { + this_zero->my_block = this_zero->right->my_block; + } + else + { + this_zero->my_block = fgk_make_block (h, this_zero); + } + + h->remaining_zeros = NULL; + + return this_zero; + } + + zero_ptr = h->remaining_zeros; + + new_internal = h->free_node++; + + new_internal->parent = zero_ptr->parent; + new_internal->right = zero_ptr->right; + new_internal->weight = 0; + new_internal->right_child = this_zero; + new_internal->left = this_zero; + + if (h->remaining_zeros == h->root_node) + { + /* This is the first element to be coded */ + h->root_node = new_internal; + this_zero->my_block = fgk_make_block (h, this_zero); + new_internal->my_block = fgk_make_block (h, new_internal); + } + else + { + new_internal->right->left = new_internal; + + if (zero_ptr->parent->right_child == zero_ptr) + { + zero_ptr->parent->right_child = new_internal; + } + else + { + zero_ptr->parent->left_child = new_internal; + } + + if (new_internal->right->weight == 1) + { + new_internal->my_block = new_internal->right->my_block; + } + else + { + new_internal->my_block = fgk_make_block (h, new_internal); + } + + this_zero->my_block = new_internal->my_block; + } + + fgk_eliminate_zero (h, this_zero); + + new_internal->left_child = h->remaining_zeros; + + this_zero->right = new_internal; + this_zero->left = h->remaining_zeros; + this_zero->parent = new_internal; + this_zero->left_child = NULL; + this_zero->right_child = NULL; + + h->remaining_zeros->parent = new_internal; + h->remaining_zeros->right = this_zero; + + return this_zero; +} + +/* When a zero frequency element is encoded, it is followed by the + * binary representation of the index into the remaining elements. + * Sets a cache to the element before it so that it can be removed + * without calling this procedure again. */ +static unsigned int fgk_find_nth_zero (fgk_stream* h, usize_t n) +{ + fgk_node *target_ptr = h->alphabet + n; + fgk_node *head_ptr = h->remaining_zeros; + unsigned int idx = 0; + + while (target_ptr != head_ptr) + { + head_ptr = head_ptr->right_child; + idx += 1; + } + + return idx; +} + +/* Splices node out of the list of zeros. */ +static void fgk_eliminate_zero (fgk_stream* h, fgk_node *node) +{ + if (h->zero_freq_count == 1) + { + return; + } + + fgk_factor_remaining(h); + + if (node->left_child == NULL) + { + h->remaining_zeros = h->remaining_zeros->right_child; + h->remaining_zeros->left_child = NULL; + } + else if (node->right_child == NULL) + { + node->left_child->right_child = NULL; + } + else + { + node->right_child->left_child = node->left_child; + node->left_child->right_child = node->right_child; + } +} + +static void fgk_init_node (fgk_node *node, usize_t i, usize_t size) +{ + if (i < size - 1) + { + node->right_child = node + 1; + } + else + { + node->right_child = NULL; + } + + if (i >= 1) + { + node->left_child = node - 1; + } + else + { + node->left_child = NULL; + } + + node->weight = 0; + node->parent = NULL; + node->right = NULL; + node->left = NULL; + node->my_block = NULL; +} + +/* The data structure used is an array of blocks, which are unions of + * free pointers and huffnode pointers. free blocks are a linked list + * of free blocks, the front of which is h->free_block. The used + * blocks are pointers to the head of each block. */ +static fgk_block* fgk_make_block (fgk_stream *h, fgk_node* lead) +{ + fgk_block *ret = h->free_block; + + XD3_ASSERT (h->free_block != NULL); + + h->free_block = h->free_block->block_freeptr; + + ret->block_leader = lead; + + return ret; +} + +/* Restores the block to the front of the free list. */ +static void fgk_free_block (fgk_stream *h, fgk_block *b) +{ + b->block_freeptr = h->free_block; + h->free_block = b; +} + +/* sets zero_freq_count, zero_freq_rem, and zero_freq_exp to satsity + * the equation given above. */ +static void fgk_factor_remaining (fgk_stream *h) +{ + unsigned int i; + + i = (--h->zero_freq_count); + h->zero_freq_exp = 0; + + while (i > 1) + { + h->zero_freq_exp += 1; + i >>= 1; + } + + i = 1 << h->zero_freq_exp; + + h->zero_freq_rem = h->zero_freq_count - i; +} + +/* receives a bit at a time and returns true when a complete code has + * been received. + */ +static inline int fgk_decode_bit (fgk_stream* h, fgk_bit b) +{ + XD3_ASSERT (b == 1 || b == 0); + + if (IS_ADAPTIVE && h->decode_ptr->weight == 0) + { + usize_t bitsreq; + + if (h->zero_freq_rem == 0) + { + bitsreq = h->zero_freq_exp; + } + else + { + bitsreq = h->zero_freq_exp + 1; + } + + h->coded_bits[h->coded_depth] = b; + h->coded_depth += 1; + + return h->coded_depth >= bitsreq; + } + else + { + if (b) + { + h->decode_ptr = h->decode_ptr->right_child; + } + else + { + h->decode_ptr = h->decode_ptr->left_child; + } + + if (h->decode_ptr->left_child == NULL) + { + /* If the weight is non-zero, finished. */ + if (h->decode_ptr->weight != 0) + { + return 1; + } + + /* zero_freq_count is dropping to 0, finished. */ + return h->zero_freq_count == 1; + } + else + { + return 0; + } + } +} + +static usize_t fgk_nth_zero (fgk_stream* h, usize_t n) +{ + fgk_node *ret = h->remaining_zeros; + + /* ERROR: if during this loop (ret->right_child == NULL) then the + * encoder's zero count is too high. Could return an error code + * now, but is probably unnecessary overhead, since the caller + * should check integrity anyway. */ + for (; n != 0 && ret->right_child != NULL; n -= 1) + { + ret = ret->right_child; + } + + return (usize_t)(ret - h->alphabet); +} + +/* once fgk_decode_bit returns 1, this retrieves an index into the + * alphabet otherwise this returns 0, indicating more bits are + * required. + */ +static int fgk_decode_data (fgk_stream* h) +{ + usize_t elt = (usize_t)(h->decode_ptr - h->alphabet); + + if (IS_ADAPTIVE && h->decode_ptr->weight == 0) { + usize_t i = 0; + usize_t n = 0; + + if (h->coded_depth > 0) + { + for (; i < h->coded_depth - 1; i += 1) + { + n |= h->coded_bits[i]; + n <<= 1; + } + } + + n |= h->coded_bits[i]; + elt = fgk_nth_zero(h, n); + } + + h->coded_depth = 0; + + if (IS_ADAPTIVE) + { + fgk_update_tree(h, elt); + } + + h->decode_ptr = h->root_node; + + return elt; +} + +static void fgk_destroy (xd3_stream *stream, + fgk_stream *h) +{ + if (h != NULL) + { + xd3_free (stream, h->alphabet); + xd3_free (stream, h->coded_bits); + xd3_free (stream, h->block_array); + xd3_free (stream, h); + } +} + +/*********************************************************************/ +/* Xdelta */ +/*********************************************************************/ + +static int +xd3_encode_fgk (xd3_stream *stream, fgk_stream *sec_stream, xd3_output *input, xd3_output *output, xd3_sec_cfg *cfg) +{ + bit_state bstate = BIT_STATE_ENCODE_INIT; + xd3_output *cur_page; + int ret; + + /* OPT: quit compression early if it looks bad */ + for (cur_page = input; cur_page; cur_page = cur_page->next_page) + { + const uint8_t *inp = cur_page->base; + const uint8_t *inp_max = inp + cur_page->next; + + while (inp < inp_max) + { + usize_t bits = fgk_encode_data (sec_stream, *inp++); + + while (bits--) + { + if ((ret = xd3_encode_bit (stream, & output, & bstate, fgk_get_encoded_bit (sec_stream)))) { return ret; } + } + } + } + + return xd3_flush_bits (stream, & output, & bstate); +} + +static int +xd3_decode_fgk (xd3_stream *stream, + fgk_stream *sec_stream, + const uint8_t **input_pos, + const uint8_t *const input_max, + uint8_t **output_pos, + const uint8_t *const output_max) +{ + bit_state bstate; + uint8_t *output = *output_pos; + const uint8_t *input = *input_pos; + + for (;;) + { + if (input == input_max) + { + stream->msg = "secondary decoder end of input"; + return XD3_INTERNAL; + } + + bstate.cur_byte = *input++; + + for (bstate.cur_mask = 1; bstate.cur_mask != 0x100; bstate.cur_mask <<= 1) + { + int done = fgk_decode_bit (sec_stream, (bstate.cur_byte & bstate.cur_mask) ? 1U : 0U); + + if (! done) { continue; } + + *output++ = fgk_decode_data (sec_stream); + + if (output == output_max) + { + /* During regression testing: */ + IF_REGRESSION ({ + int ret; + bstate.cur_mask <<= 1; + if ((ret = xd3_test_clean_bits (stream, & bstate))) { return ret; } + }); + + (*output_pos) = output; + (*input_pos) = input; + return 0; + } + } + } +} + +#endif /* _XDELTA3_FGK_ */ diff --git a/3rdparty/xdelta3-3.0.8/xdelta3-hash.h b/3rdparty/xdelta3-3.0.8/xdelta3-hash.h new file mode 100644 index 0000000..e359436 --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3-hash.h @@ -0,0 +1,170 @@ +/* xdelta 3 - delta compression tools and library + * Copyright (C) 2001, 2003, 2004, 2005, 2006, 2007. Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +#ifndef _XDELTA3_HASH_H_ +#define _XDELTA3_HASH_H_ + +#if XD3_DEBUG +#define SMALL_HASH_DEBUG1(s,inp) \ + uint32_t debug_state; \ + uint32_t debug_hval = xd3_checksum_hash (& (s)->small_hash, \ + xd3_scksum (&debug_state, (inp), (s)->smatcher.small_look)) +#define SMALL_HASH_DEBUG2(s,inp) \ + XD3_ASSERT (debug_hval == xd3_checksum_hash (& (s)->small_hash, \ + xd3_scksum (&debug_state, (inp), (s)->smatcher.small_look))) +#else +#define SMALL_HASH_DEBUG1(s,inp) +#define SMALL_HASH_DEBUG2(s,inp) +#endif /* XD3_DEBUG */ + +/* This is a good hash multiplier for 32-bit LCGs: see "linear + * congruential generators of different sizes and good lattice + * structure" */ +static const uint32_t hash_multiplier = 1597334677U; + +/*********************************************************************** + Permute stuff + ***********************************************************************/ + +#if HASH_PERMUTE == 0 +#define PERMUTE(x) (x) +#else +#define PERMUTE(x) (__single_hash[(uint32_t)x]) + +extern const uint16_t __single_hash[256]; +#endif + +/* Update the checksum state. */ +#if ADLER_LARGE_CKSUM +inline uint32_t +xd3_large_cksum_update (uint32_t cksum, + const uint8_t *base, + usize_t look) { + uint32_t old_c = PERMUTE(base[0]); + uint32_t new_c = PERMUTE(base[look]); + uint32_t low = ((cksum & 0xffff) - old_c + new_c) & 0xffff; + uint32_t high = ((cksum >> 16) - (old_c * look) + low) & 0xffff; + return (high << 16) | low; +} +#else +/* TODO: revisit this topic */ +#endif + +#if UNALIGNED_OK +#define UNALIGNED_READ32(dest,src) (*(dest)) = (*(uint32_t*)(src)) +#else +#define UNALIGNED_READ32(dest,src) memcpy((dest), (src), 4); +#endif + +/* TODO: small cksum is hard-coded for 4 bytes (i.e., "look" is unused) */ +static inline uint32_t +xd3_scksum (uint32_t *state, + const uint8_t *base, + const usize_t look) +{ + UNALIGNED_READ32(state, base); + return (*state) * hash_multiplier; +} +static inline uint32_t +xd3_small_cksum_update (uint32_t *state, + const uint8_t *base, + usize_t look) +{ + UNALIGNED_READ32(state, base+1); + return (*state) * hash_multiplier; +} + +/*********************************************************************** + Ctable stuff + ***********************************************************************/ + +static inline usize_t +xd3_checksum_hash (const xd3_hash_cfg *cfg, const usize_t cksum) +{ + return (cksum >> cfg->shift) ^ (cksum & cfg->mask); +} + +/*********************************************************************** + Cksum function + ***********************************************************************/ + +#if ADLER_LARGE_CKSUM +static inline uint32_t +xd3_lcksum (const uint8_t *seg, const usize_t ln) +{ + usize_t i = 0; + uint32_t low = 0; + uint32_t high = 0; + + for (; i < ln; i += 1) + { + low += PERMUTE(*seg++); + high += low; + } + + return ((high & 0xffff) << 16) | (low & 0xffff); +} +#else +static inline uint32_t +xd3_lcksum (const uint8_t *seg, const usize_t ln) +{ + usize_t i, j; + uint32_t h = 0; + for (i = 0, j = ln - 1; i < ln; ++i, --j) { + h += PERMUTE(seg[i]) * hash_multiplier_powers[j]; + } + return h; +} +#endif + +#if XD3_ENCODER +static usize_t +xd3_size_log2 (usize_t slots) +{ + int bits = 28; /* This should not be an unreasonable limit. */ + int i; + + for (i = 3; i <= bits; i += 1) + { + if (slots < (1U << i)) + { + /* TODO: this is compaction=1 in checksum_test.cc and maybe should + * not be fixed at -1. */ + bits = i - 1; + break; + } + } + + return bits; +} + +static void +xd3_size_hashtable (xd3_stream *stream, + usize_t slots, + xd3_hash_cfg *cfg) +{ + int bits = xd3_size_log2 (slots); + + /* TODO: there's a 32-bit assumption here */ + cfg->size = (1 << bits); + cfg->mask = (cfg->size - 1); + cfg->shift = 32 - bits; +} +#endif + +#endif diff --git a/3rdparty/xdelta3-3.0.8/xdelta3-internal.h b/3rdparty/xdelta3-3.0.8/xdelta3-internal.h new file mode 100644 index 0000000..8e9e25b --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3-internal.h @@ -0,0 +1,145 @@ +/* xdelta3 - delta compression tools and library + * Copyright (C) 2011, 2012 Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ +#ifndef XDELTA3_INTERNAL_H__ +#define XDELTA3_INTERNAL_H__ + +#include "xdelta3.h" + +typedef struct _main_file main_file; +typedef struct _main_extcomp main_extcomp; + +void main_buffree (void *ptr); +void* main_bufalloc (size_t size); +void main_file_init (main_file *xfile); +int main_file_close (main_file *xfile); +void main_file_cleanup (main_file *xfile); +int main_file_isopen (main_file *xfile); +int main_file_open (main_file *xfile, const char* name, int mode); +int main_file_exists (main_file *xfile); +int xd3_whole_append_window (xd3_stream *stream); +int xd3_main_cmdline (int argc, char **argv); +int main_file_read (main_file *ifile, + uint8_t *buf, + size_t size, + size_t *nread, + const char *msg); +int main_file_write (main_file *ofile, uint8_t *buf, + usize_t size, const char *msg); +int test_compare_files (const char* f0, const char* f1); +usize_t xd3_bytes_on_srcblk (xd3_source *src, xoff_t blkno); +xoff_t xd3_source_eof(const xd3_source *src); +uint32_t xd3_large_cksum_update (uint32_t cksum, + const uint8_t *base, + usize_t look); +int xd3_encode_init_full (xd3_stream *stream); +#if PYTHON_MODULE || SWIG_MODULE || NOT_MAIN +int xd3_main_cmdline (int argc, char **argv); +#endif + +/* main_file->mode values */ +typedef enum +{ + XO_READ = 0, + XO_WRITE = 1 +} main_file_modes; + +struct _main_file +{ +#if XD3_STDIO + FILE *file; +#elif XD3_POSIX + int file; +#elif XD3_WIN32 + HANDLE file; +#endif + + int mode; /* XO_READ and XO_WRITE */ + const char *filename; /* File name or /dev/stdin, + * /dev/stdout, /dev/stderr. */ + char *filename_copy; /* File name or /dev/stdin, + * /dev/stdout, /dev/stderr. */ + const char *realname; /* File name or /dev/stdin, + * /dev/stdout, /dev/stderr. */ + const main_extcomp *compressor; /* External compression struct. */ + int flags; /* RD_FIRST, RD_NONEXTERNAL, ... */ + xoff_t nread; /* for input position */ + xoff_t nwrite; /* for output position */ + uint8_t *snprintf_buf; /* internal snprintf() use */ + int size_known; /* Set by main_set_souze */ + xoff_t source_position; /* for avoiding seek in getblk_func */ + int seek_failed; /* after seek fails once, try FIFO */ +}; + +/* According to the internet, Windows vsnprintf() differs from most + * Unix implementations regarding the terminating 0 when the boundary + * condition is met. It doesn't matter here, we don't rely on the + * trailing 0. Besides, both Windows and DJGPP vsnprintf return -1 + * upon truncation, which isn't C99 compliant. To overcome this, + * recent MinGW runtimes provided their own vsnprintf (notice the + * absence of the '_' prefix) but they were initially buggy. So, + * always use the native '_'-prefixed version with Win32. */ +#ifdef _WIN32 +#define vsnprintf_func(str,size,fmt,args) \ + _vsnprintf_s(str,size,size-1,fmt,args) +#define snprintf_func(str,size,fmt,...) \ + _snprintf_s(str,size,size-1,fmt,__VA_ARGS__) +#else +#define vsnprintf_func vsnprintf +#define snprintf_func snprintf +#endif +#define short_sprintf(sb,fmt,...) \ + snprintf_func((sb).buf,sizeof((sb).buf),fmt,__VA_ARGS__) + +/* Type used for short snprintf calls. */ +typedef struct { + char buf[48]; +} shortbuf; + +/* Prior to SVN 303 this function was only defined in DJGPP and WIN32 + * environments and other platforms would use the builtin snprintf() + * with an arrangement of macros below. In OS X 10.6, Apply made + * snprintf() a macro, which defeated those macros (since snprintf + * would be evaluated before its argument macros were expanded, + * therefore always define xsnprintf_func. */ +#undef PRINTF_ATTRIBUTE +#ifdef __GNUC__ +/* Let's just assume no one uses gcc 2.x! */ +#define PRINTF_ATTRIBUTE(x,y) __attribute__ ((__format__ (__printf__, x, y))) +#else +#define PRINTF_ATTRIBUTE(x,y) +#endif + +/* Underlying xprintf() */ +int xsnprintf_func (char *str, int n, const char *fmt, ...) + PRINTF_ATTRIBUTE(3,4); + +/* XPR(NT "", ...) (used by main) prefixes an "xdelta3: " to the output. */ +void xprintf(const char *fmt, ...) PRINTF_ATTRIBUTE(1,2); +#define XPR xprintf +#define NT "xdelta3: " +#define NTR "" + +#ifndef UINT32_MAX +#define UINT32_MAX 4294967295U +#endif + +#ifndef UINT64_MAX +#define UINT64_MAX 18446744073709551615ULL +#endif + +#endif // XDELTA3_INTERNAL_H__ diff --git a/3rdparty/xdelta3-3.0.8/xdelta3-list.h b/3rdparty/xdelta3-3.0.8/xdelta3-list.h new file mode 100644 index 0000000..6e3125f --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3-list.h @@ -0,0 +1,130 @@ +/* xdelta 3 - delta compression tools and library + * Copyright (C) 2002, 2006, 2007. Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +#ifndef __XDELTA3_LIST__ +#define __XDELTA3_LIST__ + +#define XD3_MAKELIST(LTYPE,ETYPE,LNAME) \ + \ +static inline ETYPE* \ +LTYPE ## _entry (LTYPE* l) \ +{ \ + return (ETYPE*) ((char*) l - (ptrdiff_t) &((ETYPE*) 0)->LNAME); \ +} \ + \ +static inline void \ +LTYPE ## _init (LTYPE *l) \ +{ \ + l->next = l; \ + l->prev = l; \ +} \ + \ +static inline void \ +LTYPE ## _add (LTYPE *prev, LTYPE *next, LTYPE *ins) \ +{ \ + next->prev = ins; \ + prev->next = ins; \ + ins->next = next; \ + ins->prev = prev; \ +} \ + \ +static inline void \ +LTYPE ## _push_back (LTYPE *l, ETYPE *i) \ +{ \ + LTYPE ## _add (l->prev, l, & i->LNAME); \ +} \ + \ +static inline void \ +LTYPE ## _del (LTYPE *next, \ + LTYPE *prev) \ +{ \ + next->prev = prev; \ + prev->next = next; \ +} \ + \ +static inline ETYPE* \ +LTYPE ## _remove (ETYPE *f) \ +{ \ + LTYPE *i = f->LNAME.next; \ + LTYPE ## _del (f->LNAME.next, f->LNAME.prev); \ + return LTYPE ## _entry (i); \ +} \ + \ +static inline ETYPE* \ +LTYPE ## _pop_back (LTYPE *l) \ +{ \ + LTYPE *i = l->prev; \ + LTYPE ## _del (i->next, i->prev); \ + return LTYPE ## _entry (i); \ +} \ + \ +static inline ETYPE* \ +LTYPE ## _pop_front (LTYPE *l) \ +{ \ + LTYPE *i = l->next; \ + LTYPE ## _del (i->next, i->prev); \ + return LTYPE ## _entry (i); \ +} \ + \ +static inline int \ +LTYPE ## _empty (LTYPE *l) \ +{ \ + return l == l->next; \ +} \ + \ +static inline ETYPE* \ +LTYPE ## _front (LTYPE *f) \ +{ \ + return LTYPE ## _entry (f->next); \ +} \ + \ +static inline ETYPE* \ +LTYPE ## _back (LTYPE *f) \ +{ \ + return LTYPE ## _entry (f->prev); \ +} \ + \ +static inline int \ +LTYPE ## _end (LTYPE *f, ETYPE *i) \ +{ \ + return f == & i->LNAME; \ +} \ + \ +static inline ETYPE* \ +LTYPE ## _next (ETYPE *f) \ +{ \ + return LTYPE ## _entry (f->LNAME.next); \ +} \ + \ +static inline usize_t \ +LTYPE ## _length (LTYPE *l) \ +{ \ + LTYPE *p; \ + int c = 0; \ + \ + for (p = l->next; p != l; p = p->next) \ + { \ + c += 1; \ + } \ + \ + return c; \ +} \ + \ +typedef int unused_ ## LTYPE + +#endif diff --git a/3rdparty/xdelta3-3.0.8/xdelta3-lzma.h b/3rdparty/xdelta3-3.0.8/xdelta3-lzma.h new file mode 100644 index 0000000..5be8407 --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3-lzma.h @@ -0,0 +1,196 @@ +/* xdelta 3 - delta compression tools and library + * Copyright (C) 2012, 2013. Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +/* Note: The use of the _easy_ decoder means we're not calling the + * xd3_stream malloc hooks. TODO(jmacd) Fix if anyone cares. */ + +#ifndef _XDELTA3_LZMA_H_ +#define _XDELTA3_LZMA_H_ + +#include + +typedef struct _xd3_lzma_stream xd3_lzma_stream; + +struct _xd3_lzma_stream { + lzma_stream lzma; + lzma_options_lzma options; + lzma_filter filters[2]; +}; + +xd3_sec_stream* +xd3_lzma_alloc (xd3_stream *stream) +{ + return (xd3_sec_stream*) xd3_alloc (stream, sizeof (xd3_lzma_stream), 1); +} + +void +xd3_lzma_destroy (xd3_stream *stream, xd3_sec_stream *sec_stream) +{ + xd3_lzma_stream *ls = (xd3_lzma_stream*) sec_stream; + lzma_end (&ls->lzma); + xd3_free (stream, ls); +} + +int +xd3_lzma_init (xd3_stream *stream, xd3_lzma_stream *sec, int is_encode) +{ + int ret; + + memset (&sec->lzma, 0, sizeof(sec->lzma)); + + if (is_encode) + { + int preset = (stream->flags & XD3_COMPLEVEL_MASK) >> XD3_COMPLEVEL_SHIFT; + + if (lzma_lzma_preset(&sec->options, preset)) + { + stream->msg = "invalid lzma preset"; + return XD3_INVALID; + } + + sec->filters[0].id = LZMA_FILTER_LZMA2; + sec->filters[0].options = &sec->options; + sec->filters[1].id = LZMA_VLI_UNKNOWN; + + ret = lzma_stream_encoder (&sec->lzma, &sec->filters[0], LZMA_CHECK_NONE); + } + else + { + ret = lzma_stream_decoder (&sec->lzma, UINT64_MAX, LZMA_TELL_NO_CHECK); + } + + if (ret != LZMA_OK) + { + stream->msg = "lzma stream init failed"; + return XD3_INTERNAL; + } + + return 0; +} + +int xd3_decode_lzma (xd3_stream *stream, xd3_lzma_stream *sec, + const uint8_t **input_pos, + const uint8_t *const input_end, + uint8_t **output_pos, + const uint8_t *const output_end) +{ + uint8_t *output = *output_pos; + const uint8_t *input = *input_pos; + size_t avail_in = input_end - input; + size_t avail_out = output_end - output; + + sec->lzma.avail_in = avail_in; + sec->lzma.next_in = input; + sec->lzma.avail_out = avail_out; + sec->lzma.next_out = output; + + while (1) + { + int lret = lzma_code (&sec->lzma, LZMA_RUN); + + switch (lret) + { + case LZMA_NO_CHECK: + case LZMA_OK: + if (sec->lzma.avail_out == 0) + { + (*output_pos) = sec->lzma.next_out; + (*input_pos) = sec->lzma.next_in; + return 0; + } + break; + + default: + stream->msg = "lzma decoding error"; + return XD3_INTERNAL; + } + } +} + +#if XD3_ENCODER + +int xd3_encode_lzma (xd3_stream *stream, + xd3_lzma_stream *sec, + xd3_output *input, + xd3_output *output, + xd3_sec_cfg *cfg) + +{ + lzma_action action = LZMA_RUN; + + cfg->inefficient = 1; /* Can't skip windows */ + sec->lzma.next_in = NULL; + sec->lzma.avail_in = 0; + sec->lzma.next_out = (output->base + output->next); + sec->lzma.avail_out = (output->avail - output->next); + + while (1) + { + int lret; + size_t nwrite; + if (sec->lzma.avail_in == 0 && input != NULL) + { + sec->lzma.avail_in = input->next; + sec->lzma.next_in = input->base; + + if ((input = input->next_page) == NULL) + { + action = LZMA_SYNC_FLUSH; + } + } + + lret = lzma_code (&sec->lzma, action); + + nwrite = (output->avail - output->next) - sec->lzma.avail_out; + + if (nwrite != 0) + { + output->next += nwrite; + + if (output->next == output->avail) + { + if ((output = xd3_alloc_output (stream, output)) == NULL) + { + return ENOMEM; + } + + sec->lzma.next_out = output->base; + sec->lzma.avail_out = output->avail; + } + } + + switch (lret) + { + case LZMA_OK: + break; + + case LZMA_STREAM_END: + return 0; + + default: + stream->msg = "lzma encoding error"; + return XD3_INTERNAL; + } + } + + return 0; +} + +#endif /* XD3_ENCODER */ + +#endif /* _XDELTA3_LZMA_H_ */ diff --git a/3rdparty/xdelta3-3.0.8/xdelta3-main.h b/3rdparty/xdelta3-3.0.8/xdelta3-main.h new file mode 100644 index 0000000..090b7d9 --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3-main.h @@ -0,0 +1,4048 @@ +/* xdelta3 - delta compression tools and library + * Copyright (C) 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, + * 2009, 2010, 2011, 2012, 2013 Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +/* This is all the extra stuff you need for convenience to users in a + * command line application. It contains these major components: + * + * 1. VCDIFF tools 2. external compression support (this is + * POSIX-specific). 3. a general read/write loop that handles all of + * the Xdelta decode/encode/VCDIFF-print functions 4. command-line + * interpreter 5. an Xdelta application header which stores default + * filename, external compression settings 6. output/error printing + * 7. basic file support and OS interface + */ + +/* TODO list: 1. do exact gzip-like filename, stdout handling. make a + * .vcdiff extension, refuse to encode to stdout without -cf, etc. + * 2. Allow the user to add a comment string to the app header without + * disturbing the default behavior. + */ + +/* On error handling and printing: + * + * The xdelta library sets stream->msg to indicate what condition + * caused an internal failure, but many failures originate here and + * are printed here. The return convention is 0 for success, as + * throughout Xdelta code, but special attention is required here for + * the operating system calls with different error handling. See the + * main_file_* routines. All errors in this file have a message + * printed at the time of occurance. Since some of these calls occur + * within calls to the library, the error may end up being printed + * again with a more general error message. + */ + +/*********************************************************************/ + +#ifndef XD3_POSIX +#define XD3_POSIX 0 +#endif +#ifndef XD3_STDIO +#define XD3_STDIO 0 +#endif +#ifndef XD3_WIN32 +#define XD3_WIN32 0 +#endif +#ifndef NOT_MAIN +#define NOT_MAIN 0 +#endif + +/* Combines xd3_strerror() and strerror() */ +const char* xd3_mainerror(int err_num); + +#include "xdelta3-internal.h" + +int +xsnprintf_func (char *str, int n, const char *fmt, ...) +{ + va_list a; + int ret; + va_start (a, fmt); + ret = vsnprintf_func (str, n, fmt, a); + va_end (a); + if (ret < 0) + { + ret = n; + } + return ret; +} + +/* If none are set, default to posix. */ +#if (XD3_POSIX + XD3_STDIO + XD3_WIN32) == 0 +#undef XD3_POSIX +#define XD3_POSIX 1 +#endif + +/* Handle externally-compressed inputs. */ +#ifndef EXTERNAL_COMPRESSION +#define EXTERNAL_COMPRESSION 1 +#endif + +#define PRINTHDR_SPECIAL -4378291 + +/* The number of soft-config variables. */ +#define XD3_SOFTCFG_VARCNT 7 + +/* this is used as in XPR(NT XD3_LIB_ERRMSG (stream, ret)) to print an + * error message from the library. */ +#define XD3_LIB_ERRMSG(stream, ret) "%s: %s\n", \ + xd3_errstring (stream), xd3_mainerror (ret) + +#if XD3_POSIX +#include /* close, read, write... */ +#include +#include +#endif + +#ifndef _WIN32 +#include /* lots */ +#include /* gettimeofday() */ +#include /* stat() and fstat() */ +#else +#if defined(_MSC_VER) +#define strtoll _strtoi64 +#endif +#include +#include +#ifndef WIFEXITED +# define WIFEXITED(stat) (((*((int *) &(stat))) & 0xff) == 0) +#endif +#ifndef WEXITSTATUS +# define WEXITSTATUS(stat) (((*((int *) &(stat))) >> 8) & 0xff) +#endif +#ifndef S_ISREG +//# ifdef S_IFREG +//# define S_ISREG(m) (((m) & S_IFMT) == S_IFREG) +//# else +# define S_ISREG(m) 1 +//# endif +#endif /* !S_ISREG */ + +// For standard input/output handles +static STARTUPINFO winStartupInfo; +#endif + +/********************************************************************** + ENUMS and TYPES + *********************************************************************/ + +/* These flags (mainly pertaining to main_read() operations) are set + * in the main_file->flags variable. All are related to with external + * decompression support. + * + * RD_FIRST causes the external decompression check when the input is + * first read. + * + * RD_NONEXTERNAL disables external decompression for reading a + * compressed input, in the case of Xdelta inputs. Note: Xdelta is + * supported as an external compression type, which makes is the + * reason for this flag. An example to justify this is: to create a + * delta between two files that are VCDIFF-compressed. Two external + * Xdelta decoders are run to supply decompressed source and target + * inputs to the Xdelta encoder. */ +typedef enum +{ + RD_FIRST = (1 << 0), + RD_NONEXTERNAL = (1 << 1), + RD_DECOMPSET = (1 << 2), + RD_MAININPUT = (1 << 3), +} xd3_read_flags; + +/* Main commands. For example, CMD_PRINTHDR is the "xdelta printhdr" + * command. */ +typedef enum +{ + CMD_NONE = 0, + CMD_PRINTHDR, + CMD_PRINTHDRS, + CMD_PRINTDELTA, + CMD_RECODE, + CMD_MERGE_ARG, + CMD_MERGE, +#if XD3_ENCODER + CMD_ENCODE, +#endif + CMD_DECODE, + CMD_TEST, + CMD_CONFIG, +} xd3_cmd; + +#if XD3_ENCODER +#define CMD_DEFAULT CMD_ENCODE +#define IS_ENCODE(cmd) (cmd == CMD_ENCODE) +#else +#define CMD_DEFAULT CMD_DECODE +#define IS_ENCODE(cmd) (0) +#endif + +typedef struct _main_merge main_merge; +typedef struct _main_merge_list main_merge_list; + +/* Various strings and magic values used to detect and call external + * compression. See below for examples. */ +struct _main_extcomp +{ + const char *recomp_cmdname; + const char *recomp_options; + + const char *decomp_cmdname; + const char *decomp_options; + + const char *ident; + const char *magic; + usize_t magic_size; + int flags; +}; + +/* Merge state: */ + +struct _main_merge_list +{ + main_merge_list *next; + main_merge_list *prev; +}; + +struct _main_merge +{ + const char *filename; + + main_merge_list link; +}; + +XD3_MAKELIST(main_merge_list,main_merge,link); + +/* TODO: really need to put options in a struct so that internal + * callers can easily reset state. */ + +#define DEFAULT_VERBOSE 0 + +/* Program options: various command line flags and options. */ +static int option_stdout = 0; +static int option_force = 0; +static int option_verbose = DEFAULT_VERBOSE; +static int option_quiet = 0; +static int option_use_appheader = 1; +static uint8_t* option_appheader = NULL; +static int option_use_secondary = 0; +static const char* option_secondary = NULL; +static int option_use_checksum = 1; +static int option_use_altcodetable = 0; +static const char* option_smatch_config = NULL; +static int option_no_compress = 0; +static int option_no_output = 0; /* do not write output */ +static const char *option_source_filename = NULL; + +static int option_level = XD3_DEFAULT_LEVEL; +static usize_t option_iopt_size = XD3_DEFAULT_IOPT_SIZE; +static usize_t option_winsize = XD3_DEFAULT_WINSIZE; +/* Note: option_srcwinsz is restricted from [16Kb, 4Gb], because + * addresses in the large hash checksum are 32 bits. The flag is read + * as xoff_t, so that 4Gb != 0. */ +static xoff_t option_srcwinsz = XD3_DEFAULT_SRCWINSZ; +static usize_t option_sprevsz = XD3_DEFAULT_SPREVSZ; + +/* These variables are supressed to avoid their use w/o support. main() warns + * appropriately when external compression is not enabled. */ +#if EXTERNAL_COMPRESSION +static int num_subprocs = 0; +static int option_force2 = 0; +static int option_decompress_inputs = 1; +static int option_recompress_outputs = 1; +#endif + +/* This is for comparing "printdelta" output without attention to + * copy-instruction modes. */ +#if VCDIFF_TOOLS +static int option_print_cpymode = 1; /* Note: see reset_defaults(). */ +#endif + +/* Static variables */ +IF_DEBUG(static int main_mallocs = 0;) + +static char* program_name = NULL; +static uint8_t* appheader_used = NULL; +static uint8_t* main_bdata = NULL; +static usize_t main_bsize = 0; + +/* Hacks for VCDIFF tools, recode command. */ +static int allow_fake_source = 0; + +/* recode_stream is used by both recode/merge for reading vcdiff inputs */ +static xd3_stream *recode_stream = NULL; + +/* merge_stream is used by merge commands for storing the source encoding */ +static xd3_stream *merge_stream = NULL; + +/* This array of compressor types is compiled even if EXTERNAL_COMPRESSION is + * false just so the program knows the mapping of IDENT->NAME. */ +static main_extcomp extcomp_types[] = +{ + { "bzip2", "-c", "bzip2", "-dc", "B", "BZh", 3, 0 }, + { "gzip", "-c", "gzip", "-dc", "G", "\037\213", 2, 0 }, + { "compress", "-c", "uncompress", "-c", "Z", "\037\235", 2, 0 }, + + /* Xz is lzma with a magic number http://tukaani.org/xz/format.html */ + { "xz", "-c", "xz", "-dc", "Y", "\xfd\x37\x7a\x58\x5a\x00", 2, 0 }, +}; + +static int main_input (xd3_cmd cmd, main_file *ifile, + main_file *ofile, main_file *sfile); +static void main_get_appheader (xd3_stream *stream, main_file *ifile, + main_file *output, main_file *sfile); + +static int main_getblk_func (xd3_stream *stream, + xd3_source *source, + xoff_t blkno); +static void main_free (void *ptr); +static void* main_malloc (size_t size); + +static int main_file_stat (main_file *xfile, xoff_t *size); +static int main_file_seek (main_file *xfile, xoff_t pos); +static int main_read_primary_input (main_file *file, + uint8_t *buf, + size_t size, + size_t *nread); + +static const char* main_format_bcnt (xoff_t r, shortbuf *buf); +static int main_help (void); + +#if XD3_ENCODER +static int xd3_merge_input_output (xd3_stream *stream, + xd3_whole_state *source); +#endif + +/* The code in xdelta3-blk.h is essentially part of this unit, see + * comments there. */ +#include "xdelta3-blkcache.h" + +void (*xprintf_message_func)(const char*msg) = NULL; + +void +xprintf (const char *fmt, ...) +{ + char buf[1000]; + va_list a; + int size; + va_start (a, fmt); + size = vsnprintf_func (buf, 1000, fmt, a); + va_end (a); + if (size < 0) + { + size = sizeof(buf) - 1; + buf[size] = 0; + } + if (xprintf_message_func != NULL) { + xprintf_message_func(buf); + } else { + size_t ignore = fwrite(buf, 1, size, stderr); + (void) ignore; + } +} + +static int +main_version (void) +{ + /* $Format: " XPR(NTR \"Xdelta version $Xdelta3Version$, Copyright (C) 2007, 2008, 2009, 2010, 2011, 2012, 2013, Joshua MacDonald\\n\");" $ */ + XPR(NTR "Xdelta version 3.0.8, Copyright (C) 2007, 2008, 2009, 2010, 2011, 2012, 2013 Joshua MacDonald\n"); + XPR(NTR "Xdelta comes with ABSOLUTELY NO WARRANTY.\n"); + XPR(NTR "This is free software, and you are welcome to redistribute it\n"); + XPR(NTR "under certain conditions; see \"COPYING\" for details.\n"); + return EXIT_SUCCESS; +} + +static int +main_config (void) +{ + main_version (); + + XPR(NTR "EXTERNAL_COMPRESSION=%d\n", EXTERNAL_COMPRESSION); + XPR(NTR "GENERIC_ENCODE_TABLES=%d\n", GENERIC_ENCODE_TABLES); + XPR(NTR "GENERIC_ENCODE_TABLES_COMPUTE=%d\n", GENERIC_ENCODE_TABLES_COMPUTE); + XPR(NTR "REGRESSION_TEST=%d\n", REGRESSION_TEST); + XPR(NTR "SECONDARY_DJW=%d\n", SECONDARY_DJW); + XPR(NTR "SECONDARY_FGK=%d\n", SECONDARY_FGK); + XPR(NTR "SECONDARY_LZMA=%d\n", SECONDARY_LZMA); + XPR(NTR "UNALIGNED_OK=%d\n", UNALIGNED_OK); + XPR(NTR "VCDIFF_TOOLS=%d\n", VCDIFF_TOOLS); + XPR(NTR "XD3_ALLOCSIZE=%d\n", XD3_ALLOCSIZE); + XPR(NTR "XD3_DEBUG=%d\n", XD3_DEBUG); + XPR(NTR "XD3_ENCODER=%d\n", XD3_ENCODER); + XPR(NTR "XD3_POSIX=%d\n", XD3_POSIX); + XPR(NTR "XD3_STDIO=%d\n", XD3_STDIO); + XPR(NTR "XD3_WIN32=%d\n", XD3_WIN32); + XPR(NTR "XD3_USE_LARGEFILE64=%d\n", XD3_USE_LARGEFILE64); + XPR(NTR "XD3_DEFAULT_LEVEL=%d\n", XD3_DEFAULT_LEVEL); + XPR(NTR "XD3_DEFAULT_IOPT_SIZE=%d\n", XD3_DEFAULT_IOPT_SIZE); + XPR(NTR "XD3_DEFAULT_SPREVSZ=%d\n", XD3_DEFAULT_SPREVSZ); + XPR(NTR "XD3_DEFAULT_SRCWINSZ=%d\n", XD3_DEFAULT_SRCWINSZ); + XPR(NTR "XD3_DEFAULT_WINSIZE=%d\n", XD3_DEFAULT_WINSIZE); + XPR(NTR "XD3_HARDMAXWINSIZE=%d\n", XD3_HARDMAXWINSIZE); + XPR(NTR "sizeof(void*)=%d\n", (int)sizeof(void*)); + XPR(NTR "sizeof(int)=%d\n", (int)sizeof(int)); + XPR(NTR "sizeof(size_t)=%d\n", (int)sizeof(size_t)); + XPR(NTR "sizeof(uint32_t)=%d\n", (int)sizeof(uint32_t)); + XPR(NTR "sizeof(uint64_t)=%d\n", (int)sizeof(uint64_t)); + XPR(NTR "sizeof(usize_t)=%d\n", (int)sizeof(usize_t)); + XPR(NTR "sizeof(xoff_t)=%d\n", (int)sizeof(xoff_t)); + + return EXIT_SUCCESS; +} + +static void +reset_defaults(void) +{ + option_stdout = 0; + option_force = 0; + option_verbose = DEFAULT_VERBOSE; + option_quiet = 0; + option_appheader = NULL; + option_use_secondary = 0; + option_secondary = NULL; + option_use_altcodetable = 0; + option_smatch_config = NULL; + option_no_compress = 0; + option_no_output = 0; + option_source_filename = NULL; + program_name = NULL; + appheader_used = NULL; + main_bdata = NULL; + main_bsize = 0; + allow_fake_source = 0; + option_smatch_config = NULL; + + main_lru_reset(); + + option_use_appheader = 1; + option_use_checksum = 1; +#if EXTERNAL_COMPRESSION + option_force2 = 0; + option_decompress_inputs = 1; + option_recompress_outputs = 1; + num_subprocs = 0; +#endif +#if VCDIFF_TOOLS + option_print_cpymode = 1; +#endif + option_level = XD3_DEFAULT_LEVEL; + option_iopt_size = XD3_DEFAULT_IOPT_SIZE; + option_winsize = XD3_DEFAULT_WINSIZE; + option_srcwinsz = XD3_DEFAULT_SRCWINSZ; + option_sprevsz = XD3_DEFAULT_SPREVSZ; +} + +static void* +main_malloc1 (size_t size) +{ + void* r = malloc (size); + if (r == NULL) { XPR(NT "malloc: %s\n", xd3_mainerror (ENOMEM)); } + return r; +} + +void* main_bufalloc (size_t size) { +#if XD3_WIN32 + return VirtualAlloc(NULL, size, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE); +#else + return main_malloc1(size); +#endif +} + +static void* +main_malloc (size_t size) +{ + void *r = main_malloc1 (size); + if (r) { IF_DEBUG (main_mallocs += 1); } + return r; +} + +static void* +main_alloc (void *opaque, + size_t items, + usize_t size) +{ + return main_malloc1 (items * size); +} + +static void +main_free1 (void *opaque, void *ptr) +{ + free (ptr); +} + +static void +main_free (void *ptr) +{ + if (ptr) + { + IF_DEBUG (main_mallocs -= 1); + main_free1 (NULL, ptr); + IF_DEBUG (XD3_ASSERT(main_mallocs >= 0)); + } +} + +void main_buffree (void *ptr) { +#if XD3_WIN32 + VirtualFree(ptr, 0, MEM_RELEASE); +#else + main_free1(NULL, ptr); +#endif +} + +/* This ensures that (ret = errno) always indicates failure, in case errno was + * accidentally not set. If this prints there's a bug somewhere. */ +static int +get_errno (void) +{ +#ifndef _WIN32 + if (errno == 0) + { + XPR(NT "you found a bug: expected errno != 0\n"); + errno = XD3_INTERNAL; + } + return errno; +#else + DWORD err_num = GetLastError(); + if (err_num == NO_ERROR) + { + err_num = XD3_INTERNAL; + } + return err_num; +#endif +} + +const char* +xd3_mainerror(int err_num) { +#ifndef _WIN32 + const char* x = xd3_strerror (err_num); + if (x != NULL) + { + return x; + } + return strerror(err_num); +#else + static char err_buf[256]; + const char* x = xd3_strerror (err_num); + if (x != NULL) + { + return x; + } + memset (err_buf, 0, 256); + FormatMessage (FORMAT_MESSAGE_FROM_SYSTEM | + FORMAT_MESSAGE_IGNORE_INSERTS, + NULL, err_num, + MAKELANGID (LANG_NEUTRAL, SUBLANG_DEFAULT), + err_buf, 256, NULL); + if (err_buf[0] != 0 && err_buf[strlen(err_buf) - 1] == '\n') + { + err_buf[strlen(err_buf) - 1] = 0; + } + return err_buf; +#endif +} + +static long +get_millisecs_now (void) +{ +#ifndef _WIN32 + struct timeval tv; + + gettimeofday (& tv, NULL); + + return (tv.tv_sec) * 1000L + (tv.tv_usec) / 1000; +#else + SYSTEMTIME st; + FILETIME ft; + __int64 *pi = (__int64*)&ft; + GetLocalTime(&st); + SystemTimeToFileTime(&st, &ft); + return (long)((*pi) / 10000); +#endif +} + +/* Always >= 1 millisec, right? */ +static long +get_millisecs_since (void) +{ + static long last = 0; + long now = get_millisecs_now(); + long diff = now - last; + last = now; + return diff; +} + +static const char* +main_format_bcnt (xoff_t r, shortbuf *buf) +{ + static const char* fmts[] = { "B", "KiB", "MiB", "GiB", "TiB", "PiB", "EiB" }; + usize_t i; + + for (i = 0; i < SIZEOF_ARRAY(fmts) - 1; i += 1) + { + xoff_t new_r; + + if (r == 0) + { + short_sprintf (*buf, "0 %s", fmts[i]); + return buf->buf; + } + + if (r >= 1 && r < 10) + { + short_sprintf (*buf, "%.2f %s", (double) r, fmts[i]); + return buf->buf; + } + + if (r >= 10 && r < 100) + { + short_sprintf (*buf, "%.1f %s", (double) r, fmts[i]); + return buf->buf; + } + + if (r >= 100 && r < 1000) + { + short_sprintf (*buf, "%"Q"u %s", r, fmts[i]); + return buf->buf; + } + + new_r = r / 1024; + + if (new_r < 10) + { + short_sprintf (*buf, "%.2f %s", (double) r / 1024.0, fmts[i + 1]); + return buf->buf; + } + + if (new_r < 100) + { + short_sprintf (*buf, "%.1f %s", (double) r / 1024.0, fmts[i + 1]); + return buf->buf; + } + + r = new_r; + } + XD3_ASSERT (0); + return ""; +} + +static char* +main_format_rate (xoff_t bytes, long millis, shortbuf *buf) +{ + xoff_t r = (xoff_t)(1.0 * bytes / (1.0 * millis / 1000.0)); + static shortbuf lbuf; + + main_format_bcnt (r, &lbuf); + short_sprintf (*buf, "%s/s", lbuf.buf); + return buf->buf; +} + +static char* +main_format_millis (long millis, shortbuf *buf) +{ + if (millis < 1000) + { + short_sprintf (*buf, "%lu ms", millis); + } + else if (millis < 10000) + { + short_sprintf (*buf, "%.1f sec", millis / 1000.0); + } + else + { + short_sprintf (*buf, "%lu sec", millis / 1000L); + } + return buf->buf; +} + +/* A safe version of strtol for xoff_t. */ +static int +main_strtoxoff (const char* s, xoff_t *xo, char which) +{ + char *e; + xoff_t x; + + XD3_ASSERT(s && *s != 0); + + { + /* Should check LONG_MIN, LONG_MAX, LLONG_MIN, LLONG_MAX? */ +#if SIZEOF_XOFF_T == 4 + long xx = strtol (s, &e, 0); +#else + long long xx = strtoll (s, &e, 0); +#endif + + if (xx < 0) + { + XPR(NT "-%c: negative integer: %s\n", which, s); + return EXIT_FAILURE; + } + + x = xx; + } + + if (*e != 0) + { + XPR(NT "-%c: invalid integer: %s\n", which, s); + return EXIT_FAILURE; + } + + (*xo) = x; + return 0; +} + +static int +main_atoux (const char* arg, xoff_t *xo, xoff_t low, + xoff_t high, char which) +{ + xoff_t x; + int ret; + + if ((ret = main_strtoxoff (arg, & x, which))) { return ret; } + + if (x < low) + { + XPR(NT "-%c: minimum value: %"Q"u\n", which, low); + return EXIT_FAILURE; + } + if (high != 0 && x > high) + { + XPR(NT "-%c: maximum value: %"Q"u\n", which, high); + return EXIT_FAILURE; + } + (*xo) = x; + return 0; +} + +static int +main_atou (const char* arg, usize_t *uo, usize_t low, + usize_t high, char which) +{ + int ret; + xoff_t xo; + if ((ret = main_atoux (arg, &xo, low, high, which))) + { + return ret; + } + *uo = (usize_t)xo; + return 0; +} + +/****************************************************************** + FILE BASICS + ******************************************************************/ + +/* With all the variation in file system-call semantics, arguments, + * return values and error-handling for the POSIX and STDIO file APIs, + * the insides of these functions make me sick, which is why these + * wrappers exist. */ + +#define XOPEN_OPNAME (xfile->mode == XO_READ ? "read" : "write") +#define XOPEN_STDIO (xfile->mode == XO_READ ? "rb" : "wb") +#define XOPEN_POSIX (xfile->mode == XO_READ ? \ + O_RDONLY : O_WRONLY | O_CREAT | O_TRUNC) +#define XOPEN_MODE (xfile->mode == XO_READ ? 0 : 0666) + +#define XF_ERROR(op, name, ret) \ + do { if (!option_quiet) { XPR(NT "file %s failed: %s: %s: %s\n", (op), \ + XOPEN_OPNAME, (name), xd3_mainerror (ret)); } } while (0) + +#if XD3_STDIO +#define XFNO(f) fileno(f->file) +#define XSTDOUT_XF(f) { (f)->file = stdout; (f)->filename = "/dev/stdout"; } +#define XSTDIN_XF(f) { (f)->file = stdin; (f)->filename = "/dev/stdin"; } + +#elif XD3_POSIX +#define XFNO(f) f->file +#define XSTDOUT_XF(f) \ + { (f)->file = STDOUT_FILENO; (f)->filename = "/dev/stdout"; } +#define XSTDIN_XF(f) \ + { (f)->file = STDIN_FILENO; (f)->filename = "/dev/stdin"; } + +#elif XD3_WIN32 +#define XFNO(f) -1 +#define XSTDOUT_XF(f) { \ + (f)->file = GetStdHandle(STD_OUTPUT_HANDLE); \ + (f)->filename = "(stdout)"; \ + } +#define XSTDIN_XF(f) { \ + (f)->file = GetStdHandle(STD_INPUT_HANDLE); \ + (f)->filename = "(stdin)"; \ + } +#endif + +void +main_file_init (main_file *xfile) +{ + memset (xfile, 0, sizeof (*xfile)); + +#if XD3_POSIX + xfile->file = -1; +#endif +#if XD3_WIN32 + xfile->file = INVALID_HANDLE_VALUE; +#endif +} + +int +main_file_isopen (main_file *xfile) +{ +#if XD3_STDIO + return xfile->file != NULL; + +#elif XD3_POSIX + return xfile->file != -1; + +#elif XD3_WIN32 + return xfile->file != INVALID_HANDLE_VALUE; +#endif +} + +int +main_file_close (main_file *xfile) +{ + int ret = 0; + + if (! main_file_isopen (xfile)) + { + return 0; + } + +#if XD3_STDIO + ret = fclose (xfile->file); + xfile->file = NULL; + +#elif XD3_POSIX + ret = close (xfile->file); + xfile->file = -1; + +#elif XD3_WIN32 + if (!CloseHandle(xfile->file)) { + ret = get_errno (); + } + xfile->file = INVALID_HANDLE_VALUE; +#endif + + if (ret != 0) { XF_ERROR ("close", xfile->filename, ret = get_errno ()); } + return ret; +} + +void +main_file_cleanup (main_file *xfile) +{ + XD3_ASSERT (xfile != NULL); + + if (main_file_isopen (xfile)) + { + main_file_close (xfile); + } + + if (xfile->snprintf_buf != NULL) + { + main_free(xfile->snprintf_buf); + xfile->snprintf_buf = NULL; + } + + if (xfile->filename_copy != NULL) + { + main_free(xfile->filename_copy); + xfile->filename_copy = NULL; + } +} + +int +main_file_open (main_file *xfile, const char* name, int mode) +{ + int ret = 0; + + xfile->mode = mode; + + XD3_ASSERT (name != NULL); + XD3_ASSERT (! main_file_isopen (xfile)); + if (name[0] == 0) + { + XPR(NT "invalid file name: empty string\n"); + return XD3_INVALID; + } + +#if XD3_STDIO + xfile->file = fopen (name, XOPEN_STDIO); + + ret = (xfile->file == NULL) ? get_errno () : 0; + +#elif XD3_POSIX + /* TODO: Should retry this call if interrupted, similar to read/write */ + if ((ret = open (name, XOPEN_POSIX, XOPEN_MODE)) < 0) + { + ret = get_errno (); + } + else + { + xfile->file = ret; + ret = 0; + } + +#elif XD3_WIN32 + xfile->file = CreateFile(name, + (mode == XO_READ) ? GENERIC_READ : GENERIC_WRITE, + FILE_SHARE_READ, + NULL, + (mode == XO_READ) ? + OPEN_EXISTING : + (option_force ? CREATE_ALWAYS : CREATE_NEW), + FILE_ATTRIBUTE_NORMAL, + NULL); + if (xfile->file == INVALID_HANDLE_VALUE) + { + ret = get_errno (); + } +#endif + if (ret) { XF_ERROR ("open", name, ret); } + else { xfile->realname = name; xfile->nread = 0; } + return ret; +} + +int +main_file_stat (main_file *xfile, xoff_t *size) +{ + int ret = 0; +#if XD3_WIN32 + if (GetFileType(xfile->file) != FILE_TYPE_DISK) + { + return -1; + } +# if (_WIN32_WINNT >= 0x0500) + { + LARGE_INTEGER li; + if (GetFileSizeEx(xfile->file, &li) == 0) + { + return get_errno (); + } + *size = li.QuadPart; + } +# else + { + DWORD filesize = GetFileSize(xfile->file, NULL); + if (filesize == INVALID_FILE_SIZE) + { + return get_errno () + } + *size = filesize; + } +# endif +#else + struct stat sbuf; + if (fstat (XFNO (xfile), & sbuf) < 0) + { + ret = get_errno (); + return ret; + } + + if (! S_ISREG (sbuf.st_mode)) + { + return ESPIPE; + } + (*size) = sbuf.st_size; +#endif + return ret; +} + +int +main_file_exists (main_file *xfile) +{ + struct stat sbuf; + return stat (xfile->filename, & sbuf) == 0 && S_ISREG (sbuf.st_mode); +} + +#if (XD3_POSIX || EXTERNAL_COMPRESSION) +/* POSIX-generic code takes a function pointer to read() or write(). + * This calls the function repeatedly until the buffer is full or EOF. + * The NREAD parameter is not set for write, NULL is passed. Return + * is signed, < 0 indicate errors, otherwise byte count. */ +typedef int (xd3_posix_func) (int fd, uint8_t *buf, usize_t size); + +static int +xd3_posix_io (int fd, uint8_t *buf, size_t size, + xd3_posix_func *func, size_t *nread) +{ + int ret; + size_t nproc = 0; + + while (nproc < size) + { + size_t tryread = min(size - nproc, 1U << 30); + ssize_t result = (*func) (fd, buf + nproc, tryread); + + if (result < 0) + { + ret = get_errno (); + if (ret != EAGAIN && ret != EINTR) + { + return ret; + } + continue; + } + + if (nread != NULL && result == 0) { break; } + + nproc += result; + } + if (nread != NULL) { (*nread) = nproc; } + return 0; +} +#endif + +#if XD3_WIN32 +static int +xd3_win32_io (HANDLE file, uint8_t *buf, size_t size, + int is_read, size_t *nread) +{ + int ret = 0; + size_t nproc = 0; + + while (nproc < size) + { + DWORD nproc2 = 0; /* hmm */ + DWORD nremain = size - nproc; + if ((is_read ? + ReadFile (file, buf + nproc, nremain, &nproc2, NULL) : + WriteFile (file, buf + nproc, nremain, &nproc2, NULL)) == 0) + { + ret = get_errno(); + if (ret != ERROR_HANDLE_EOF && ret != ERROR_BROKEN_PIPE) + { + return ret; + } + /* By falling through here, we'll break this loop in the + * read case in case of eof or broken pipe. */ + } + + nproc += nproc2; + + if (nread != NULL && nproc2 == 0) { break; } + } + if (nread != NULL) { (*nread) = nproc; } + return 0; +} +#endif + +/* POSIX is unbuffered, while STDIO is buffered. main_file_read() + * should always be called on blocks. */ +int +main_file_read (main_file *ifile, + uint8_t *buf, + size_t size, + size_t *nread, + const char *msg) +{ + int ret = 0; + +#if XD3_STDIO + size_t result; + + result = fread (buf, 1, size, ifile->file); + + if (result < size && ferror (ifile->file)) + { + ret = get_errno (); + } + else + { + *nread = result; + } + +#elif XD3_POSIX + ret = xd3_posix_io (ifile->file, buf, size, (xd3_posix_func*) &read, nread); +#elif XD3_WIN32 + ret = xd3_win32_io (ifile->file, buf, size, 1 /* is_read */, nread); +#endif + + if (ret) + { + XPR(NT "%s: %s: %s\n", msg, ifile->filename, xd3_mainerror (ret)); + } + else + { + if (option_verbose > 4) { XPR(NT "read %s: %zu bytes\n", + ifile->filename, (*nread)); } + ifile->nread += (*nread); + } + + return ret; +} + +int +main_file_write (main_file *ofile, uint8_t *buf, usize_t size, const char *msg) +{ + int ret = 0; + +#if XD3_STDIO + usize_t result; + + result = fwrite (buf, 1, size, ofile->file); + + if (result != size) { ret = get_errno (); } + +#elif XD3_POSIX + ret = xd3_posix_io (ofile->file, buf, size, (xd3_posix_func*) &write, NULL); + +#elif XD3_WIN32 + ret = xd3_win32_io (ofile->file, buf, size, 0, NULL); + +#endif + + if (ret) + { + XPR(NT "%s: %s: %s\n", msg, ofile->filename, xd3_mainerror (ret)); + } + else + { + if (option_verbose > 5) { XPR(NT "write %s: %u bytes\n", + ofile->filename, size); } + ofile->nwrite += size; + } + + return ret; +} + +static int +main_file_seek (main_file *xfile, xoff_t pos) +{ + int ret = 0; + +#if XD3_STDIO + if (fseek (xfile->file, pos, SEEK_SET) != 0) { ret = get_errno (); } + +#elif XD3_POSIX + if ((xoff_t) lseek (xfile->file, pos, SEEK_SET) != pos) + { ret = get_errno (); } + +#elif XD3_WIN32 +# if (_WIN32_WINNT >= 0x0500) + LARGE_INTEGER move, out; + move.QuadPart = pos; + if (SetFilePointerEx(xfile->file, move, &out, FILE_BEGIN) == 0) + { + ret = get_errno (); + } +# else + if (SetFilePointer(xfile->file, (LONG)pos, NULL, FILE_BEGIN) == + INVALID_SET_FILE_POINTER) + { + ret = get_errno (); + } +# endif +#endif + + return ret; +} + +/* This function simply writes the stream output buffer, if there is + * any, for encode, decode and recode commands. (The VCDIFF tools use + * main_print_func()). */ +static int +main_write_output (xd3_stream* stream, main_file *ofile) +{ + int ret; + + if (option_no_output) + { + return 0; + } + + if (stream->avail_out > 0 && + (ret = main_file_write (ofile, stream->next_out, + stream->avail_out, "write failed"))) + { + return ret; + } + + return 0; +} + +static int +main_set_secondary_flags (xd3_config *config) +{ + int ret; + if (option_use_secondary) + { + /* The default secondary compressor is DJW, if it's compiled. */ + if (option_secondary == NULL) + { + if (SECONDARY_DJW) + { + config->flags |= XD3_SEC_DJW; + } + } + else + { + if (strcmp (option_secondary, "fgk") == 0 && SECONDARY_FGK) + { + config->flags |= XD3_SEC_FGK; + } + else if (strcmp (option_secondary, "lzma") == 0 && SECONDARY_LZMA) + { + config->flags |= XD3_SEC_LZMA; + } + else if (strncmp (option_secondary, "djw", 3) == 0 && SECONDARY_DJW) + { + usize_t level = XD3_DEFAULT_SECONDARY_LEVEL; + + config->flags |= XD3_SEC_DJW; + + if (strlen (option_secondary) > 3 && + (ret = main_atou (option_secondary + 3, + &level, + 0, 9, 'S')) != 0 && + !option_quiet) + { + return XD3_INVALID; + } + + /* XD3_SEC_NOXXXX flags disable secondary compression on + * a per-section basis. For djw, ngroups=1 indicates + * minimum work, ngroups=0 uses default settings, which + * is > 1 groups by default. */ + if (level < 1) { config->flags |= XD3_SEC_NODATA; } + if (level < 7) { config->sec_data.ngroups = 1; } + else { config->sec_data.ngroups = 0; } + + if (level < 3) { config->flags |= XD3_SEC_NOINST; } + if (level < 8) { config->sec_inst.ngroups = 1; } + else { config->sec_inst.ngroups = 0; } + + if (level < 5) { config->flags |= XD3_SEC_NOADDR; } + if (level < 9) { config->sec_addr.ngroups = 1; } + else { config->sec_addr.ngroups = 0; } + } + else if (strcmp (option_secondary, "none") == 0 && SECONDARY_DJW) + { + /* No secondary */ + } + else + { + if (!option_quiet) + { + XPR(NT "unrecognized secondary compressor type: %s\n", + option_secondary); + return XD3_INVALID; + } + } + } + } + + return 0; +} + +/****************************************************************** + VCDIFF TOOLS + *****************************************************************/ + +#if VCDIFF_TOOLS +#include "xdelta3-merge.h" + +/* The following macros let VCDIFF print using main_file_write(), + * for example: + * + * VC(UT "trying to be portable: %d\n", x)VE; + */ +#define SNPRINTF_BUFSIZE 1024 +#define VC do { if (((ret = xsnprintf_func +#define UT (char*)xfile->snprintf_buf, SNPRINTF_BUFSIZE, +#define VE ) >= SNPRINTF_BUFSIZE \ + && (ret = main_print_overflow(ret)) != 0) \ + || (ret = main_file_write(xfile, xfile->snprintf_buf, \ + (usize_t)ret, "print")) != 0) \ + { return ret; } } while (0) + +static int +main_print_overflow (int x) +{ + XPR(NT "internal print buffer overflow: %d bytes\n", x); + return XD3_INTERNAL; +} + +/* This function prints a single VCDIFF window. */ +static int +main_print_window (xd3_stream* stream, main_file *xfile) +{ + int ret; + usize_t size = 0; + + VC(UT " Offset Code Type1 Size1 @Addr1 + Type2 Size2 @Addr2\n")VE; + + while (stream->inst_sect.buf < stream->inst_sect.buf_max) + { + usize_t code = stream->inst_sect.buf[0]; + const uint8_t *addr_before = stream->addr_sect.buf; + const uint8_t *inst_before = stream->inst_sect.buf; + usize_t addr_bytes; + usize_t inst_bytes; + usize_t size_before = size; + + if ((ret = xd3_decode_instruction (stream))) + { + XPR(NT "instruction decode error at %"Q"u: %s\n", + stream->dec_winstart + size, stream->msg); + return ret; + } + + addr_bytes = (usize_t)(stream->addr_sect.buf - addr_before); + inst_bytes = (usize_t)(stream->inst_sect.buf - inst_before); + + VC(UT " %06"Q"u %03u %s %6u", stream->dec_winstart + size, + option_print_cpymode ? code : 0, + xd3_rtype_to_string ((xd3_rtype) stream->dec_current1.type, + option_print_cpymode), + stream->dec_current1.size)VE; + + if (stream->dec_current1.type != XD3_NOOP) + { + if (stream->dec_current1.type >= XD3_CPY) + { + if (stream->dec_current1.addr >= stream->dec_cpylen) + { + VC(UT " T@%-6u", + stream->dec_current1.addr - stream->dec_cpylen)VE; + } + else + { + VC(UT " S@%-6"Q"u", + stream->dec_cpyoff + stream->dec_current1.addr)VE; + } + } + else + { + VC(UT " ")VE; + } + + size += stream->dec_current1.size; + } + + if (stream->dec_current2.type != XD3_NOOP) + { + VC(UT " %s %6u", + xd3_rtype_to_string ((xd3_rtype) stream->dec_current2.type, + option_print_cpymode), + stream->dec_current2.size)VE; + + if (stream->dec_current2.type >= XD3_CPY) + { + if (stream->dec_current2.addr >= stream->dec_cpylen) + { + VC(UT " T@%-6u", + stream->dec_current2.addr - stream->dec_cpylen)VE; + } + else + { + VC(UT " S@%-6"Q"u", + stream->dec_cpyoff + stream->dec_current2.addr)VE; + } + } + + size += stream->dec_current2.size; + } + + VC(UT "\n")VE; + + if (option_verbose && + addr_bytes + inst_bytes >= (size - size_before) && + (stream->dec_current1.type >= XD3_CPY || + stream->dec_current2.type >= XD3_CPY)) + { + VC(UT " %06"Q"u (inefficiency) %u encoded as %u bytes\n", + stream->dec_winstart + size_before, + size - size_before, + addr_bytes + inst_bytes)VE; + } + } + + if (stream->dec_tgtlen != size && (stream->flags & XD3_SKIP_WINDOW) == 0) + { + XPR(NT "target window size inconsistency"); + return XD3_INTERNAL; + } + + if (stream->dec_position != stream->dec_maxpos) + { + XPR(NT "target window position inconsistency"); + return XD3_INTERNAL; + } + + if (stream->addr_sect.buf != stream->addr_sect.buf_max) + { + XPR(NT "address section inconsistency"); + return XD3_INTERNAL; + } + + return 0; +} + +static int +main_print_vcdiff_file (main_file *xfile, main_file *file, const char *type) +{ + int ret; /* Used by above macros */ + if (file->filename) + { + VC(UT "XDELTA filename (%s): %s\n", type, + file->filename)VE; + } + if (file->compressor) + { + VC(UT "XDELTA ext comp (%s): %s\n", type, + file->compressor->recomp_cmdname)VE; + } + return 0; +} + +/* This function prints a VCDIFF input, mainly for debugging purposes. */ +static int +main_print_func (xd3_stream* stream, main_file *xfile) +{ + int ret; + + if (option_no_output) + { + return 0; + } + + if (xfile->snprintf_buf == NULL) + { + if ((xfile->snprintf_buf = + (uint8_t*)main_malloc(SNPRINTF_BUFSIZE)) == NULL) + { + return ENOMEM; + } + } + + if (stream->dec_winstart == 0) + { + VC(UT "VCDIFF version: 0\n")VE; + VC(UT "VCDIFF header size: %d\n", + stream->dec_hdrsize)VE; + VC(UT "VCDIFF header indicator: ")VE; + if ((stream->dec_hdr_ind & VCD_SECONDARY) != 0) + VC(UT "VCD_SECONDARY ")VE; + if ((stream->dec_hdr_ind & VCD_CODETABLE) != 0) + VC(UT "VCD_CODETABLE ")VE; + if ((stream->dec_hdr_ind & VCD_APPHEADER) != 0) + VC(UT "VCD_APPHEADER ")VE; + if (stream->dec_hdr_ind == 0) + VC(UT "none")VE; + VC(UT "\n")VE; + + IF_SEC(VC(UT "VCDIFF secondary compressor: %s\n", + stream->sec_type ? stream->sec_type->name : "none")VE); + IF_NSEC(VC(UT "VCDIFF secondary compressor: unsupported\n")VE); + + if (stream->dec_hdr_ind & VCD_APPHEADER) + { + uint8_t *apphead; + usize_t appheadsz; + ret = xd3_get_appheader (stream, & apphead, & appheadsz); + + if (ret == 0 && appheadsz > 0) + { + int sq = option_quiet; + main_file i, o, s; + XD3_ASSERT (apphead != NULL); + VC(UT "VCDIFF application header: ")VE; + if ((ret = main_file_write (xfile, apphead, + appheadsz, "print")) != 0) + { return ret; } + VC(UT "\n")VE; + + main_file_init (& i); + main_file_init (& o); + main_file_init (& s); + option_quiet = 1; + main_get_appheader (stream, &i, & o, & s); + option_quiet = sq; + if ((ret = main_print_vcdiff_file (xfile, & o, "output"))) + { return ret; } + if ((ret = main_print_vcdiff_file (xfile, & s, "source"))) + { return ret; } + main_file_cleanup (& i); + main_file_cleanup (& o); + main_file_cleanup (& s); + } + } + } + else + { + VC(UT "\n")VE; + } + + VC(UT "VCDIFF window number: %"Q"u\n", stream->current_window)VE; + VC(UT "VCDIFF window indicator: ")VE; + if ((stream->dec_win_ind & VCD_SOURCE) != 0) VC(UT "VCD_SOURCE ")VE; + if ((stream->dec_win_ind & VCD_TARGET) != 0) VC(UT "VCD_TARGET ")VE; + if ((stream->dec_win_ind & VCD_ADLER32) != 0) VC(UT "VCD_ADLER32 ")VE; + if (stream->dec_win_ind == 0) VC(UT "none")VE; + VC(UT "\n")VE; + + if ((stream->dec_win_ind & VCD_ADLER32) != 0) + { + VC(UT "VCDIFF adler32 checksum: %08X\n", + (usize_t)stream->dec_adler32)VE; + } + + if (stream->dec_del_ind != 0) + { + VC(UT "VCDIFF delta indicator: ")VE; + if ((stream->dec_del_ind & VCD_DATACOMP) != 0) VC(UT "VCD_DATACOMP ")VE; + if ((stream->dec_del_ind & VCD_INSTCOMP) != 0) VC(UT "VCD_INSTCOMP ")VE; + if ((stream->dec_del_ind & VCD_ADDRCOMP) != 0) VC(UT "VCD_ADDRCOMP ")VE; + if (stream->dec_del_ind == 0) VC(UT "none")VE; + VC(UT "\n")VE; + } + + if (stream->dec_winstart != 0) + { + VC(UT "VCDIFF window at offset: %"Q"u\n", stream->dec_winstart)VE; + } + + if (SRCORTGT (stream->dec_win_ind)) + { + VC(UT "VCDIFF copy window length: %u\n", + (usize_t)stream->dec_cpylen)VE; + VC(UT "VCDIFF copy window offset: %"Q"u\n", + stream->dec_cpyoff)VE; + } + + VC(UT "VCDIFF delta encoding length: %u\n", + (usize_t)stream->dec_enclen)VE; + VC(UT "VCDIFF target window length: %u\n", + (usize_t)stream->dec_tgtlen)VE; + + VC(UT "VCDIFF data section length: %u\n", + (usize_t)stream->data_sect.size)VE; + VC(UT "VCDIFF inst section length: %u\n", + (usize_t)stream->inst_sect.size)VE; + VC(UT "VCDIFF addr section length: %u\n", + (usize_t)stream->addr_sect.size)VE; + + ret = 0; + if ((stream->flags & XD3_JUST_HDR) != 0) + { + /* Print a header -- finished! */ + ret = PRINTHDR_SPECIAL; + } + else if ((stream->flags & XD3_SKIP_WINDOW) == 0) + { + ret = main_print_window (stream, xfile); + } + + return ret; +} + +static int +main_recode_copy (xd3_stream* stream, + xd3_output* output, + xd3_desect* input) +{ + int ret; + + XD3_ASSERT(output != NULL); + XD3_ASSERT(output->next_page == NULL); + + if ((ret = xd3_decode_allocate (recode_stream, + input->size, + &output->base, + &output->avail))) + { + XPR(NT XD3_LIB_ERRMSG (stream, ret)); + return ret; + } + + memcpy (output->base, + /* Note: decoder advances buf, so get base of buffer with + * buf_max - size */ + input->buf_max - input->size, + input->size); + output->next = input->size; + return 0; +} + +// Re-encode one window +static int +main_recode_func (xd3_stream* stream, main_file *ofile) +{ + int ret; + xd3_source decode_source; + + XD3_ASSERT(stream->dec_state == DEC_FINISH); + XD3_ASSERT(recode_stream->enc_state == ENC_INIT || + recode_stream->enc_state == ENC_INPUT); + + // Copy partial decoder output to partial encoder inputs + if ((ret = main_recode_copy (recode_stream, + DATA_HEAD(recode_stream), + &stream->data_sect)) || + (ret = main_recode_copy (recode_stream, + INST_HEAD(recode_stream), + &stream->inst_sect)) || + (ret = main_recode_copy (recode_stream, + ADDR_HEAD(recode_stream), + &stream->addr_sect))) + { + return ret; + } + + // This jumps to xd3_emit_hdr() + recode_stream->enc_state = ENC_FLUSH; + recode_stream->avail_in = stream->dec_tgtlen; + + if (SRCORTGT (stream->dec_win_ind)) + { + recode_stream->src = & decode_source; + decode_source.srclen = stream->dec_cpylen; + decode_source.srcbase = stream->dec_cpyoff; + } + + if (option_use_checksum && + (stream->dec_win_ind & VCD_ADLER32) != 0) + { + recode_stream->flags |= XD3_ADLER32_RECODE; + recode_stream->recode_adler32 = stream->dec_adler32; + } + + if (option_use_appheader != 0 && + option_appheader != NULL) + { + xd3_set_appheader (recode_stream, option_appheader, + (usize_t) strlen ((char*) option_appheader)); + } + else if (option_use_appheader != 0 && + option_appheader == NULL) + { + if (stream->dec_appheader != NULL) + { + xd3_set_appheader (recode_stream, + stream->dec_appheader, stream->dec_appheadsz); + } + } + + // Output loop + for (;;) + { + switch((ret = xd3_encode_input (recode_stream))) + { + case XD3_INPUT: { + /* finished recoding one window */ + stream->total_out = recode_stream->total_out; + return 0; + } + case XD3_OUTPUT: { + /* main_file_write below */ + break; + } + case XD3_GOTHEADER: + case XD3_WINSTART: + case XD3_WINFINISH: { + /* ignore */ + continue; + } + case XD3_GETSRCBLK: + case 0: { + return XD3_INTERNAL; + } + default: + return ret; + } + + if ((ret = main_write_output (recode_stream, ofile))) + { + return ret; + } + + xd3_consume_output (recode_stream); + } +} +#endif /* VCDIFF_TOOLS */ + +/******************************************************************* + VCDIFF merging + ******************************************************************/ + +#if VCDIFF_TOOLS +/* Modifies static state. */ +static int +main_init_recode_stream (void) +{ + int ret; + int stream_flags = XD3_ADLER32_NOVER | XD3_SKIP_EMIT; + int recode_flags; + xd3_config recode_config; + + XD3_ASSERT (recode_stream == NULL); + + if ((recode_stream = (xd3_stream*) main_malloc(sizeof(xd3_stream))) == NULL) + { + return ENOMEM; + } + + recode_flags = (stream_flags & XD3_SEC_TYPE); + + recode_config.alloc = main_alloc; + recode_config.freef = main_free1; + + xd3_init_config(&recode_config, recode_flags); + + if ((ret = main_set_secondary_flags (&recode_config)) || + (ret = xd3_config_stream (recode_stream, &recode_config)) || + (ret = xd3_encode_init_partial (recode_stream)) || + (ret = xd3_whole_state_init (recode_stream))) + { + XPR(NT XD3_LIB_ERRMSG (recode_stream, ret)); + xd3_free_stream (recode_stream); + recode_stream = NULL; + return ret; + } + + return 0; +} + +/* This processes the sequence of -m arguments. The final input + * is processed as part of the ordinary main_input() loop. */ +static int +main_merge_arguments (main_merge_list* merges) +{ + int ret = 0; + int count = 0; + main_merge *merge = NULL; + xd3_stream merge_input; + + if (main_merge_list_empty (merges)) + { + return 0; + } + + if ((ret = xd3_config_stream (& merge_input, NULL)) || + (ret = xd3_whole_state_init (& merge_input))) + { + XPR(NT XD3_LIB_ERRMSG (& merge_input, ret)); + return ret; + } + + merge = main_merge_list_front (merges); + while (!main_merge_list_end (merges, merge)) + { + main_file mfile; + main_file_init (& mfile); + mfile.filename = merge->filename; + mfile.flags = RD_NONEXTERNAL; + + if ((ret = main_file_open (& mfile, merge->filename, XO_READ))) + { + goto error; + } + + ret = main_input (CMD_MERGE_ARG, & mfile, NULL, NULL); + + if (ret == 0) + { + if (count++ == 0) + { + /* The first merge source is the next merge input. */ + xd3_swap_whole_state (& recode_stream->whole_target, + & merge_input.whole_target); + } + else + { + /* Merge the recode_stream with merge_input. */ + ret = xd3_merge_input_output (recode_stream, + & merge_input.whole_target); + + /* Save the next merge source in merge_input. */ + xd3_swap_whole_state (& recode_stream->whole_target, + & merge_input.whole_target); + } + } + + main_file_cleanup (& mfile); + + if (recode_stream != NULL) + { + xd3_free_stream (recode_stream); + main_free (recode_stream); + recode_stream = NULL; + } + + if (main_bdata != NULL) + { + main_buffree (main_bdata); + main_bdata = NULL; + main_bsize = 0; + } + + if (ret != 0) + { + goto error; + } + + merge = main_merge_list_next (merge); + } + + XD3_ASSERT (merge_stream == NULL); + + if ((merge_stream = (xd3_stream*) main_malloc (sizeof(xd3_stream))) == NULL) + { + ret = ENOMEM; + goto error; + } + + if ((ret = xd3_config_stream (merge_stream, NULL)) || + (ret = xd3_whole_state_init (merge_stream))) + { + XPR(NT XD3_LIB_ERRMSG (& merge_input, ret)); + goto error; + } + + xd3_swap_whole_state (& merge_stream->whole_target, + & merge_input.whole_target); + ret = 0; + error: + xd3_free_stream (& merge_input); + return ret; +} + +/* This processes each window of the final merge input. This routine + * does not output, it buffers the entire delta into memory. */ +static int +main_merge_func (xd3_stream* stream, main_file *no_write) +{ + int ret; + + if ((ret = xd3_whole_append_window (stream))) + { + return ret; + } + + return 0; +} + + +/* This is called after all windows have been read, as a final step in + * main_input(). This is only called for the final merge step. */ +static int +main_merge_output (xd3_stream *stream, main_file *ofile) +{ + int ret; + usize_t inst_pos = 0; + xoff_t output_pos = 0; + xd3_source recode_source; + usize_t window_num = 0; + int at_least_once = 0; + + /* merge_stream is set if there were arguments. this stream's input + * needs to be applied to the merge_stream source. */ + if ((merge_stream != NULL) && + (ret = xd3_merge_input_output (stream, + & merge_stream->whole_target))) + { + XPR(NT XD3_LIB_ERRMSG (stream, ret)); + return ret; + } + + if (option_use_appheader != 0 && + option_appheader != NULL) + { + xd3_set_appheader (recode_stream, option_appheader, + (usize_t) strlen ((char*) option_appheader)); + } + + /* Enter the ENC_INPUT state and bypass the next_in == NULL test + * and (leftover) input buffering logic. */ + XD3_ASSERT(recode_stream->enc_state == ENC_INIT); + recode_stream->enc_state = ENC_INPUT; + recode_stream->next_in = main_bdata; + recode_stream->flags |= XD3_FLUSH; + + /* This encodes the entire target. */ + while (inst_pos < stream->whole_target.instlen || !at_least_once) + { + xoff_t window_start = output_pos; + int window_srcset = 0; + xoff_t window_srcmin = 0; + xoff_t window_srcmax = 0; + usize_t window_pos = 0; + usize_t window_size; + + /* at_least_once ensures that we encode at least one window, + * which handles the 0-byte case. */ + at_least_once = 1; + + XD3_ASSERT (recode_stream->enc_state == ENC_INPUT); + + if ((ret = xd3_encode_input (recode_stream)) != XD3_WINSTART) + { + XPR(NT "invalid merge state: %s\n", xd3_mainerror (ret)); + return XD3_INVALID; + } + + /* Window sizes must match from the input to the output, so that + * target copies are in-range (and so that checksums carry + * over). */ + XD3_ASSERT (window_num < stream->whole_target.wininfolen); + window_size = stream->whole_target.wininfo[window_num].length; + + /* Output position should also match. */ + if (output_pos != stream->whole_target.wininfo[window_num].offset) + { + XPR(NT "internal merge error: offset mismatch\n"); + return XD3_INVALID; + } + + if (option_use_checksum && + (stream->dec_win_ind & VCD_ADLER32) != 0) + { + recode_stream->flags |= XD3_ADLER32_RECODE; + recode_stream->recode_adler32 = + stream->whole_target.wininfo[window_num].adler32; + } + + window_num++; + + if (main_bsize < window_size) + { + main_buffree (main_bdata); + main_bdata = NULL; + main_bsize = 0; + if ((main_bdata = (uint8_t*) + main_bufalloc (window_size)) == NULL) + { + return ENOMEM; + } + main_bsize = window_size; + } + + /* This encodes a single target window. */ + while (window_pos < window_size && + inst_pos < stream->whole_target.instlen) + { + xd3_winst *inst = &stream->whole_target.inst[inst_pos]; + usize_t take = min(inst->size, window_size - window_pos); + xoff_t addr; + + switch (inst->type) + { + case XD3_RUN: + if ((ret = xd3_emit_run (recode_stream, window_pos, take, + &stream->whole_target.adds[inst->addr]))) + { + return ret; + } + break; + + case XD3_ADD: + /* Adds are implicit, put them into the input buffer. */ + memcpy (main_bdata + window_pos, + stream->whole_target.adds + inst->addr, take); + break; + + default: /* XD3_COPY + copy mode */ + if (inst->mode != 0) + { + if (window_srcset) { + window_srcmin = min(window_srcmin, inst->addr); + window_srcmax = max(window_srcmax, inst->addr + take); + } else { + window_srcset = 1; + window_srcmin = inst->addr; + window_srcmax = inst->addr + take; + } + addr = inst->addr; + } + else + { + XD3_ASSERT (inst->addr >= window_start); + addr = inst->addr - window_start; + } + IF_DEBUG2 (XPR(NTR "[merge copy] winpos %u take %u addr %"Q"u mode %u\n", + window_pos, take, addr, inst->mode)); + if ((ret = xd3_found_match (recode_stream, window_pos, take, + addr, inst->mode != 0))) + { + return ret; + } + break; + } + + window_pos += take; + output_pos += take; + + if (take == inst->size) + { + inst_pos += 1; + } + else + { + /* Modify the instruction for the next pass. */ + if (inst->type != XD3_RUN) + { + inst->addr += take; + } + inst->size -= take; + } + } + + xd3_avail_input (recode_stream, main_bdata, window_pos); + + recode_stream->enc_state = ENC_INSTR; + + if (window_srcset) { + recode_stream->srcwin_decided = 1; + recode_stream->src = &recode_source; + recode_source.srclen = (usize_t)(window_srcmax - window_srcmin); + recode_source.srcbase = window_srcmin; + recode_stream->taroff = recode_source.srclen; + + XD3_ASSERT (recode_source.srclen != 0); + } else { + recode_stream->srcwin_decided = 0; + recode_stream->src = NULL; + recode_stream->taroff = 0; + } + + for (;;) + { + switch ((ret = xd3_encode_input (recode_stream))) + { + case XD3_INPUT: { + goto done_window; + } + case XD3_OUTPUT: { + /* main_file_write below */ + break; + } + case XD3_GOTHEADER: + case XD3_WINSTART: + case XD3_WINFINISH: { + /* ignore */ + continue; + } + case XD3_GETSRCBLK: + case 0: { + return XD3_INTERNAL; + } + default: + return ret; + } + + if ((ret = main_write_output(recode_stream, ofile))) + { + return ret; + } + + xd3_consume_output (recode_stream); + } + done_window: + (void) 0; + } + + return 0; +} +#endif + +/******************************************************************* + Input decompression, output recompression + ******************************************************************/ + +#if EXTERNAL_COMPRESSION +/* This is tricky POSIX-specific code with lots of fork(), pipe(), + * dup(), waitpid(), and exec() business. Most of this code + * originated in PRCS1, which did automatic package-file + * decompression. It works with both XD3_POSIX and XD3_STDIO file + * disciplines. + * + * To automatically detect compressed inputs requires a child process + * to reconstruct the input stream, which was advanced in order to + * detect compression, because it may not be seekable. In other + * words, the main program reads part of the input stream, and if it + * detects a compressed input it then forks a pipe copier process, + * which copies the first-read block out of the main-program's memory, + * then streams the remaining compressed input into the + * input-decompression pipe. + */ + +#include +#include +#include +#include + +/* Remember which pipe FD is which. */ +#define PIPE_READ_FD 0 +#define PIPE_WRITE_FD 1 +#define MAX_SUBPROCS 4 /* max(source + copier + output, + source + copier + input + copier). */ +static pid_t ext_subprocs[MAX_SUBPROCS]; + +/* Like write(), applies to a fd instead of a main_file, for the pipe + * copier subprocess. Does not print an error, to facilitate ignoring + * trailing garbage, see main_pipe_copier(). */ +static int +main_pipe_write (int outfd, uint8_t *exist_buf, usize_t remain) +{ + int ret; + + if ((ret = xd3_posix_io (outfd, exist_buf, remain, + (xd3_posix_func*) &write, NULL))) + { + return ret; + } + + return 0; +} + +/* A simple error-reporting waitpid interface. */ +static int +main_waitpid_check(pid_t pid) +{ + int status; + int ret = 0; + + if (waitpid (pid, & status, 0) < 0) + { + ret = get_errno (); + XPR(NT "external compression [pid %d] wait: %s\n", + pid, xd3_mainerror (ret)); + } + else if (! WIFEXITED (status)) + { + // SIGPIPE will be delivered to the child process whenever it + // writes data after this process closes the pipe, + // happens if xdelta does not require access to the entire + // source file. Considered normal. + if (! WIFSIGNALED (status) || WTERMSIG (status) != SIGPIPE) + { + ret = ECHILD; + XPR(NT "external compression [pid %d] signal %d\n", pid, + WIFSIGNALED (status) ? WTERMSIG (status) : WSTOPSIG (status)); + } + else if (option_verbose) + { + XPR(NT "external compression sigpipe\n"); + } + } + else if (WEXITSTATUS (status) != 0) + { + ret = ECHILD; + if (option_verbose > 1) + { + /* Presumably, the error was printed by the subprocess. */ + XPR(NT "external compression [pid %d] exit %d\n", + pid, WEXITSTATUS (status)); + } + } + + return ret; +} + +/* Wait for any existing child processes to check for abnormal exit. */ +static int +main_external_compression_finish (void) +{ + int i; + int ret; + + for (i = 0; i < num_subprocs; i += 1) + { + if (! ext_subprocs[i]) { continue; } + + if ((ret = main_waitpid_check (ext_subprocs[i]))) + { + return ret; + } + + ext_subprocs[i] = 0; + } + + return 0; +} + +/* Kills any outstanding compression process. */ +static void +main_external_compression_cleanup (void) +{ + int i; + + for (i = 0; i < num_subprocs; i += 1) + { + if (! ext_subprocs[i]) { continue; } + + kill (ext_subprocs[i], SIGTERM); + + ext_subprocs[i] = 0; + } +} + +/* This runs as a forked process of main_input_decompress_setup() to + * copy input to the decompression process. First, the available + * input is copied out of the existing buffer, then the buffer is + * reused to continue reading from the compressed input file. */ +static int +main_pipe_copier (uint8_t *pipe_buf, + usize_t pipe_bufsize, + size_t nread, + main_file *ifile, + int outfd) +{ + int ret; + xoff_t skipped = 0; + + /* Prevent SIGPIPE signals, allow EPIPE return values instead. This + * is safe to comment-out, except that the -F flag will not work + * properly (the parent would need to treat WTERMSIG(status) == + * SIGPIPE). */ + struct sigaction sa; + sa.sa_handler = SIG_IGN; + sigaction (SIGPIPE, &sa, NULL); + + for (;;) + { + /* force_drain will be set when option_force and EPIPE cause us + * to skip data. This is reset each time through the loop, so + * the break condition below works. */ + int force_drain = 0; + if (nread > 0 && (ret = main_pipe_write (outfd, pipe_buf, nread))) + { + if (ret == EPIPE) + { + /* This causes the loop to continue reading until nread + * == 0. */ + skipped += nread; + force_drain = 1; + } + else + { + XPR(NT "pipe write failed: %s\n", xd3_mainerror (ret)); + return ret; + } + } + + if (nread < pipe_bufsize && !force_drain) + { + break; + } + + if ((ret = main_file_read (ifile, pipe_buf, pipe_bufsize, + & nread, "pipe read failed")) < 0) + { + return ret; + } + } + + if (option_verbose && skipped != 0) + { + XPR(NT "skipping %"Q"u bytes in %s\n", + skipped, ifile->filename); + } + return 0; +} + +/* This function is called after we have read some amount of data from + * the input file and detected a compressed input. Here we start a + * decompression subprocess by forking twice. The first process runs + * the decompression command, the second process copies data to the + * input of the first. */ +static int +main_input_decompress_setup (const main_extcomp *decomp, + main_file *ifile, + uint8_t *input_buf, + usize_t input_bufsize, + uint8_t *pipe_buf, + usize_t pipe_bufsize, + usize_t pipe_avail, + size_t *nread) +{ + /* The two pipes: input and output file descriptors. */ + int outpipefd[2], inpipefd[2]; + int input_fd = -1; /* The resulting input_fd (output of decompression). */ + pid_t decomp_id, copier_id; /* The two subprocs. */ + int ret; + + outpipefd[0] = outpipefd[1] = -1; + inpipefd[0] = inpipefd[1] = -1; + + if (pipe (outpipefd) || pipe (inpipefd)) + { + XPR(NT "pipe failed: %s\n", xd3_mainerror (ret = get_errno ())); + goto pipe_cleanup; + } + + if ((decomp_id = fork ()) < 0) + { + XPR(NT "fork failed: %s\n", xd3_mainerror (ret = get_errno ())); + goto pipe_cleanup; + } + + /* The first child runs the decompression process: */ + if (decomp_id == 0) + { + if (option_verbose > 2) + { + XPR(NT "external decompression pid %d\n", getpid ()); + } + + /* Setup pipes: write to the outpipe, read from the inpipe. */ + if (dup2 (outpipefd[PIPE_WRITE_FD], STDOUT_FILENO) < 0 || + dup2 (inpipefd[PIPE_READ_FD], STDIN_FILENO) < 0 || + close (outpipefd[PIPE_READ_FD]) || + close (outpipefd[PIPE_WRITE_FD]) || + close (inpipefd[PIPE_READ_FD]) || + close (inpipefd[PIPE_WRITE_FD]) || + execlp (decomp->decomp_cmdname, decomp->decomp_cmdname, + decomp->decomp_options, + option_force2 ? "-f" : NULL, + NULL)) + { + XPR(NT "child process %s failed to execute: %s\n", + decomp->decomp_cmdname, xd3_mainerror (get_errno ())); + } + + _exit (127); + } + + XD3_ASSERT(num_subprocs < MAX_SUBPROCS); + ext_subprocs[num_subprocs++] = decomp_id; + + if ((copier_id = fork ()) < 0) + { + XPR(NT "fork failed: %s\n", xd3_mainerror (ret = get_errno ())); + goto pipe_cleanup; + } + + /* The second child runs the copier process: */ + if (copier_id == 0) + { + int exitval = 0; + + if (option_verbose > 2) + { + XPR(NT "child pipe-copier pid %d\n", getpid ()); + } + + if (close (inpipefd[PIPE_READ_FD]) || + close (outpipefd[PIPE_READ_FD]) || + close (outpipefd[PIPE_WRITE_FD]) || + main_pipe_copier (pipe_buf, pipe_bufsize, pipe_avail, + ifile, inpipefd[PIPE_WRITE_FD]) || + close (inpipefd[PIPE_WRITE_FD])) + { + XPR(NT "child copier process failed: %s\n", + xd3_mainerror (get_errno ())); + exitval = 1; + } + + _exit (exitval); + } + + XD3_ASSERT(num_subprocs < MAX_SUBPROCS); + ext_subprocs[num_subprocs++] = copier_id; + + /* The parent closes both pipes after duplicating the output of + * compression. */ + input_fd = dup (outpipefd[PIPE_READ_FD]); + + if (input_fd < 0 || + main_file_close (ifile) || + close (outpipefd[PIPE_READ_FD]) || + close (outpipefd[PIPE_WRITE_FD]) || + close (inpipefd[PIPE_READ_FD]) || + close (inpipefd[PIPE_WRITE_FD])) + { + XPR(NT "dup/close failed: %s\n", xd3_mainerror (ret = get_errno ())); + goto pipe_cleanup; + } + +#if XD3_STDIO + /* Note: fdopen() acquires the fd, closes it when finished. */ + if ((ifile->file = fdopen (input_fd, "r")) == NULL) + { + XPR(NT "fdopen failed: %s\n", xd3_mainerror (ret = get_errno ())); + goto pipe_cleanup; + } + +#elif XD3_POSIX + ifile->file = input_fd; +#endif + + ifile->compressor = decomp; + + /* Now the input file is decompressed. */ + return main_file_read (ifile, input_buf, input_bufsize, + nread, "input decompression failed"); + + pipe_cleanup: + close (input_fd); + close (outpipefd[PIPE_READ_FD]); + close (outpipefd[PIPE_WRITE_FD]); + close (inpipefd[PIPE_READ_FD]); + close (inpipefd[PIPE_WRITE_FD]); + return ret; +} + + +/* This routine is called when the first buffer of input data is read + * by the main program (unless input decompression is disabled by + * command-line option). If it recognizes the magic number of a known + * input type it invokes decompression. + * + * Skips decompression if the decompression type or the file type is + * RD_NONEXTERNAL. + * + * Behaves exactly like main_file_read, otherwise. + * + * This function uses a separate buffer to read the first small block + * of input. If a compressed input is detected, the separate buffer + * is passed to the pipe copier. This avoids using the same size + * buffer in both cases. */ +static int +main_secondary_decompress_check (main_file *file, + uint8_t *input_buf, + size_t input_size, + size_t *nread) +{ + int ret; + usize_t i; + usize_t try_read = min (input_size, XD3_ALLOCSIZE); + size_t check_nread = 0; + uint8_t check_buf[XD3_ALLOCSIZE]; /* TODO: stack limit */ + const main_extcomp *decompressor = NULL; + + if ((ret = main_file_read (file, check_buf, + try_read, + & check_nread, "input read failed"))) + { + return ret; + } + + if (file->flags & RD_DECOMPSET) + { + /* This allows the application header to override the magic + * number, for whatever reason. */ + decompressor = file->compressor; + } + else + { + for (i = 0; i < SIZEOF_ARRAY (extcomp_types); i += 1) + { + const main_extcomp *decomp = & extcomp_types[i]; + + if (check_nread > decomp->magic_size) + { + /* The following expr checks if we are trying to read a + * VCDIFF input, in which case do not treat it as + * "secondary" decompression. */ + int skip_this_type = (decomp->flags & RD_NONEXTERNAL) && + (file->flags & RD_NONEXTERNAL); + + if (skip_this_type) + { + continue; + } + + if (memcmp (check_buf, decomp->magic, decomp->magic_size) == 0) + { + decompressor = decomp; + break; + } + } + } + } + + if (decompressor != NULL) + { + if (! option_quiet) + { + XPR(NT "externally compressed input: %s %s%s < %s\n", + decompressor->decomp_cmdname, + decompressor->decomp_options, + (option_force2 ? " -f" : ""), + file->filename); + if (file->flags & RD_MAININPUT) + { + XPR(NT + "WARNING: the encoder is automatically decompressing the input file;\n"); + XPR(NT + "WARNING: the decoder will automatically recompress the output file;\n"); + XPR(NT + "WARNING: this may result in different compressed data and checksums\n"); + XPR(NT + "WARNING: despite being identical data; if this is an issue, use -D\n"); + XPR(NT + "WARNING: to avoid decompression and/or use -R to avoid recompression\n"); + XPR(NT + "WARNING: and/or manually decompress the input file; if you know the\n"); + XPR(NT + "WARNING: compression settings that will produce identical output\n"); + XPR(NT + "WARNING: you may set those flags using the environment (e.g., GZIP=-9)\n"); + } + } + + file->size_known = 0; + return main_input_decompress_setup (decompressor, file, + input_buf, input_size, + check_buf, XD3_ALLOCSIZE, + check_nread, nread); + } + + /* Now read the rest of the input block. */ + (*nread) = 0; + + if (check_nread == try_read) + { + ret = main_file_read (file, + input_buf + try_read, + input_size - try_read, + nread, + "input read failed"); + } + + memcpy (input_buf, check_buf, check_nread); + + (*nread) += check_nread; + + return 0; +} + +/* Initiate re-compression of the output stream. This is easier than + * input decompression because we know beforehand that the stream will + * be compressed, whereas the input has already been read when we + * decide it should be decompressed. Thus, it only requires one + * subprocess and one pipe. */ +static int +main_recompress_output (main_file *ofile) +{ + pid_t recomp_id; /* One subproc. */ + int pipefd[2]; /* One pipe. */ + int output_fd = -1; + int ret; + const main_extcomp *recomp = ofile->compressor; + + pipefd[0] = pipefd[1] = -1; + + if (pipe (pipefd)) + { + XPR(NT "pipe failed: %s\n", xd3_mainerror (ret = get_errno ())); + goto pipe_cleanup; + } + + if ((recomp_id = fork ()) < 0) + { + XPR(NT "fork failed: %s\n", xd3_mainerror (ret = get_errno ())); + goto pipe_cleanup; + } + + /* The child runs the recompression process: */ + if (recomp_id == 0) + { + if (option_verbose > 2) + { + XPR(NT "external recompression pid %d\n", getpid ()); + } + + /* Setup pipes: write to the output file, read from the pipe. */ + if (dup2 (XFNO (ofile), STDOUT_FILENO) < 0 || + dup2 (pipefd[PIPE_READ_FD], STDIN_FILENO) < 0 || + close (pipefd[PIPE_READ_FD]) || + close (pipefd[PIPE_WRITE_FD]) || + execlp (recomp->recomp_cmdname, recomp->recomp_cmdname, + recomp->recomp_options, + option_force2 ? "-f" : NULL, + NULL)) + { + XPR(NT "child process %s failed to execute: %s\n", + recomp->recomp_cmdname, xd3_mainerror (get_errno ())); + } + + _exit (127); + } + + XD3_ASSERT(num_subprocs < MAX_SUBPROCS); + ext_subprocs[num_subprocs++] = recomp_id; + + /* The parent closes both pipes after duplicating the output-fd for + * writing to the compression pipe. */ + output_fd = dup (pipefd[PIPE_WRITE_FD]); + + if (output_fd < 0 || + main_file_close (ofile) || + close (pipefd[PIPE_READ_FD]) || + close (pipefd[PIPE_WRITE_FD])) + { + XPR(NT "close failed: %s\n", xd3_mainerror (ret = get_errno ())); + goto pipe_cleanup; + } + +#if XD3_STDIO + /* Note: fdopen() acquires the fd, closes it when finished. */ + if ((ofile->file = fdopen (output_fd, "w")) == NULL) + { + XPR(NT "fdopen failed: %s\n", xd3_mainerror (ret = get_errno ())); + goto pipe_cleanup; + } + +#elif XD3_POSIX + ofile->file = output_fd; +#endif + + /* Now the output file will be compressed. */ + return 0; + + pipe_cleanup: + close (output_fd); + close (pipefd[PIPE_READ_FD]); + close (pipefd[PIPE_WRITE_FD]); + return ret; +} +#endif /* EXTERNAL_COMPRESSION */ + +/* Identify the compressor that was used based on its ident string, + * which is passed in the application header. */ +static const main_extcomp* +main_ident_compressor (const char *ident) +{ + usize_t i; + + for (i = 0; i < SIZEOF_ARRAY (extcomp_types); i += 1) + { + if (strcmp (extcomp_types[i].ident, ident) == 0) + { + return & extcomp_types[i]; + } + } + + return NULL; +} + +/* Return the main_extcomp record to use for this identifier, if possible. */ +static const main_extcomp* +main_get_compressor (const char *ident) +{ + const main_extcomp *ext = main_ident_compressor (ident); + + if (ext == NULL) + { + if (! option_quiet) + { + XPR(NT "warning: cannot recompress output: " + "unrecognized external compression ID: %s\n", ident); + } + return NULL; + } + else if (! EXTERNAL_COMPRESSION) + { + if (! option_quiet) + { + XPR(NT "warning: external support not compiled: " + "original input was compressed: %s\n", ext->recomp_cmdname); + } + return NULL; + } + else + { + return ext; + } +} + +/********************************************************************* + APPLICATION HEADER + *******************************************************************/ + +#if XD3_ENCODER +static const char* +main_apphead_string (const char* x) +{ + const char *y; + + if (x == NULL) { return ""; } + + if (strcmp (x, "/dev/stdin") == 0 || + strcmp (x, "/dev/stdout") == 0 || + strcmp (x, "/dev/stderr") == 0) { return "-"; } + + // TODO: this is not portable + return (y = strrchr (x, '/')) == NULL ? x : y + 1; +} + +static int +main_set_appheader (xd3_stream *stream, main_file *input, main_file *sfile) +{ + /* The user may disable the application header. Once the appheader + * is set, this disables setting it again. */ + if (appheader_used || ! option_use_appheader) { return 0; } + + /* The user may specify the application header, otherwise format the + default header. */ + if (option_appheader) + { + appheader_used = option_appheader; + } + else + { + const char *iname; + const char *icomp; + const char *sname; + const char *scomp; + usize_t len; + + iname = main_apphead_string (input->filename); + icomp = (input->compressor == NULL) ? "" : input->compressor->ident; + len = (usize_t) strlen (iname) + (usize_t) strlen (icomp) + 2; + + if (sfile->filename != NULL) + { + sname = main_apphead_string (sfile->filename); + scomp = (sfile->compressor == NULL) ? "" : sfile->compressor->ident; + len += (usize_t) strlen (sname) + (usize_t) strlen (scomp) + 2; + } + else + { + sname = scomp = ""; + } + + if ((appheader_used = (uint8_t*) main_malloc (len)) == NULL) + { + return ENOMEM; + } + + if (sfile->filename == NULL) + { + snprintf_func ((char*)appheader_used, len, "%s/%s", iname, icomp); + } + else + { + snprintf_func ((char*)appheader_used, len, "%s/%s/%s/%s", + iname, icomp, sname, scomp); + } + } + + xd3_set_appheader (stream, appheader_used, + (usize_t) strlen ((char*)appheader_used)); + + return 0; +} +#endif + +static void +main_get_appheader_params (main_file *file, char **parsed, + int output, const char *type, + main_file *other) +{ + /* Set the filename if it was not specified. If output, option_stdout (-c) + * overrides. */ + if (file->filename == NULL && + ! (output && option_stdout) && + strcmp (parsed[0], "-") != 0) + { + file->filename = parsed[0]; + + if (other->filename != NULL) { + /* Take directory from the other file, if it has one. */ + /* TODO: This results in nonsense names like /dev/foo.tar.gz + * and probably the filename-default logic interferes with + * multi-file operation and the standard file extension? + * Possibly the name header is bad, should be off by default. + * Possibly we just want to remember external/compression + * settings. */ + const char *last_slash = strrchr(other->filename, '/'); + + if (last_slash != NULL) { + usize_t dlen = (usize_t) (last_slash - other->filename); + + XD3_ASSERT(file->filename_copy == NULL); + file->filename_copy = + (char*) main_malloc(dlen + 2 + (usize_t) strlen(file->filename)); + + strncpy(file->filename_copy, other->filename, dlen); + file->filename_copy[dlen] = '/'; + strcpy(file->filename_copy + dlen + 1, parsed[0]); + + file->filename = file->filename_copy; + } + } + + if (! option_quiet) + { + XPR(NT "using default %s filename: %s\n", type, file->filename); + } + } + + /* Set the compressor, initiate de/recompression later. */ + if (file->compressor == NULL && *parsed[1] != 0) + { + file->flags |= RD_DECOMPSET; + file->compressor = main_get_compressor (parsed[1]); + } +} + +static void +main_get_appheader (xd3_stream *stream, main_file *ifile, + main_file *output, main_file *sfile) +{ + uint8_t *apphead; + usize_t appheadsz; + int ret; + + /* The user may disable the application header. Once the appheader + * is set, this disables setting it again. */ + if (! option_use_appheader) { return; } + + ret = xd3_get_appheader (stream, & apphead, & appheadsz); + + /* Ignore failure, it only means we haven't received a header yet. */ + if (ret != 0) { return; } + + if (appheadsz > 0) + { + char *start = (char*)apphead; + char *slash; + int place = 0; + char *parsed[4]; + + memset (parsed, 0, sizeof (parsed)); + + while ((slash = strchr (start, '/')) != NULL) + { + *slash = 0; + parsed[place++] = start; + start = slash + 1; + } + + parsed[place++] = start; + + /* First take the output parameters. */ + if (place == 2 || place == 4) + { + main_get_appheader_params (output, parsed, 1, "output", ifile); + } + + /* Then take the source parameters. */ + if (place == 4) + { + main_get_appheader_params (sfile, parsed+2, 0, "source", ifile); + } + } + + option_use_appheader = 0; + return; +} + +/********************************************************************* + Main I/O routines + **********************************************************************/ + +/* This function acts like the above except it may also try to + * recognize a compressed input (source or target) when the first + * buffer of data is read. The EXTERNAL_COMPRESSION code is called to + * search for magic numbers. */ +static int +main_read_primary_input (main_file *file, + uint8_t *buf, + size_t size, + size_t *nread) +{ +#if EXTERNAL_COMPRESSION + if (option_decompress_inputs && file->flags & RD_FIRST) + { + file->flags &= ~RD_FIRST; + return main_secondary_decompress_check (file, buf, size, nread); + } +#endif + + return main_file_read (file, buf, size, nread, "input read failed"); +} + +/* Open the main output file, sets a default file name, initiate + * recompression. This function is expected to fprint any error + * messages. */ +static int +main_open_output (xd3_stream *stream, main_file *ofile) +{ + int ret; + + if (option_no_output) + { + return 0; + } + + if (ofile->filename == NULL) + { + XSTDOUT_XF (ofile); + + if (option_verbose > 1) + { + XPR(NT "using standard output: %s\n", ofile->filename); + } + } + else + { + /* Stat the file to check for overwrite. */ + if (option_force == 0 && main_file_exists (ofile)) + { + if (!option_quiet) + { + XPR(NT "to overwrite output file specify -f: %s\n", + ofile->filename); + } + return EEXIST; + } + + if ((ret = main_file_open (ofile, ofile->filename, XO_WRITE))) + { + return ret; + } + + if (option_verbose > 1) { XPR(NT "output %s\n", ofile->filename); } + } + +#if EXTERNAL_COMPRESSION + /* Do output recompression. */ + if (ofile->compressor != NULL && option_recompress_outputs == 1) + { + if (! option_quiet) + { + XPR(NT "externally compressed output: %s %s%s > %s\n", + ofile->compressor->recomp_cmdname, + ofile->compressor->recomp_options, + (option_force2 ? " -f" : ""), + ofile->filename); + } + + if ((ret = main_recompress_output (ofile))) + { + return ret; + } + } +#endif + + return 0; +} + +static usize_t +main_get_winsize (main_file *ifile) { + xoff_t file_size = 0; + usize_t size = option_winsize; + static shortbuf iszbuf; + + if (main_file_stat (ifile, &file_size) == 0) + { + size = (usize_t) min(file_size, (xoff_t) size); + } + + size = max(size, XD3_ALLOCSIZE); + + if (option_verbose > 1) + { + XPR(NT "input %s window size %s\n", + ifile->filename, + main_format_bcnt (size, &iszbuf)); + } + + return size; +} + +/********************************************************************* + Main routines + ********************************************************************/ + +/* This is a generic input function. It calls the xd3_encode_input or + * xd3_decode_input functions and makes calls to the various input + * handling routines above, which coordinate external decompression. + */ +static int +main_input (xd3_cmd cmd, + main_file *ifile, + main_file *ofile, + main_file *sfile) +{ + int ret; + xd3_stream stream; + size_t nread = 0; + usize_t winsize; + int stream_flags = 0; + xd3_config config; + xd3_source source; + xoff_t last_total_in = 0; + xoff_t last_total_out = 0; + long start_time; + int stdout_only = 0; + int (*input_func) (xd3_stream*); + int (*output_func) (xd3_stream*, main_file *); + + memset (& stream, 0, sizeof (stream)); + memset (& source, 0, sizeof (source)); + memset (& config, 0, sizeof (config)); + + config.alloc = main_alloc; + config.freef = main_free1; + + config.iopt_size = option_iopt_size; + config.sprevsz = option_sprevsz; + + do_src_fifo = 0; + + start_time = get_millisecs_now (); + + if (option_use_checksum) { stream_flags |= XD3_ADLER32; } + + /* main_input setup. */ + switch ((int) cmd) + { +#if VCDIFF_TOOLS + if (1) { case CMD_PRINTHDR: stream_flags |= XD3_JUST_HDR; } + else if (1) { case CMD_PRINTHDRS: stream_flags |= XD3_SKIP_WINDOW; } + else { case CMD_PRINTDELTA: stream_flags |= XD3_SKIP_EMIT; } + ifile->flags |= RD_NONEXTERNAL; + input_func = xd3_decode_input; + output_func = main_print_func; + stream_flags |= XD3_ADLER32_NOVER; + stdout_only = 1; + break; + + case CMD_RECODE: + case CMD_MERGE: + case CMD_MERGE_ARG: + /* No source will be read */ + stream_flags |= XD3_ADLER32_NOVER | XD3_SKIP_EMIT; + ifile->flags |= RD_NONEXTERNAL; + input_func = xd3_decode_input; + + if ((ret = main_init_recode_stream ())) + { + return EXIT_FAILURE; + } + + if (cmd == CMD_RECODE) { output_func = main_recode_func; } + else { output_func = main_merge_func; } + break; +#endif /* VCDIFF_TOOLS */ + +#if XD3_ENCODER + case CMD_ENCODE: + do_src_fifo = 1; + input_func = xd3_encode_input; + output_func = main_write_output; + + if (option_no_compress) { stream_flags |= XD3_NOCOMPRESS; } + if (option_use_altcodetable) { stream_flags |= XD3_ALT_CODE_TABLE; } + if (option_smatch_config) + { + const char *s = option_smatch_config; + char *e; + int values[XD3_SOFTCFG_VARCNT]; + int got; + + config.smatch_cfg = XD3_SMATCH_SOFT; + + for (got = 0; got < XD3_SOFTCFG_VARCNT; got += 1, s = e + 1) + { + values[got] = strtol (s, &e, 10); + + if ((values[got] < 0) || + (e == s) || + (got < XD3_SOFTCFG_VARCNT-1 && *e == 0) || + (got == XD3_SOFTCFG_VARCNT-1 && *e != 0)) + { + XPR(NT "invalid string match specifier (-C) %d: %s\n", + got, s); + return EXIT_FAILURE; + } + } + + config.smatcher_soft.large_look = values[0]; + config.smatcher_soft.large_step = values[1]; + config.smatcher_soft.small_look = values[2]; + config.smatcher_soft.small_chain = values[3]; + config.smatcher_soft.small_lchain = values[4]; + config.smatcher_soft.max_lazy = values[5]; + config.smatcher_soft.long_enough = values[6]; + } + else + { + if (option_verbose > 2) + { + XPR(NT "compression level: %d\n", option_level); + } + if (option_level == 0) + { + stream_flags |= XD3_NOCOMPRESS; + config.smatch_cfg = XD3_SMATCH_FASTEST; + } + else if (option_level == 1) + { config.smatch_cfg = XD3_SMATCH_FASTEST; } + else if (option_level == 2) + { config.smatch_cfg = XD3_SMATCH_FASTER; } + else if (option_level <= 5) + { config.smatch_cfg = XD3_SMATCH_FAST; } + else if (option_level == 6) + { config.smatch_cfg = XD3_SMATCH_DEFAULT; } + else + { config.smatch_cfg = XD3_SMATCH_SLOW; } + } + break; +#endif + case CMD_DECODE: + if (option_use_checksum == 0) { stream_flags |= XD3_ADLER32_NOVER; } + ifile->flags |= RD_NONEXTERNAL; + input_func = xd3_decode_input; + output_func = main_write_output; + break; + default: + XPR(NT "internal error\n"); + return EXIT_FAILURE; + } + + main_bsize = winsize = main_get_winsize (ifile); + + if ((main_bdata = (uint8_t*) main_bufalloc (winsize)) == NULL) + { + return EXIT_FAILURE; + } + + config.winsize = winsize; + config.getblk = main_getblk_func; + config.flags = stream_flags; + + if ((ret = main_set_secondary_flags (&config)) || + (ret = xd3_config_stream (& stream, & config))) + { + XPR(NT XD3_LIB_ERRMSG (& stream, ret)); + return EXIT_FAILURE; + } + +#if VCDIFF_TOOLS + if ((cmd == CMD_MERGE || cmd == CMD_MERGE_ARG) && + (ret = xd3_whole_state_init (& stream))) + { + XPR(NT XD3_LIB_ERRMSG (& stream, ret)); + return EXIT_FAILURE; + } +#endif + + if (cmd != CMD_DECODE) + { + /* When not decoding, set source now. The decoder delays this + * step until XD3_GOTHEADER. */ + if (sfile && sfile->filename != NULL) + { + if ((ret = main_set_source (& stream, cmd, sfile, & source))) + { + return EXIT_FAILURE; + } + + XD3_ASSERT(stream.src != NULL); + } + } + + if (cmd == CMD_PRINTHDR || + cmd == CMD_PRINTHDRS || + cmd == CMD_PRINTDELTA || + cmd == CMD_RECODE) + { + if (sfile->filename == NULL) + { + allow_fake_source = 1; + sfile->filename = ""; + main_set_source (& stream, cmd, sfile, & source); + } + } + + /* This times each window. */ + get_millisecs_since (); + + /* Main input loop. */ + do + { + xoff_t input_offset; + xoff_t input_remain; + usize_t try_read; + + input_offset = ifile->nread; + + input_remain = XOFF_T_MAX - input_offset; + + try_read = (usize_t) min ((xoff_t) config.winsize, input_remain); + + if ((ret = main_read_primary_input (ifile, main_bdata, + try_read, & nread))) + { + return EXIT_FAILURE; + } + + /* If we've reached EOF tell the stream to flush. */ + if (nread < try_read) + { + stream.flags |= XD3_FLUSH; + } + +#if XD3_ENCODER + /* After the first main_read_primary_input completes, we know + * all the information needed to encode the application + * header. */ + if (cmd == CMD_ENCODE && + (ret = main_set_appheader (& stream, ifile, sfile))) + { + return EXIT_FAILURE; + } +#endif + xd3_avail_input (& stream, main_bdata, nread); + + /* If we read zero bytes after encoding at least one window... */ + if (nread == 0 && stream.current_window > 0) { + break; + } + + again: + ret = input_func (& stream); + + switch (ret) + { + case XD3_INPUT: + continue; + + case XD3_GOTHEADER: + { + XD3_ASSERT (stream.current_window == 0); + + /* Need to process the appheader as soon as possible. It may + * contain a suggested default filename/decompression routine for + * the ofile, and it may contain default/decompression routine for + * the sources. */ + if (cmd == CMD_DECODE) + { + /* May need to set the sfile->filename if none was given. */ + main_get_appheader (& stream, ifile, ofile, sfile); + + /* Now open the source file. */ + if ((sfile->filename != NULL) && + (ret = main_set_source (& stream, cmd, sfile, & source))) + { + return EXIT_FAILURE; + } + } + } + /* FALLTHROUGH */ + case XD3_WINSTART: + { + /* e.g., set or unset XD3_SKIP_WINDOW. */ + goto again; + } + + case XD3_OUTPUT: + { + /* Defer opening the output file until the stream produces its + * first output for both encoder and decoder, this way we + * delay long enough for the decoder to receive the + * application header. (Or longer if there are skipped + * windows, but I can't think of any reason not to delay + * open.) */ + if (ofile != NULL && + ! main_file_isopen (ofile) && + (ret = main_open_output (& stream, ofile)) != 0) + { + return EXIT_FAILURE; + } + + if ((ret = output_func (& stream, ofile)) && + (ret != PRINTHDR_SPECIAL)) + { + return EXIT_FAILURE; + } + + if (ret == PRINTHDR_SPECIAL) + { + xd3_abort_stream (& stream); + ret = EXIT_SUCCESS; + goto done; + } + + ret = 0; + + xd3_consume_output (& stream); + goto again; + } + + case XD3_WINFINISH: + { + if (IS_ENCODE (cmd) || cmd == CMD_DECODE || cmd == CMD_RECODE) + { + if (! option_quiet && IS_ENCODE (cmd) && + main_file_isopen (sfile)) + { + /* Warn when no source copies are found */ + if (option_verbose && ! xd3_encoder_used_source (& stream)) + { + XPR(NT "warning: input window %"Q"u..%"Q"u has " + "no source copies\n", + stream.current_window * winsize, + (stream.current_window+1) * winsize); + XD3_ASSERT (stream.src != NULL); + } + + /* Limited i-buffer size affects source copies + * when the sourcewin is decided early. */ + if (option_verbose > 1 && + stream.srcwin_decided_early && + stream.i_slots_used > stream.iopt_size) + { + XPR(NT "warning: input position %"Q"u overflowed " + "instruction buffer, needed %u (vs. %u), " + "consider changing -I\n", + stream.current_window * winsize, + stream.i_slots_used, stream.iopt_size); + } + } + + if (option_verbose) + { + shortbuf rrateavg, wrateavg, tm; + shortbuf rdb, wdb; + shortbuf trdb, twdb; + shortbuf srcpos; + long millis = get_millisecs_since (); + usize_t this_read = (usize_t)(stream.total_in - + last_total_in); + usize_t this_write = (usize_t)(stream.total_out - + last_total_out); + last_total_in = stream.total_in; + last_total_out = stream.total_out; + + if (option_verbose > 1) + { + XPR(NT "%"Q"u: in %s (%s): out %s (%s): " + "total in %s: out %s: %s: srcpos %s\n", + stream.current_window, + main_format_bcnt (this_read, &rdb), + main_format_rate (this_read, millis, &rrateavg), + main_format_bcnt (this_write, &wdb), + main_format_rate (this_write, millis, &wrateavg), + main_format_bcnt (stream.total_in, &trdb), + main_format_bcnt (stream.total_out, &twdb), + main_format_millis (millis, &tm), + main_format_bcnt (sfile->source_position, &srcpos)); + } + else + { + XPR(NT "%"Q"u: in %s: out %s: total in %s: " + "out %s: %s\n", + stream.current_window, + main_format_bcnt (this_read, &rdb), + main_format_bcnt (this_write, &wdb), + main_format_bcnt (stream.total_in, &trdb), + main_format_bcnt (stream.total_out, &twdb), + main_format_millis (millis, &tm)); + } + } + } + goto again; + } + + default: + /* input_func() error */ + XPR(NT XD3_LIB_ERRMSG (& stream, ret)); + if (! option_quiet && ret == XD3_INVALID_INPUT) + { + XPR(NT "normally this indicates that the source file is incorrect\n"); + XPR(NT "please verify the source file with sha1sum or equivalent\n"); + } + return EXIT_FAILURE; + } + } + while (nread == config.winsize); +done: + /* Close the inputs. (ifile must be open, sfile may be open) */ + main_file_close (ifile); + if (sfile != NULL) + { + main_file_close (sfile); + } + +#if VCDIFF_TOOLS + if (cmd == CMD_MERGE && + (ret = main_merge_output (& stream, ofile))) + { + return EXIT_FAILURE; + } + + if (cmd == CMD_MERGE_ARG) + { + xd3_swap_whole_state (& stream.whole_target, + & recode_stream->whole_target); + } +#endif /* VCDIFF_TOOLS */ + + /* If output file is not open yet because of delayed-open, it means + * we never encountered a window in the delta, but it could have had + * a VCDIFF header? TODO: solve this elsewhere. For now, it prints + * "nothing to output" below, but the check doesn't happen in case + * of option_no_output. */ + if (! option_no_output && ofile != NULL) + { + if (!stdout_only && ! main_file_isopen (ofile)) + { + XPR(NT "nothing to output: %s\n", ifile->filename); + return EXIT_FAILURE; + } + + /* Have to close the output before calling + * main_external_compression_finish, or else it hangs. */ + if (main_file_close (ofile) != 0) + { + return EXIT_FAILURE; + } + } + +#if EXTERNAL_COMPRESSION + if ((ret = main_external_compression_finish ())) + { + XPR(NT "external compression commands failed\n"); + return EXIT_FAILURE; + } +#endif + + if ((ret = xd3_close_stream (& stream))) + { + XPR(NT XD3_LIB_ERRMSG (& stream, ret)); + return EXIT_FAILURE; + } + +#if XD3_ENCODER + if (option_verbose > 1 && cmd == CMD_ENCODE) + { + XPR(NT "scanner configuration: %s\n", stream.smatcher.name); + XPR(NT "target hash table size: %u\n", stream.small_hash.size); + if (sfile != NULL && sfile->filename != NULL) + { + XPR(NT "source hash table size: %u\n", stream.large_hash.size); + } + } + + if (option_verbose > 2 && cmd == CMD_ENCODE) + { + XPR(NT "source copies: %"Q"u (%"Q"u bytes)\n", + stream.n_scpy, stream.l_scpy); + XPR(NT "target copies: %"Q"u (%"Q"u bytes)\n", + stream.n_tcpy, stream.l_tcpy); + XPR(NT "adds: %"Q"u (%"Q"u bytes)\n", stream.n_add, stream.l_add); + XPR(NT "runs: %"Q"u (%"Q"u bytes)\n", stream.n_run, stream.l_run); + } +#endif + + xd3_free_stream (& stream); + + if (option_verbose) + { + shortbuf tm; + long end_time = get_millisecs_now (); + xoff_t nwrite = ofile != NULL ? ofile->nwrite : 0; + + XPR(NT "finished in %s; input %"Q"u output %"Q"u bytes (%0.2f%%)\n", + main_format_millis (end_time - start_time, &tm), + ifile->nread, nwrite, 100.0 * nwrite / ifile->nread); + } + + return EXIT_SUCCESS; +} + +/* free memory before exit, reset single-use variables. */ +static void +main_cleanup (void) +{ + if (appheader_used != NULL && + appheader_used != option_appheader) + { + main_free (appheader_used); + appheader_used = NULL; + } + + main_buffree (main_bdata); + main_bdata = NULL; + main_bsize = 0; + + main_lru_cleanup(); + + if (recode_stream != NULL) + { + xd3_free_stream (recode_stream); + main_free (recode_stream); + recode_stream = NULL; + } + + if (merge_stream != NULL) + { + xd3_free_stream (merge_stream); + main_free (merge_stream); + merge_stream = NULL; + } + + XD3_ASSERT (main_mallocs == 0); +} + +static void +setup_environment (int argc, + char **argv, + int *argc_out, + char ***argv_out, + char ***argv_free, + char **env_free) +{ + int n, i, i0; + char *p, *v = getenv("XDELTA"); + if (v == NULL) { + (*argc_out) = argc; + (*argv_out) = argv; + (*argv_free) = NULL; + (*env_free) = NULL; + return; + } + + (*env_free) = (char*) main_malloc((usize_t) strlen(v) + 1); + strcpy(*env_free, v); + + /* Space needed for extra args, at least # of spaces */ + n = argc + 1; + for (p = *env_free; *p != 0; ) { + if (*p++ == ' ') { + n++; + } + } + + (*argv_free) = (char**) main_malloc(sizeof(char*) * (n + 1)); + (*argv_out) = (*argv_free); + (*argv_out)[0] = argv[0]; + (*argv_out)[n] = NULL; + + i = 1; + for (p = *env_free; *p != 0; ) { + (*argv_out)[i++] = p; + while (*p != ' ' && *p != 0) { + p++; + } + while (*p == ' ') { + *p++ = 0; + } + } + + for (i0 = 1; i0 < argc; i0++) { + (*argv_out)[i++] = argv[i0]; + } + + /* Counting spaces is an upper bound, argv stays NULL terminated. */ + (*argc_out) = i; + while (i <= n) { + (*argv_out)[i++] = NULL; + } +} + +#if PYTHON_MODULE || SWIG_MODULE || NOT_MAIN +int xd3_main_cmdline (int argc, char **argv) +#else +int main (int argc, char **argv) +#endif +{ + static const char *flags = + "0123456789cdefhnqvDFJNORTVs:m:B:C:E:I:L:O:M:P:W:A::S::"; + xd3_cmd cmd; + main_file ifile; + main_file ofile; + main_file sfile; + main_merge_list merge_order; + main_merge *merge; + int my_optind; + const char *my_optarg; + const char *my_optstr; + const char *sfilename; + int env_argc; + char **env_argv; + char **free_argv; /* malloc() in setup_environment() */ + char *free_value; /* malloc() in setup_environment() */ + int ret; + +#ifdef _WIN32 + GetStartupInfo(&winStartupInfo); + setvbuf(stderr, NULL, _IONBF, 0); /* Do not buffer stderr */ +#endif + + main_file_init (& ifile); + main_file_init (& ofile); + main_file_init (& sfile); + main_merge_list_init (& merge_order); + + reset_defaults(); + + free_argv = NULL; + free_value = NULL; + setup_environment(argc, argv, &env_argc, &env_argv, + &free_argv, &free_value); + cmd = CMD_NONE; + sfilename = NULL; + my_optind = 1; + argv = env_argv; + argc = env_argc; + program_name = env_argv[0]; + + takearg: + my_optarg = NULL; + my_optstr = argv[my_optind]; + + /* This doesn't use getopt() because it makes trouble for -P & python which + * reenter main() and thus care about freeing all memory. I never had much + * trust for getopt anyway, it's too opaque. This implements a fairly + * standard non-long-option getopt with support for named operations (e.g., + * "xdelta3 [encode|decode|printhdr...] < in > out"). */ + if (my_optstr) + { + if (*my_optstr == '-') { my_optstr += 1; } + else if (cmd == CMD_NONE) { goto nonflag; } + else { my_optstr = NULL; } + } + while (my_optstr) + { + const char *s; + my_optarg = NULL; + if ((ret = *my_optstr++) == 0) { my_optind += 1; goto takearg; } + + /* Option handling: first check for one ':' following the option in + * flags, then check for two. The syntax allows: + * + * 1. -Afoo defines optarg="foo" + * 2. -A foo defines optarg="foo" + * 3. -A "" defines optarg="" (allows empty-string) + * 4. -A [EOA or -moreargs] error (mandatory case) + * 5. -A [EOA -moreargs] defines optarg=NULL (optional case) + * 6. -A=foo defines optarg="foo" + * 7. -A= defines optarg="" (mandatory case) + * 8. -A= defines optarg=NULL (optional case) + * + * See tests in test_command_line_arguments(). + */ + s = strchr (flags, ret); + if (s && s[1] && s[1] == ':') + { + int option = s[2] && s[2] == ':'; + + /* Case 1, set optarg to the remaining characters. */ + my_optarg = my_optstr; + my_optstr = ""; + + /* Case 2-5 */ + if (*my_optarg == 0) + { + /* Condition 4-5 */ + int have_arg = (my_optind < (argc - 1) && + *argv[my_optind+1] != '-'); + + if (! have_arg) + { + if (! option) + { + /* Case 4 */ + XPR(NT "-%c: requires an argument\n", ret); + ret = EXIT_FAILURE; + goto cleanup; + } + /* Case 5. */ + my_optarg = NULL; + } + else + { + /* Case 2-3. */ + my_optarg = argv[++my_optind]; + } + } + /* Case 6-8. */ + else if (*my_optarg == '=') + { + /* Remove the = in all cases. */ + my_optarg += 1; + + if (option && *my_optarg == 0) + { + /* Case 8. */ + my_optarg = NULL; + } + } + } + + switch (ret) + { + /* case: if no '-' was found, maybe check for a command name. */ + nonflag: + if (strcmp (my_optstr, "decode") == 0) { cmd = CMD_DECODE; } + else if (strcmp (my_optstr, "encode") == 0) + { +#if XD3_ENCODER + cmd = CMD_ENCODE; +#else + XPR(NT "encoder support not compiled\n"); + return EXIT_FAILURE; +#endif + } + else if (strcmp (my_optstr, "config") == 0) { cmd = CMD_CONFIG; } +#if REGRESSION_TEST + else if (strcmp (my_optstr, "test") == 0) { cmd = CMD_TEST; } +#endif +#if VCDIFF_TOOLS + else if (strcmp (my_optstr, "printhdr") == 0) { cmd = CMD_PRINTHDR; } + else if (strcmp (my_optstr, "printhdrs") == 0) + { cmd = CMD_PRINTHDRS; } + else if (strcmp (my_optstr, "printdelta") == 0) + { cmd = CMD_PRINTDELTA; } + else if (strcmp (my_optstr, "recode") == 0) { cmd = CMD_RECODE; } + else if (strcmp (my_optstr, "merge") == 0) { cmd = CMD_MERGE; } +#endif + + /* If no option was found and still no command, let the default + * command be encode. The remaining args are treated as + * filenames. */ + if (cmd == CMD_NONE) + { + cmd = CMD_DEFAULT; + my_optstr = NULL; + break; + } + else + { + /* But if we find a command name, continue the getopt loop. */ + my_optind += 1; + goto takearg; + } + + /* gzip-like options */ + case '0': case '1': case '2': case '3': case '4': + case '5': case '6': case '7': case '8': case '9': + option_level = ret - '0'; + break; + case 'f': option_force = 1; break; + case 'F': +#if EXTERNAL_COMPRESSION + option_force2 = 1; +#else + XPR(NT "warning: -F option ignored, " + "external compression support was not compiled\n"); + break; +#endif + case 'v': option_verbose += 1; option_quiet = 0; break; + case 'q': option_quiet = 1; option_verbose = 0; break; + case 'c': option_stdout = 1; break; + case 'd': + if (cmd == CMD_NONE) { cmd = CMD_DECODE; } + else { ret = main_help (); goto exit; } + break; + case 'e': +#if XD3_ENCODER + if (cmd == CMD_NONE) { cmd = CMD_ENCODE; } + else { ret = main_help (); goto exit; } + break; +#else + XPR(NT "encoder support not compiled\n"); + return EXIT_FAILURE; +#endif + + case 'n': option_use_checksum = 0; break; + case 'N': option_no_compress = 1; break; + case 'T': option_use_altcodetable = 1; break; + case 'C': option_smatch_config = my_optarg; break; + case 'J': option_no_output = 1; break; + case 'S': if (my_optarg == NULL) + { + option_use_secondary = 1; + option_secondary = "none"; + } + else + { + option_use_secondary = 1; + option_secondary = my_optarg; + } + break; + case 'A': if (my_optarg == NULL) { option_use_appheader = 0; } + else { option_appheader = (uint8_t*) my_optarg; } break; + case 'B': { + xoff_t bsize; + if ((ret = main_atoux (my_optarg, & bsize, + XD3_MINSRCWINSZ, XD3_MAXSRCWINSZ, 'B'))) + { + goto exit; + } + option_srcwinsz = bsize; + break; + } + case 'I': + if ((ret = main_atou (my_optarg, & option_iopt_size, 0, + 0, 'I'))) + { + goto exit; + } + break; + case 'P': + if ((ret = main_atou (my_optarg, & option_sprevsz, 0, + 0, 'P'))) + { + goto exit; + } + break; + case 'W': + if ((ret = main_atou (my_optarg, & option_winsize, XD3_ALLOCSIZE, + XD3_HARDMAXWINSIZE, 'W'))) + { + goto exit; + } + break; + case 'D': +#if EXTERNAL_COMPRESSION == 0 + if (option_verbose > 0) + { + XPR(NT "warning: -D option ignored, " + "external compression support was not compiled\n"); + } +#else + option_decompress_inputs = 0; +#endif + break; + case 'R': +#if EXTERNAL_COMPRESSION == 0 + if (option_verbose > 0) + { + XPR(NT "warning: -R option ignored, " + "external compression support was not compiled\n"); + } +#else + option_recompress_outputs = 0; +#endif + break; + case 's': + if (sfilename != NULL) + { + XPR(NT "specify only one source file\n"); + goto cleanup; + } + + sfilename = my_optarg; + break; + case 'm': + if ((merge = (main_merge*) + main_malloc (sizeof (main_merge))) == NULL) + { + goto cleanup; + } + main_merge_list_push_back (& merge_order, merge); + merge->filename = my_optarg; + break; + case 'V': + ret = main_version (); goto exit; + default: + ret = main_help (); goto exit; + } + } + + option_source_filename = sfilename; + + /* In case there were no arguments, set the default command. */ + if (cmd == CMD_NONE) { cmd = CMD_DEFAULT; } + + argc -= my_optind; + argv += my_optind; + + /* There may be up to two more arguments. */ + if (argc > 2) + { + XPR(NT "too many filenames: %s ...\n", argv[2]); + goto cleanup; + } + + ifile.flags = RD_FIRST | RD_MAININPUT; + sfile.flags = RD_FIRST; + sfile.filename = option_source_filename; + + /* The infile takes the next argument, if there is one. But if not, infile + * is set to stdin. */ + if (argc > 0) + { + ifile.filename = argv[0]; + + if ((ret = main_file_open (& ifile, ifile.filename, XO_READ))) + { + goto cleanup; + } + } + else + { + XSTDIN_XF (& ifile); + } + + /* The ofile takes the following argument, if there is one. But if not, it + * is left NULL until the application header is processed. It will be set + * in main_open_output. */ + if (argc > 1) + { + /* Check for conflicting arguments. */ + if (option_stdout && ! option_quiet) + { + XPR(NT "warning: -c option overrides output filename: %s\n", + argv[1]); + } + + if (! option_stdout) { ofile.filename = argv[1]; } + } + +#if VCDIFF_TOOLS + if (cmd == CMD_MERGE && + (ret = main_merge_arguments (&merge_order))) + { + goto cleanup; + } +#endif /* VCDIFF_TOOLS */ + + switch (cmd) + { + case CMD_PRINTHDR: + case CMD_PRINTHDRS: + case CMD_PRINTDELTA: +#if XD3_ENCODER + case CMD_ENCODE: + case CMD_RECODE: + case CMD_MERGE: +#endif + case CMD_DECODE: + ret = main_input (cmd, & ifile, & ofile, & sfile); + break; + +#if REGRESSION_TEST + case CMD_TEST: + main_config (); + ret = xd3_selftest (); + break; +#endif + + case CMD_CONFIG: + ret = main_config (); + break; + + default: + ret = main_help (); + break; + } + + if (0) + { + cleanup: + ret = EXIT_FAILURE; + exit: + (void)0; + } + +#if EXTERNAL_COMPRESSION + main_external_compression_cleanup (); +#endif + + main_file_cleanup (& ifile); + main_file_cleanup (& ofile); + main_file_cleanup (& sfile); + + while (! main_merge_list_empty (& merge_order)) + { + merge = main_merge_list_pop_front (& merge_order); + main_free (merge); + } + + main_free (free_argv); + main_free (free_value); + + main_cleanup (); + + fflush (stdout); + fflush (stderr); + return ret; +} + +static int +main_help (void) +{ + main_version(); + + /* Note: update wiki when command-line features change */ + XPR(NTR "usage: xdelta3 [command/options] [input [output]]\n"); + XPR(NTR "make patch:\n"); + XPR(NTR "\n"); + XPR(NTR " xdelta3.exe -e -s old_file new_file delta_file\n"); + XPR(NTR "\n"); + XPR(NTR "apply patch:\n"); + XPR(NTR "\n"); + XPR(NTR " xdelta3.exe -d -s old_file delta_file decoded_new_file\n"); + XPR(NTR "\n"); + XPR(NTR "special command names:\n"); + XPR(NTR " config prints xdelta3 configuration\n"); + XPR(NTR " decode decompress the input\n"); + XPR(NTR " encode compress the input%s\n", + XD3_ENCODER ? "" : " [Not compiled]"); +#if REGRESSION_TEST + XPR(NTR " test run the builtin tests\n"); +#endif +#if VCDIFF_TOOLS + XPR(NTR "special commands for VCDIFF inputs:\n"); + XPR(NTR " printdelta print information about the entire delta\n"); + XPR(NTR " printhdr print information about the first window\n"); + XPR(NTR " printhdrs print information about all windows\n"); + XPR(NTR " recode encode with new application/secondary settings\n"); + XPR(NTR " merge merge VCDIFF inputs (see below)\n"); +#endif + XPR(NTR "merge patches:\n"); + XPR(NTR "\n"); + XPR(NTR " xdelta3 merge -m 1.vcdiff -m 2.vcdiff 3.vcdiff merged.vcdiff\n"); + XPR(NTR "\n"); + XPR(NTR "standard options:\n"); + XPR(NTR " -0 .. -9 compression level\n"); + XPR(NTR " -c use stdout\n"); + XPR(NTR " -d decompress\n"); + XPR(NTR " -e compress%s\n", + XD3_ENCODER ? "" : " [Not compiled]"); + XPR(NTR " -f force (overwrite, ignore trailing garbage)\n"); +#if EXTERNAL_COMPRESSION + XPR(NTR " -F force the external-compression subprocess\n"); +#endif + XPR(NTR " -h show help\n"); + XPR(NTR " -q be quiet\n"); + XPR(NTR " -v be verbose (max 2)\n"); + XPR(NTR " -V show version\n"); + + XPR(NTR "memory options:\n"); + XPR(NTR " -B bytes source window size\n"); + XPR(NTR " -W bytes input window size\n"); + XPR(NTR " -P size compression duplicates window\n"); + XPR(NTR " -I size instruction buffer size (0 = unlimited)\n"); + + XPR(NTR "compression options:\n"); + XPR(NTR " -s source source file to copy from (if any)\n"); + XPR(NTR " -S [djw|fgk] enable/disable secondary compression\n"); + XPR(NTR " -N disable small string-matching compression\n"); + XPR(NTR " -D disable external decompression (encode/decode)\n"); + XPR(NTR " -R disable external recompression (decode)\n"); + XPR(NTR " -n disable checksum (encode/decode)\n"); + XPR(NTR " -C soft config (encode, undocumented)\n"); + XPR(NTR " -A [apphead] disable/provide application header (encode)\n"); + XPR(NTR " -J disable output (check/compute only)\n"); + XPR(NTR " -T use alternate code table (test)\n"); + XPR(NTR " -m arguments for \"merge\"\n"); + + XPR(NTR "the XDELTA environment variable may contain extra args:\n"); + XPR(NTR " XDELTA=\"-s source-x.y.tar.gz\" \\\n"); + XPR(NTR " tar --use-compress-program=xdelta3 \\\n"); + XPR(NTR " -cf target-x.z.tar.gz.vcdiff target-x.y\n"); + return EXIT_FAILURE; +} diff --git a/3rdparty/xdelta3-3.0.8/xdelta3-merge.h b/3rdparty/xdelta3-3.0.8/xdelta3-merge.h new file mode 100644 index 0000000..e65978d --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3-merge.h @@ -0,0 +1,579 @@ +/* xdelta 3 - delta compression tools and library + * Copyright (C) 2007. Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +#ifndef _XDELTA3_MERGE_H_ +#define _XDELTA3_MERGE_H_ + +int xd3_merge_inputs (xd3_stream *stream, + xd3_whole_state *source, + xd3_whole_state *input); + +static int +xd3_whole_state_init (xd3_stream *stream) +{ + XD3_ASSERT (stream->whole_target.adds == NULL); + XD3_ASSERT (stream->whole_target.inst == NULL); + XD3_ASSERT (stream->whole_target.wininfo == NULL); + XD3_ASSERT (stream->whole_target.length == 0); + + stream->whole_target.adds_alloc = XD3_ALLOCSIZE; + stream->whole_target.inst_alloc = XD3_ALLOCSIZE; + stream->whole_target.wininfo_alloc = XD3_ALLOCSIZE; + + if ((stream->whole_target.adds = (uint8_t*) + xd3_alloc (stream, stream->whole_target.adds_alloc, 1)) == NULL || + (stream->whole_target.inst = (xd3_winst*) + xd3_alloc (stream, stream->whole_target.inst_alloc, 1)) == NULL || + (stream->whole_target.wininfo = (xd3_wininfo*) + xd3_alloc (stream, stream->whole_target.wininfo_alloc, 1)) == NULL) + { + return ENOMEM; + } + return 0; +} + +static void +xd3_swap_whole_state (xd3_whole_state *a, + xd3_whole_state *b) +{ + xd3_whole_state tmp; + XD3_ASSERT (a->inst != NULL && a->adds != NULL); + XD3_ASSERT (b->inst != NULL && b->adds != NULL); + XD3_ASSERT (b->wininfo != NULL && b->wininfo != NULL); + memcpy (&tmp, a, sizeof (xd3_whole_state)); + memcpy (a, b, sizeof (xd3_whole_state)); + memcpy (b, &tmp, sizeof (xd3_whole_state)); +} + +static int +xd3_realloc_buffer (xd3_stream *stream, + usize_t current_units, + usize_t unit_size, + usize_t new_units, + usize_t *alloc_size, + void **alloc_ptr) +{ + usize_t needed; + usize_t new_alloc; + usize_t cur_size; + uint8_t *new_buf; + + needed = (current_units + new_units) * unit_size; + + if (needed <= *alloc_size) + { + return 0; + } + + cur_size = current_units * unit_size; + new_alloc = xd3_round_blksize (needed * 2, XD3_ALLOCSIZE); + + if ((new_buf = (uint8_t*) xd3_alloc (stream, new_alloc, 1)) == NULL) + { + return ENOMEM; + } + + if (cur_size != 0) + { + memcpy (new_buf, *alloc_ptr, cur_size); + } + + if (*alloc_ptr != NULL) + { + xd3_free (stream, *alloc_ptr); + } + + *alloc_size = new_alloc; + *alloc_ptr = new_buf; + + return 0; +} + +/* allocate one new output instruction */ +static int +xd3_whole_alloc_winst (xd3_stream *stream, + xd3_winst **winstp) +{ + int ret; + + if ((ret = xd3_realloc_buffer (stream, + stream->whole_target.instlen, + sizeof (xd3_winst), + 1, + & stream->whole_target.inst_alloc, + (void**) & stream->whole_target.inst))) + { + return ret; + } + + *winstp = &stream->whole_target.inst[stream->whole_target.instlen++]; + + return 0; +} + +static int +xd3_whole_alloc_adds (xd3_stream *stream, + usize_t count) +{ + return xd3_realloc_buffer (stream, + stream->whole_target.addslen, + 1, + count, + & stream->whole_target.adds_alloc, + (void**) & stream->whole_target.adds); +} + +static int +xd3_whole_alloc_wininfo (xd3_stream *stream, + xd3_wininfo **wininfop) +{ + int ret; + + if ((ret = xd3_realloc_buffer (stream, + stream->whole_target.wininfolen, + sizeof (xd3_wininfo), + 1, + & stream->whole_target.wininfo_alloc, + (void**) & stream->whole_target.wininfo))) + { + return ret; + } + + *wininfop = &stream->whole_target.wininfo[stream->whole_target.wininfolen++]; + + return 0; +} + +static int +xd3_whole_append_inst (xd3_stream *stream, + xd3_hinst *inst) +{ + int ret; + xd3_winst *winst; + + if ((ret = xd3_whole_alloc_winst (stream, &winst))) + { + return ret; + } + + winst->type = inst->type; + winst->mode = 0; + winst->size = inst->size; + winst->position = stream->whole_target.length; + stream->whole_target.length += inst->size; + + if (((inst->type == XD3_ADD) || (inst->type == XD3_RUN)) && + (ret = xd3_whole_alloc_adds (stream, + (inst->type == XD3_RUN ? 1 : inst->size)))) + { + return ret; + } + + switch (inst->type) + { + case XD3_RUN: + winst->addr = stream->whole_target.addslen; + stream->whole_target.adds[stream->whole_target.addslen++] = + *stream->data_sect.buf++; + break; + + case XD3_ADD: + winst->addr = stream->whole_target.addslen; + memcpy (stream->whole_target.adds + stream->whole_target.addslen, + stream->data_sect.buf, + inst->size); + stream->data_sect.buf += inst->size; + stream->whole_target.addslen += inst->size; + break; + + default: + if (inst->addr < stream->dec_cpylen) + { + winst->mode = SRCORTGT (stream->dec_win_ind); + winst->addr = stream->dec_cpyoff + inst->addr; + } + else + { + winst->addr = (stream->dec_winstart + + inst->addr - + stream->dec_cpylen); + } + break; + } + + return 0; +} + +int +xd3_whole_append_window (xd3_stream *stream) +{ + int ret; + xd3_wininfo *wininfo; + + if ((ret = xd3_whole_alloc_wininfo (stream, &wininfo))) { return ret; } + + wininfo->length = stream->dec_tgtlen; + wininfo->offset = stream->dec_winstart; + wininfo->adler32 = stream->dec_adler32; + + while (stream->inst_sect.buf < stream->inst_sect.buf_max) + { + if ((ret = xd3_decode_instruction (stream))) + { + return ret; + } + + if ((stream->dec_current1.type != XD3_NOOP) && + (ret = xd3_whole_append_inst (stream, + & stream->dec_current1))) + { + return ret; + } + + if ((stream->dec_current2.type != XD3_NOOP) && + (ret = xd3_whole_append_inst (stream, + & stream->dec_current2))) + { + return ret; + } + } + + return 0; +} + +/* xd3_merge_input_output applies *source to *stream, returns the + * result in stream. */ +int xd3_merge_input_output (xd3_stream *stream, + xd3_whole_state *source) +{ + int ret; + xd3_stream tmp_stream; + memset (& tmp_stream, 0, sizeof (tmp_stream)); + if ((ret = xd3_config_stream (& tmp_stream, NULL)) || + (ret = xd3_whole_state_init (& tmp_stream)) || + (ret = xd3_merge_inputs (& tmp_stream, + source, + & stream->whole_target))) + { + XPR(NT XD3_LIB_ERRMSG (&tmp_stream, ret)); + return ret; + } + + /* the output is in tmp_stream.whole_state, swap into input */ + xd3_swap_whole_state (& stream->whole_target, + & tmp_stream.whole_target); + /* total allocation counts are preserved */ + xd3_free_stream (& tmp_stream); + return 0; +} + +static int +xd3_merge_run (xd3_stream *stream, + xd3_whole_state *target, + xd3_winst *iinst) +{ + int ret; + xd3_winst *oinst; + + if ((ret = xd3_whole_alloc_winst (stream, &oinst)) || + (ret = xd3_whole_alloc_adds (stream, 1))) + { + return ret; + } + + oinst->type = iinst->type; + oinst->mode = iinst->mode; + oinst->size = iinst->size; + oinst->addr = stream->whole_target.addslen; + + XD3_ASSERT (stream->whole_target.length == iinst->position); + oinst->position = stream->whole_target.length; + stream->whole_target.length += iinst->size; + + stream->whole_target.adds[stream->whole_target.addslen++] = + target->adds[iinst->addr]; + + return 0; +} + +static int +xd3_merge_add (xd3_stream *stream, + xd3_whole_state *target, + xd3_winst *iinst) +{ + int ret; + xd3_winst *oinst; + + if ((ret = xd3_whole_alloc_winst (stream, &oinst)) || + (ret = xd3_whole_alloc_adds (stream, iinst->size))) + { + return ret; + } + + oinst->type = iinst->type; + oinst->mode = iinst->mode; + oinst->size = iinst->size; + oinst->addr = stream->whole_target.addslen; + + XD3_ASSERT (stream->whole_target.length == iinst->position); + oinst->position = stream->whole_target.length; + stream->whole_target.length += iinst->size; + + memcpy(stream->whole_target.adds + stream->whole_target.addslen, + target->adds + iinst->addr, + iinst->size); + + stream->whole_target.addslen += iinst->size; + + return 0; +} + +static int +xd3_merge_target_copy (xd3_stream *stream, + xd3_winst *iinst) +{ + int ret; + xd3_winst *oinst; + + if ((ret = xd3_whole_alloc_winst (stream, &oinst))) + { + return ret; + } + + XD3_ASSERT (stream->whole_target.length == iinst->position); + + memcpy (oinst, iinst, sizeof (*oinst)); + return 0; +} + +static int +xd3_merge_find_position (xd3_stream *stream, + xd3_whole_state *source, + xoff_t address, + usize_t *inst_num) +{ + usize_t low; + usize_t high; + + if (address >= source->length) + { + stream->msg = "Invalid copy offset in merge"; + return XD3_INVALID_INPUT; + } + + low = 0; + high = source->instlen; + + while (low != high) + { + xoff_t mid_lpos; + xoff_t mid_hpos; + usize_t mid = low + (high - low) / 2; + mid_lpos = source->inst[mid].position; + + if (address < mid_lpos) + { + high = mid; + continue; + } + + mid_hpos = mid_lpos + source->inst[mid].size; + + if (address >= mid_hpos) + { + low = mid + 1; + continue; + } + + *inst_num = mid; + return 0; + } + + stream->msg = "Internal error in merge"; + return XD3_INTERNAL; +} + +static int +xd3_merge_source_copy (xd3_stream *stream, + xd3_whole_state *source, + const xd3_winst *iinst_orig) +{ + int ret; + xd3_winst iinst; + usize_t sinst_num; + + memcpy (& iinst, iinst_orig, sizeof (iinst)); + + XD3_ASSERT (iinst.mode == VCD_SOURCE); + + if ((ret = xd3_merge_find_position (stream, source, + iinst.addr, &sinst_num))) + { + return ret; + } + + while (iinst.size > 0) + { + xd3_winst *sinst; + xd3_winst *minst; + usize_t sinst_offset; + usize_t sinst_left; + usize_t this_take; + + XD3_ASSERT (sinst_num < source->instlen); + + sinst = &source->inst[sinst_num]; + + XD3_ASSERT (iinst.addr >= sinst->position); + + sinst_offset = (usize_t)(iinst.addr - sinst->position); + + XD3_ASSERT (sinst->size > sinst_offset); + + sinst_left = sinst->size - sinst_offset; + this_take = min (iinst.size, sinst_left); + + XD3_ASSERT (this_take > 0); + + if ((ret = xd3_whole_alloc_winst (stream, &minst))) + { + return ret; + } + + minst->size = this_take; + minst->type = sinst->type; + minst->position = iinst.position; + minst->mode = 0; + + switch (sinst->type) + { + case XD3_RUN: + if ((ret = xd3_whole_alloc_adds (stream, 1))) + { + return ret; + } + + minst->addr = stream->whole_target.addslen; + stream->whole_target.adds[stream->whole_target.addslen++] = + source->adds[sinst->addr]; + break; + case XD3_ADD: + if ((ret = xd3_whole_alloc_adds (stream, this_take))) + { + return ret; + } + + minst->addr = stream->whole_target.addslen; + memcpy(stream->whole_target.adds + stream->whole_target.addslen, + source->adds + sinst->addr + sinst_offset, + this_take); + stream->whole_target.addslen += this_take; + break; + default: + if (sinst->mode != 0) + { + minst->mode = sinst->mode; + minst->addr = sinst->addr + sinst_offset; + } + else + { + // TODO: this is slow because of the recursion, which + // could reach a depth equal to the number of target + // copies, and this is compression-inefficient because + // it can produce duplicate adds. + xd3_winst tinst; + tinst.type = XD3_CPY; + tinst.mode = iinst.mode; + tinst.addr = sinst->addr + sinst_offset; + tinst.size = this_take; + tinst.position = iinst.position; + + // The instruction allocated in this frame will not be used. + stream->whole_target.instlen -= 1; + + if ((ret = xd3_merge_source_copy (stream, source, &tinst))) + { + return ret; + } + } + break; + } + + iinst.position += this_take; + iinst.addr += this_take; + iinst.size -= this_take; + sinst_num += 1; + } + + return 0; +} + +/* xd3_merge_inputs() applies *input to *source, returns its result in + * stream. */ +int xd3_merge_inputs (xd3_stream *stream, + xd3_whole_state *source, + xd3_whole_state *input) +{ + int ret = 0; + usize_t i; + size_t input_i; + + for (i = 0; i < input->wininfolen; ++i) { + xd3_wininfo *copyinfo; + + if ((ret = xd3_whole_alloc_wininfo (stream, ©info))) { return ret; } + + *copyinfo = input->wininfo[i]; + } + + /* iterate over each instruction. */ + for (input_i = 0; ret == 0 && input_i < input->instlen; ++input_i) + { + xd3_winst *iinst = &input->inst[input_i]; + + switch (iinst->type) + { + case XD3_RUN: + ret = xd3_merge_run (stream, input, iinst); + break; + case XD3_ADD: + ret = xd3_merge_add (stream, input, iinst); + break; + default: + /* TODO: VCD_TARGET support is completely untested all + * throughout. */ + if (iinst->mode == 0 || iinst->mode == VCD_TARGET) + { + ret = xd3_merge_target_copy (stream, iinst); + } + else + { + ret = xd3_merge_source_copy (stream, source, iinst); + } + + /* The whole_target.length is not updated in the xd3_merge*copy + * routine because of recursion in xd3_merge_source_copy. */ + stream->whole_target.length += iinst->size; + break; + } + } + + return ret; +} + +#endif diff --git a/3rdparty/xdelta3-3.0.8/xdelta3-second.h b/3rdparty/xdelta3-3.0.8/xdelta3-second.h new file mode 100644 index 0000000..1c8a406 --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3-second.h @@ -0,0 +1,315 @@ +/* xdelta 3 - delta compression tools and library + * Copyright (C) 2002, 2003, 2006, 2007, 2013. Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +#ifndef _XDELTA3_SECOND_H_ +#define _XDELTA3_SECOND_H_ + +static inline void xd3_bit_state_encode_init (bit_state *bits) +{ + bits->cur_byte = 0; + bits->cur_mask = 1; +} + +static inline int xd3_decode_bits (xd3_stream *stream, + bit_state *bits, + const uint8_t **input, + const uint8_t *input_max, + usize_t nbits, + usize_t *valuep) +{ + usize_t value = 0; + usize_t vmask = 1 << nbits; + + if (bits->cur_mask == 0x100) { goto next_byte; } + + for (;;) + { + do + { + vmask >>= 1; + + if (bits->cur_byte & bits->cur_mask) + { + value |= vmask; + } + + bits->cur_mask <<= 1; + + if (vmask == 1) { goto done; } + } + while (bits->cur_mask != 0x100); + + next_byte: + + if (*input == input_max) + { + stream->msg = "secondary decoder end of input"; + return XD3_INTERNAL; + } + + bits->cur_byte = *(*input)++; + bits->cur_mask = 1; + } + + done: + + IF_DEBUG2 (DP(RINT "(d) %u ", value)); + + (*valuep) = value; + return 0; +} + +#if REGRESSION_TEST +/* There may be extra bits at the end of secondary decompression, this macro + * checks for non-zero bits. This is overly strict, but helps pass the + * single-bit-error regression test. */ +static int +xd3_test_clean_bits (xd3_stream *stream, bit_state *bits) +{ + for (; bits->cur_mask != 0x100; bits->cur_mask <<= 1) + { + if (bits->cur_byte & bits->cur_mask) + { + stream->msg = "secondary decoder garbage"; + return XD3_INTERNAL; + } + } + + return 0; +} +#endif + +static int +xd3_get_secondary (xd3_stream *stream, xd3_sec_stream **sec_streamp, + int is_encode) +{ + if (*sec_streamp == NULL) + { + int ret; + + if ((*sec_streamp = stream->sec_type->alloc (stream)) == NULL) + { + stream->msg = "error initializing secondary stream"; + return XD3_INVALID; + } + + if ((ret = stream->sec_type->init (stream, *sec_streamp, is_encode)) != 0) + { + return ret; + } + } + + return 0; +} + +static int +xd3_decode_secondary (xd3_stream *stream, + xd3_desect *sect, + xd3_sec_stream **sec_streamp) +{ + uint32_t dec_size; + uint8_t *out_used; + int ret; + + if ((ret = xd3_get_secondary (stream, sec_streamp, 0)) != 0) + { + return ret; + } + + /* Decode the size, allocate the buffer. */ + if ((ret = xd3_read_size (stream, & sect->buf, + sect->buf_max, & dec_size)) || + (ret = xd3_decode_allocate (stream, dec_size, + & sect->copied2, & sect->alloc2))) + { + return ret; + } + + out_used = sect->copied2; + + if ((ret = stream->sec_type->decode (stream, *sec_streamp, + & sect->buf, sect->buf_max, + & out_used, out_used + dec_size))) + { + return ret; + } + + if (sect->buf != sect->buf_max) + { + stream->msg = "secondary decoder finished with unused input"; + return XD3_INTERNAL; + } + + if (out_used != sect->copied2 + dec_size) + { + stream->msg = "secondary decoder short output"; + return XD3_INTERNAL; + } + + sect->buf = sect->copied2; + sect->buf_max = sect->copied2 + dec_size; + sect->size = dec_size; + + return 0; +} + +#if XD3_ENCODER +static inline int xd3_encode_bit (xd3_stream *stream, + xd3_output **output, + bit_state *bits, + usize_t bit) +{ + int ret; + + if (bit) + { + bits->cur_byte |= bits->cur_mask; + } + + /* OPT: Might help to buffer more than 8 bits at once. */ + if (bits->cur_mask == 0x80) + { + if ((ret = xd3_emit_byte (stream, output, bits->cur_byte)) != 0) + { + return ret; + } + + bits->cur_mask = 1; + bits->cur_byte = 0; + } + else + { + bits->cur_mask <<= 1; + } + + return 0; +} + +static inline int xd3_flush_bits (xd3_stream *stream, + xd3_output **output, + bit_state *bits) +{ + return (bits->cur_mask == 1) ? 0 : + xd3_emit_byte (stream, output, bits->cur_byte); +} + +static inline int xd3_encode_bits (xd3_stream *stream, + xd3_output **output, + bit_state *bits, + usize_t nbits, + usize_t value) +{ + int ret; + usize_t mask = 1 << nbits; + + XD3_ASSERT (nbits > 0); + XD3_ASSERT (nbits < sizeof (usize_t) * 8); + XD3_ASSERT (value < mask); + + do + { + mask >>= 1; + + if ((ret = xd3_encode_bit (stream, output, bits, value & mask))) + { + return ret; + } + } + while (mask != 1); + + IF_DEBUG2 (DP(RINT "(e) %u ", value)); + + return 0; +} + +static int +xd3_encode_secondary (xd3_stream *stream, + xd3_output **head, + xd3_output **tail, + xd3_sec_stream **sec_streamp, + xd3_sec_cfg *cfg, + int *did_it) +{ + xd3_output *tmp_head; + xd3_output *tmp_tail; + + usize_t comp_size; + usize_t orig_size; + + int ret; + + orig_size = xd3_sizeof_output (*head); + + if (orig_size < SECONDARY_MIN_INPUT) { return 0; } + + if ((ret = xd3_get_secondary (stream, sec_streamp, 1)) != 0) + { + return ret; + } + + tmp_head = xd3_alloc_output (stream, NULL); + + /* Encode the size, encode the data. Encoding the size makes it + * simpler, but is a little gross. Should not need the entire + * section in contiguous memory, but it is much easier this way. */ + if ((ret = xd3_emit_size (stream, & tmp_head, orig_size)) || + (ret = stream->sec_type->encode (stream, *sec_streamp, *head, + tmp_head, cfg))) + { + goto getout; + } + + /* If the secondary compressor determines it's no good, it returns + * XD3_NOSECOND. */ + + /* Setup tmp_tail, comp_size */ + tmp_tail = tmp_head; + comp_size = tmp_head->next; + + while (tmp_tail->next_page != NULL) + { + tmp_tail = tmp_tail->next_page; + comp_size += tmp_tail->next; + } + + XD3_ASSERT (comp_size == xd3_sizeof_output (tmp_head)); + XD3_ASSERT (tmp_tail != NULL); + + if (comp_size < (orig_size - SECONDARY_MIN_SAVINGS) || cfg->inefficient) + { + IF_DEBUG1(DP(RINT "secondary saved %u bytes: %u -> %u (%0.2f%%)\n", + orig_size - comp_size, orig_size, comp_size, + 100.0 * (double) comp_size / (double) orig_size)); + + xd3_free_output (stream, *head); + + *head = tmp_head; + *tail = tmp_tail; + *did_it = 1; + } + else + { + getout: + if (ret == XD3_NOSECOND) { ret = 0; } + xd3_free_output (stream, tmp_head); + } + + return ret; +} +#endif /* XD3_ENCODER */ +#endif /* _XDELTA3_SECOND_H_ */ diff --git a/3rdparty/xdelta3-3.0.8/xdelta3-test.h b/3rdparty/xdelta3-3.0.8/xdelta3-test.h new file mode 100644 index 0000000..e9848b6 --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3-test.h @@ -0,0 +1,2872 @@ +/* xdelta 3 - delta compression tools and library Copyright (C) 2001, + * 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012. + * Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +/* This is public-domain Mersenne Twister code, + * attributed to Michael Brundage. Thanks! + * http://www.qbrundage.com/michaelb/pubs/essays/random_number_generation.html + */ +static const uint32_t TEST_SEED1 = 5489UL; +#define MT_LEN 624 +#define MT_IA 397 +static const uint32_t UPPER_MASK = 0x80000000; +static const uint32_t LOWER_MASK = 0x7FFFFFFF; +static const uint32_t MATRIX_A = 0x9908B0DF; + +#ifndef SHELL_TESTS +#define SHELL_TESTS 1 +#endif + +typedef struct mtrand mtrand; + +struct mtrand { + int mt_index_; + uint32_t mt_buffer_[MT_LEN]; +}; + +int test_compare_files (const char* tgt, const char *rec); +void mt_init(mtrand *mt, uint32_t seed); +uint32_t mt_random (mtrand *mt); +int test_setup (void); + +void mt_init(mtrand *mt, uint32_t seed) { + int i; + mt->mt_buffer_[0] = seed; + mt->mt_index_ = MT_LEN; + for (i = 1; i < MT_LEN; i++) { + /* See Knuth TAOCP Vol2. 3rd Ed. P.106 for multiplier. */ + /* In the previous versions, MSBs of the seed affect */ + /* only MSBs of the array mt[]. */ + /* 2002/01/09 modified by Makoto Matsumoto */ + mt->mt_buffer_[i] = + (1812433253UL * (mt->mt_buffer_[i-1] ^ + (mt->mt_buffer_[i-1] >> 30)) + i); + } +} + +uint32_t mt_random (mtrand *mt) { + uint32_t y; + unsigned long mag01[2]; + mag01[0] = 0; + mag01[1] = MATRIX_A; + + if (mt->mt_index_ >= MT_LEN) { + int kk; + + for (kk = 0; kk < MT_LEN - MT_IA; kk++) { + y = (mt->mt_buffer_[kk] & UPPER_MASK) | + (mt->mt_buffer_[kk + 1] & LOWER_MASK); + mt->mt_buffer_[kk] = mt->mt_buffer_[kk + MT_IA] ^ + (y >> 1) ^ mag01[y & 0x1UL]; + } + for (;kk < MT_LEN - 1; kk++) { + y = (mt->mt_buffer_[kk] & UPPER_MASK) | + (mt->mt_buffer_[kk + 1] & LOWER_MASK); + mt->mt_buffer_[kk] = mt->mt_buffer_[kk + (MT_IA - MT_LEN)] ^ + (y >> 1) ^ mag01[y & 0x1UL]; + } + y = (mt->mt_buffer_[MT_LEN - 1] & UPPER_MASK) | + (mt->mt_buffer_[0] & LOWER_MASK); + mt->mt_buffer_[MT_LEN - 1] = mt->mt_buffer_[MT_IA - 1] ^ + (y >> 1) ^ mag01[y & 0x1UL]; + mt->mt_index_ = 0; + } + + y = mt->mt_buffer_[mt->mt_index_++]; + + y ^= (y >> 11); + y ^= (y << 7) & 0x9d2c5680UL; + y ^= (y << 15) & 0xefc60000UL; + y ^= (y >> 18); + + return y; +} + +static mtrand static_mtrand; + +#include + +static uint32_t +mt_exp_rand (uint32_t mean, uint32_t max_value) +{ + double mean_d = mean; + double erand = log (1.0 / (mt_random (&static_mtrand) / + (double)UINT32_MAX)); + uint32_t x = (uint32_t) (mean_d * erand + 0.5); + + return min (x, max_value); +} + +#if SHELL_TESTS +#include +#endif + +#define MSG_IS(x) (stream->msg != NULL && strcmp ((x), stream->msg) == 0) + +static const usize_t TWO_MEGS_AND_DELTA = (3 << 20); +static const usize_t ADDR_CACHE_ROUNDS = 10000; + +static const usize_t TEST_FILE_MEAN = 16384; +static const double TEST_ADD_MEAN = 128; +static const double TEST_ADD_MAX = 512; +static const double TEST_ADD_RATIO = 0.1; +static const double TEST_EPSILON = 0.25; + +#define TESTBUFSIZE (1024 * 16) + +#define TESTFILESIZE (1024) + +static char TEST_TARGET_FILE[TESTFILESIZE]; +static char TEST_SOURCE_FILE[TESTFILESIZE]; +static char TEST_DELTA_FILE[TESTFILESIZE]; +static char TEST_RECON_FILE[TESTFILESIZE]; +static char TEST_RECON2_FILE[TESTFILESIZE]; +static char TEST_COPY_FILE[TESTFILESIZE]; +static char TEST_NOPERM_FILE[TESTFILESIZE]; + +#define CHECK(cond) if (!(cond)) { XPR(NT "check failure: " #cond); abort(); } + +#if SHELL_TESTS +/* Use a fixed soft config so that test values are fixed. See also + * test_compress_text(). */ +static const char* test_softcfg_str = "-C9,3,4,8,2,36,70"; +#endif + +/*********************************************************************** + TEST HELPERS + ***********************************************************************/ + +static void DOT (void) { XPR(NTR "."); } +static int do_cmd (xd3_stream *stream, const char *buf) +{ + int ret; + if ((ret = system (buf)) != 0) + { + if (WIFEXITED (ret)) + { + stream->msg = "command exited non-zero"; + IF_DEBUG1 (XPR(NT "command was: %s\n", buf)); + } + else + { + stream->msg = "abnormal command termination"; + } + return XD3_INTERNAL; + } + return 0; +} + +static int do_fail (xd3_stream *stream, const char *buf) +{ + int ret; + ret = system (buf); + if (! WIFEXITED (ret) || WEXITSTATUS (ret) != 1) + { + stream->msg = "command should have not succeeded"; + XPR(NT "command was %s\n", buf); + return XD3_INTERNAL; + } + return 0; +} + +/* Test that the exponential distribution actually produces its mean. */ +static int +test_random_numbers (xd3_stream *stream, int ignore) +{ + usize_t i; + usize_t sum = 0; + usize_t mean = 50; + usize_t n_rounds = 1000000; + double average, error; + double allowed_error = 0.1; + + mt_init (& static_mtrand, 0x9f73f7fe); + + for (i = 0; i < n_rounds; i += 1) + { + sum += mt_exp_rand (mean, USIZE_T_MAX); + } + + average = (double) sum / (double) n_rounds; + error = average - (double) mean; + + if (error < allowed_error && error > -allowed_error) + { + return 0; + } + + /*XPR(NT "error is %f\n", error);*/ + stream->msg = "random distribution looks broken"; + return XD3_INTERNAL; +} + +static void +test_unlink (char* file) +{ + int ret; + if ((ret = unlink (file)) != 0 && errno != ENOENT) + { + XPR(NT "unlink %s failed: %s\n", file, strerror(ret)); + } +} + +static void +test_cleanup (void) +{ +#if 1 + test_unlink (TEST_TARGET_FILE); + test_unlink (TEST_SOURCE_FILE); + test_unlink (TEST_DELTA_FILE); + test_unlink (TEST_RECON_FILE); + test_unlink (TEST_RECON2_FILE); + test_unlink (TEST_COPY_FILE); + test_unlink (TEST_NOPERM_FILE); +#endif +} + +int test_setup (void) +{ + static int x = 0; + x++; + snprintf_func (TEST_TARGET_FILE, TESTFILESIZE, "/tmp/xdtest.target.%d", x); + snprintf_func (TEST_SOURCE_FILE, TESTFILESIZE, "/tmp/xdtest.source.%d", x); + snprintf_func (TEST_DELTA_FILE, TESTFILESIZE, "/tmp/xdtest.delta.%d", x); + snprintf_func (TEST_RECON_FILE, TESTFILESIZE, "/tmp/xdtest.recon.%d", x); + snprintf_func (TEST_RECON2_FILE, TESTFILESIZE, "/tmp/xdtest.recon2.%d", x); + snprintf_func (TEST_COPY_FILE, TESTFILESIZE, "/tmp/xdtest.copy.%d", x); + snprintf_func (TEST_NOPERM_FILE, TESTFILESIZE, "/tmp/xdtest.noperm.%d", x); + test_cleanup(); + return 0; +} + +static int +test_make_inputs (xd3_stream *stream, xoff_t *ss_out, xoff_t *ts_out) +{ + usize_t ts = (mt_random (&static_mtrand) % TEST_FILE_MEAN) + TEST_FILE_MEAN / 2; + usize_t ss = (mt_random (&static_mtrand) % TEST_FILE_MEAN) + TEST_FILE_MEAN / 2; + uint8_t *buf = (uint8_t*) malloc (ts + ss), *sbuf = buf, *tbuf = buf + ss; + usize_t sadd = 0, sadd_max = (usize_t)(ss * TEST_ADD_RATIO); + FILE *tf = NULL, *sf = NULL; + usize_t i, j; + int ret; + + if (buf == NULL) { return ENOMEM; } + + if ((tf = fopen (TEST_TARGET_FILE, "w")) == NULL || + (ss_out != NULL && (sf = fopen (TEST_SOURCE_FILE, "w")) == NULL)) + { + stream->msg = "write failed"; + ret = get_errno (); + goto failure; + } + + if (ss_out != NULL) + { + for (i = 0; i < ss; ) + { + sbuf[i++] = (uint8_t) mt_random (&static_mtrand); + } + } + + /* Then modify the data to produce copies, everything not copied is + * an add. The following logic produces the TEST_ADD_RATIO. The + * variable SADD contains the number of adds so far, which should + * not exceed SADD_MAX. */ + + /* XPR(NT "ss = %u ts = %u\n", ss, ts); */ + for (i = 0; i < ts; ) + { + usize_t left = ts - i; + usize_t next = mt_exp_rand ((uint32_t) TEST_ADD_MEAN, + (uint32_t) TEST_ADD_MAX); + usize_t add_left = sadd_max - sadd; + double add_prob = (left == 0) ? 0 : (add_left / (double) left); + int do_copy; + + next = min (left, next); + do_copy = (next > add_left || + (mt_random (&static_mtrand) / \ + (double)USIZE_T_MAX) >= add_prob); + + if (ss_out == NULL) + { + do_copy &= (i > 0); + } + else + { + do_copy &= (ss - next) > 0; + } + + if (do_copy) + { + /* Copy */ + size_t offset = mt_random (&static_mtrand) % ((ss_out == NULL) ? + i : + (ss - next)); + /* XPR(NT "[%u] copy %u at %u ", i, next, offset); */ + + for (j = 0; j < next; j += 1) + { + char c = ((ss_out == NULL) ? tbuf : sbuf)[offset + j]; + /* XPR(NT "%x%x", (c >> 4) & 0xf, c & 0xf); */ + tbuf[i++] = c; + } + /* XPR(NT "\n"); */ + } + else + { + /* Add */ + /* XPR(NT "[%u] add %u ", i, next); */ + for (j = 0; j < next; j += 1) + { + char c = (char) mt_random (&static_mtrand); + /* XPR(NT "%x%x", (c >> 4) & 0xf, c & 0xf); */ + tbuf[i++] = c; + } + /* XPR(NT "\n"); */ + sadd += next; + } + } + + /* XPR(NT "sadd = %u max = %u\n", sadd, sadd_max); */ + + if ((fwrite (tbuf, 1, ts, tf) != ts) || + (ss_out != NULL && (fwrite (sbuf, 1, ss, sf) != ss))) + { + stream->msg = "write failed"; + ret = get_errno (); + goto failure; + } + + if ((ret = fclose (tf)) || (ss_out != NULL && (ret = fclose (sf)))) + { + stream->msg = "close failed"; + ret = get_errno (); + goto failure; + } + + if (ts_out) { (*ts_out) = ts; } + if (ss_out) { (*ss_out) = ss; } + + failure: + free (buf); + return ret; +} + +int +test_compare_files (const char* tgt, const char *rec) +{ + FILE *orig, *recons; + static uint8_t obuf[TESTBUFSIZE], rbuf[TESTBUFSIZE]; + xoff_t offset = 0; + size_t i; + size_t oc, rc; + xoff_t diffs = 0; + + if ((orig = fopen (tgt, "r")) == NULL) + { + XPR(NT "open %s failed\n", tgt); + return get_errno (); + } + + if ((recons = fopen (rec, "r")) == NULL) + { + XPR(NT "open %s failed\n", rec); + return get_errno (); + } + + for (;;) + { + oc = fread (obuf, 1, TESTBUFSIZE, orig); + rc = fread (rbuf, 1, TESTBUFSIZE, recons); + + if (oc != rc) + { + return XD3_INTERNAL; + } + + if (oc == 0) + { + break; + } + + for (i = 0; i < oc; i += 1) + { + if (obuf[i] != rbuf[i]) + { + XPR(NT "byte %u (read %u @ %"Q"u) %d != %d\n", + (int)i, (int)oc, offset, obuf[i], rbuf[i]); + diffs++; + return XD3_INTERNAL; + } + } + + offset += oc; + } + + fclose (orig); + fclose (recons); + if (diffs != 0) + { + return XD3_INTERNAL; + } + return 0; +} + +static int +test_save_copy (const char *origname) +{ + char buf[TESTBUFSIZE]; + int ret; + + snprintf_func (buf, TESTBUFSIZE, "cp -f %s %s", origname, TEST_COPY_FILE); + + if ((ret = system (buf)) != 0) + { + return XD3_INTERNAL; + } + + return 0; +} + +static int +test_file_size (const char* file, xoff_t *size) +{ + struct stat sbuf; + int ret; + (*size) = 0; + + if (stat (file, & sbuf) < 0) + { + ret = get_errno (); + XPR(NT "stat failed: %s: %s\n", file, strerror (ret)); + return ret; + } + + if (! S_ISREG (sbuf.st_mode)) + { + ret = XD3_INTERNAL; + XPR(NT "not a regular file: %s: %s\n", file, strerror (ret)); + return ret; + } + + (*size) = sbuf.st_size; + return 0; +} + +/*********************************************************************** + READ OFFSET + ***********************************************************************/ + +/* Common test for read_integer errors: encodes a 64-bit value and + * then attempts to read as a 32-bit value. If TRUNC is non-zero, + * attempts to get errors by shortening the input, otherwise it should + * overflow. Expects XD3_INTERNAL and MSG. */ +static int +test_read_integer_error (xd3_stream *stream, usize_t trunto, const char *msg) +{ + uint64_t eval = 1ULL << 34; + uint32_t rval; + xd3_output *buf = NULL; + const uint8_t *max; + const uint8_t *inp; + int ret; + + buf = xd3_alloc_output (stream, buf); + + if ((ret = xd3_emit_uint64_t (stream, & buf, eval))) + { + goto fail; + } + + again: + + inp = buf->base; + max = buf->base + buf->next - trunto; + + if ((ret = xd3_read_uint32_t (stream, & inp, max, & rval)) != + XD3_INVALID_INPUT || + !MSG_IS (msg)) + { + ret = XD3_INTERNAL; + } + else if (trunto && trunto < buf->next) + { + trunto += 1; + goto again; + } + else + { + ret = 0; + } + + fail: + xd3_free_output (stream, buf); + return ret; +} + +/* Test integer overflow using the above routine. */ +static int +test_decode_integer_overflow (xd3_stream *stream, int unused) +{ + return test_read_integer_error (stream, 0, "overflow in read_intger"); +} + +/* Test integer EOI using the above routine. */ +static int +test_decode_integer_end_of_input (xd3_stream *stream, int unused) +{ + return test_read_integer_error (stream, 1, "end-of-input in read_integer"); +} + +/* Test that emit_integer/decode_integer/sizeof_integer/read_integer + * work on correct inputs. Tests powers of (2^7), plus or minus, up + * to the maximum value. */ +#define TEST_ENCODE_DECODE_INTEGER(TYPE,ONE,MAX) \ + xd3_output *rbuf = NULL; \ + xd3_output *dbuf = NULL; \ + TYPE values[64]; \ + usize_t nvalues = 0; \ + usize_t i; \ + int ret = 0; \ + \ + for (i = 0; i < (sizeof (TYPE) * 8); i += 7) \ + { \ + values[nvalues++] = (ONE << i) - ONE; \ + values[nvalues++] = (ONE << i); \ + values[nvalues++] = (ONE << i) + ONE; \ + } \ + \ + values[nvalues++] = MAX-ONE; \ + values[nvalues++] = MAX; \ + \ + rbuf = xd3_alloc_output (stream, rbuf); \ + dbuf = xd3_alloc_output (stream, dbuf); \ + \ + for (i = 0; i < nvalues; i += 1) \ + { \ + const uint8_t *max; \ + const uint8_t *inp; \ + TYPE val; \ + \ + DOT (); \ + rbuf->next = 0; \ + \ + if ((ret = xd3_emit_ ## TYPE (stream, & rbuf, values[i])) || \ + (ret = xd3_emit_ ## TYPE (stream, & dbuf, values[i]))) \ + { \ + goto fail; \ + } \ + \ + inp = rbuf->base; \ + max = rbuf->base + rbuf->next; \ + \ + if (rbuf->next != xd3_sizeof_ ## TYPE (values[i])) \ + { \ + ret = XD3_INTERNAL; \ + goto fail; \ + } \ + \ + if ((ret = xd3_read_ ## TYPE (stream, & inp, max, & val))) \ + { \ + goto fail; \ + } \ + \ + if (val != values[i]) \ + { \ + ret = XD3_INTERNAL; \ + goto fail; \ + } \ + \ + DOT (); \ + } \ + \ + stream->next_in = dbuf->base; \ + stream->avail_in = dbuf->next; \ + \ + for (i = 0; i < nvalues; i += 1) \ + { \ + TYPE val; \ + \ + if ((ret = xd3_decode_ ## TYPE (stream, & val))) \ + { \ + goto fail; \ + } \ + \ + if (val != values[i]) \ + { \ + ret = XD3_INTERNAL; \ + goto fail; \ + } \ + } \ + \ + if (stream->avail_in != 0) \ + { \ + ret = XD3_INTERNAL; \ + goto fail; \ + } \ + \ + fail: \ + xd3_free_output (stream, rbuf); \ + xd3_free_output (stream, dbuf); \ + \ + return ret + +static int +test_encode_decode_uint32_t (xd3_stream *stream, int unused) +{ + TEST_ENCODE_DECODE_INTEGER(uint32_t,1U,UINT32_MAX); +} + +static int +test_encode_decode_uint64_t (xd3_stream *stream, int unused) +{ + TEST_ENCODE_DECODE_INTEGER(uint64_t,1ULL,UINT64_MAX); +} + +static int +test_usize_t_overflow (xd3_stream *stream, int unused) +{ + if (USIZE_T_OVERFLOW (USIZE_T_MAX, 0)) { goto fail; } + if (USIZE_T_OVERFLOW (0, USIZE_T_MAX)) { goto fail; } + if (USIZE_T_OVERFLOW (USIZE_T_MAX / 2, USIZE_T_MAX / 2)) { goto fail; } + if (USIZE_T_OVERFLOW (USIZE_T_MAX / 2, USIZE_T_MAX / 2 + 1)) { goto fail; } + + if (! USIZE_T_OVERFLOW (USIZE_T_MAX, 1)) { goto fail; } + if (! USIZE_T_OVERFLOW (1, USIZE_T_MAX)) { goto fail; } + if (! USIZE_T_OVERFLOW (USIZE_T_MAX / 2 + 1, USIZE_T_MAX / 2 + 1)) { goto fail; } + + return 0; + + fail: + stream->msg = "incorrect overflow computation"; + return XD3_INTERNAL; +} + +static int +test_forward_match (xd3_stream *stream, int unused) +{ + usize_t i; + uint8_t buf1[256], buf2[256]; + + memset(buf1, 0, 256); + memset(buf2, 0, 256); + + for (i = 0; i < 256; i++) + { + CHECK(xd3_forward_match(buf1, buf2, i) == (int)i); + } + + for (i = 0; i < 255; i++) + { + buf2[i] = 1; + CHECK(xd3_forward_match(buf1, buf2, 256) == (int)i); + buf2[i] = 0; + } + + return 0; +} + +/*********************************************************************** + Address cache + ***********************************************************************/ + +static int +test_address_cache (xd3_stream *stream, int unused) +{ + int ret; + usize_t i; + usize_t offset; + usize_t *addrs; + uint8_t *big_buf, *buf_max; + const uint8_t *buf; + xd3_output *outp; + uint8_t *modes; + int mode_counts[16]; + + stream->acache.s_near = stream->code_table_desc->near_modes; + stream->acache.s_same = stream->code_table_desc->same_modes; + + if ((ret = xd3_encode_init_partial (stream))) { return ret; } + + addrs = (usize_t*) xd3_alloc (stream, sizeof (usize_t), ADDR_CACHE_ROUNDS); + modes = (uint8_t*) xd3_alloc (stream, sizeof (uint8_t), ADDR_CACHE_ROUNDS); + + memset (mode_counts, 0, sizeof (mode_counts)); + memset (modes, 0, ADDR_CACHE_ROUNDS); + + addrs[0] = 0; + + mt_init (& static_mtrand, 0x9f73f7fc); + + /* First pass: encode addresses */ + xd3_init_cache (& stream->acache); + + for (offset = 1; offset < ADDR_CACHE_ROUNDS; offset += 1) + { + double p; + usize_t addr; + usize_t prev_i; + usize_t nearby; + + p = (mt_random (&static_mtrand) / (double)USIZE_T_MAX); + prev_i = mt_random (&static_mtrand) % offset; + nearby = (mt_random (&static_mtrand) % 256) % offset; + nearby = max (1U, nearby); + + if (p < 0.1) { addr = addrs[offset-nearby]; } + else if (p < 0.4) { addr = min (addrs[prev_i] + nearby, offset-1); } + else { addr = prev_i; } + + if ((ret = xd3_encode_address (stream, addr, offset, & modes[offset]))) { return ret; } + + addrs[offset] = addr; + mode_counts[modes[offset]] += 1; + } + + /* Copy addresses into a contiguous buffer. */ + big_buf = (uint8_t*) xd3_alloc (stream, xd3_sizeof_output (ADDR_HEAD (stream)), 1); + + for (offset = 0, outp = ADDR_HEAD (stream); outp != NULL; offset += outp->next, outp = outp->next_page) + { + memcpy (big_buf + offset, outp->base, outp->next); + } + + buf_max = big_buf + offset; + buf = big_buf; + + /* Second pass: decode addresses */ + xd3_init_cache (& stream->acache); + + for (offset = 1; offset < ADDR_CACHE_ROUNDS; offset += 1) + { + uint32_t addr; + + if ((ret = xd3_decode_address (stream, offset, modes[offset], & buf, buf_max, & addr))) { return ret; } + + if (addr != addrs[offset]) + { + stream->msg = "incorrect decoded address"; + return XD3_INTERNAL; + } + } + + /* Check that every byte, mode was used. */ + if (buf != buf_max) + { + stream->msg = "address bytes not used"; + return XD3_INTERNAL; + } + + for (i = 0; i < (2 + stream->acache.s_same + stream->acache.s_near); i += 1) + { + if (mode_counts[i] == 0) + { + stream->msg = "address mode not used"; + return XD3_INTERNAL; + } + } + + xd3_free (stream, modes); + xd3_free (stream, addrs); + xd3_free (stream, big_buf); + + return 0; +} + +/*********************************************************************** + Encode and decode with single bit error + ***********************************************************************/ + +/* It compresses from 256 to around 185 bytes. + * Avoids matching addresses that are a single-bit difference. + * Avoids matching address 0. */ +static const uint8_t test_text[] = +"this is a story\n" +"abouttttttttttt\n" +"- his is a stor\n" +"- about nothing " +" all. boutique -" +"his story is a -" +"about " +"what happens all" +" the time what -" +"am I ttttttt the" +" person said, so" +" what, per son -" +" gory story is -" +" about nothing -" +"tttttt to test -" +"his sto nothing"; + +static const uint8_t test_apphead[] = "header test"; + +static int +test_compress_text (xd3_stream *stream, + uint8_t *encoded, + usize_t *encoded_size) +{ + int ret; + xd3_config cfg; + int oflags = stream->flags; + int flags = stream->flags | XD3_FLUSH; + + xd3_free_stream (stream); + xd3_init_config (& cfg, flags); + + /* This configuration is fixed so that the "expected non-error" the counts in + * decompress_single_bit_errors are too. See test_coftcfg_str. */ + cfg.smatch_cfg = XD3_SMATCH_SOFT; + cfg.smatcher_soft.name = "test"; + cfg.smatcher_soft.large_look = 64; /* no source, not used */ + cfg.smatcher_soft.large_step = 64; /* no source, not used */ + cfg.smatcher_soft.small_look = 4; + cfg.smatcher_soft.small_chain = 128; + cfg.smatcher_soft.small_lchain = 16; + cfg.smatcher_soft.max_lazy = 8; + cfg.smatcher_soft.long_enough = 128; + + xd3_config_stream (stream, & cfg); + + (*encoded_size) = 0; + + xd3_set_appheader (stream, test_apphead, + (usize_t) strlen ((char*) test_apphead)); + + if ((ret = xd3_encode_stream (stream, test_text, sizeof (test_text), + encoded, encoded_size, 4*sizeof (test_text)))) { goto fail; } + + if ((ret = xd3_close_stream (stream))) { goto fail; } + + fail: + xd3_free_stream (stream); + xd3_init_config (& cfg, oflags); + xd3_config_stream (stream, & cfg); + return ret; +} + +static int +test_decompress_text (xd3_stream *stream, uint8_t *enc, usize_t enc_size, usize_t test_desize) +{ + xd3_config cfg; + char decoded[sizeof (test_text)]; + uint8_t *apphead; + usize_t apphead_size; + usize_t decoded_size; + const char *msg; + int ret; + usize_t pos = 0; + int flags = stream->flags; + usize_t take; + + input: + /* Test decoding test_desize input bytes at a time */ + take = min (enc_size - pos, test_desize); + CHECK(take > 0); + + xd3_avail_input (stream, enc + pos, take); + again: + ret = xd3_decode_input (stream); + + pos += take; + take = 0; + + switch (ret) + { + case XD3_OUTPUT: + break; + case XD3_WINSTART: + case XD3_GOTHEADER: + goto again; + case XD3_INPUT: + if (pos < enc_size) { goto input; } + /* else fallthrough */ + case XD3_WINFINISH: + default: + goto fail; + } + + CHECK(ret == XD3_OUTPUT); + CHECK(pos == enc_size); + + if (stream->avail_out != sizeof (test_text)) + { + stream->msg = "incorrect output size"; + ret = XD3_INTERNAL; + goto fail; + } + + decoded_size = stream->avail_out; + memcpy (decoded, stream->next_out, stream->avail_out); + + xd3_consume_output (stream); + + if ((ret = xd3_get_appheader (stream, & apphead, & apphead_size))) { goto fail; } + + if (apphead_size != strlen ((char*) test_apphead) || + memcmp (apphead, test_apphead, strlen ((char*) test_apphead)) != 0) + { + stream->msg = "incorrect appheader"; + ret = XD3_INTERNAL; + goto fail; + } + + if ((ret = xd3_decode_input (stream)) != XD3_WINFINISH || + (ret = xd3_close_stream (stream)) != 0) + { + goto fail; + } + + if (decoded_size != sizeof (test_text) || + memcmp (decoded, test_text, sizeof (test_text)) != 0) + { + stream->msg = "incorrect output text"; + ret = EIO; + } + + fail: + msg = stream->msg; + xd3_free_stream (stream); + xd3_init_config (& cfg, flags); + xd3_config_stream (stream, & cfg); + stream->msg = msg; + + return ret; +} + +static int +test_decompress_single_bit_error (xd3_stream *stream, int expected_non_failures) +{ + int ret; + usize_t i; + uint8_t encoded[4*sizeof (test_text)]; /* make room for alt code table */ + usize_t encoded_size; + int non_failures = 0; + int cksum = (stream->flags & XD3_ADLER32) != 0; + +//#define DEBUG_TEST_FAILURES +#ifndef DEBUG_TEST_FAILURES +#define TEST_FAILURES() +#else + /* For checking non-failure cases by hand, enable this macro and run + * xdelta printdelta with print_cpymode disabled. Every non-failure + * should change a copy address mode, which doesn't cause a failure + * because the address cache starts out with all zeros. + + ./xdelta3 test + for i in test_text.xz.*; do ./xdelta3 printdelta $i > $i.out; + diff $i.out test_text.xz.0.out; done + + */ + system ("rm -rf test_text.*"); + { + char buf[TESTBUFSIZE]; + FILE *f; + snprintf_func (buf, TESTBUFSIZE, "test_text"); + f = fopen (buf, "w"); + fwrite (test_text,1,sizeof (test_text),f); + fclose (f); + } +#define TEST_FAILURES() \ + do { \ + char buf[TESTBUFSIZE]; \ + FILE *f; \ + snprintf_func (buf, TESTBUFSIZE, "test_text.xz.%d", non_failures); \ + f = fopen (buf, "w"); \ + fwrite (encoded,1,encoded_size,f); \ + fclose (f); \ + } while (0) +#endif + + stream->sec_data.inefficient = 1; + stream->sec_inst.inefficient = 1; + stream->sec_addr.inefficient = 1; + + /* Encode text, test correct input */ + if ((ret = test_compress_text (stream, encoded, & encoded_size))) + { + /*stream->msg = "without error: encode failure";*/ + return ret; + } + + if ((ret = test_decompress_text (stream, encoded, encoded_size, + sizeof (test_text) / 4))) + { + /*stream->msg = "without error: decode failure";*/ + return ret; + } + + TEST_FAILURES(); + + for (i = 0; i < encoded_size*8; i += 1) + { + /* Single bit error. */ + encoded[i/8] ^= 1 << (i%8); + + if ((ret = test_decompress_text (stream, encoded, + encoded_size, sizeof (test_text))) == 0) + { + non_failures += 1; +#ifdef DEBUG_TEST_FAILURES + XPR(NT "%u[%u] non-failure %u\n", i/8, i%8, non_failures); +#endif + TEST_FAILURES(); + } + else + { + /*XPR(NT "%u[%u] failure: %s\n", i/8, i%8, stream->msg);*/ + } + + /* decompress_text returns EIO when the final memcmp() fails, but that + * should never happen with checksumming on. */ + if (cksum && ret == EIO) + { + /*XPR(NT "%u[%u] cksum mismatch\n", i/8, i%8);*/ + stream->msg = "checksum mismatch"; + return XD3_INTERNAL; + } + + /* Undo single bit error. */ + encoded[i/8] ^= 1 << (i%8); + } + + /* Test correct input again */ + if ((ret = test_decompress_text (stream, encoded, encoded_size, 1))) + { + /*stream->msg = "without error: decode failure";*/ + return ret; + } + + /* Check expected non-failures */ + if (non_failures != expected_non_failures) + { + XPR(NT "non-failures %u; expected %u", + non_failures, expected_non_failures); + stream->msg = "incorrect"; + return XD3_INTERNAL; + } + + DOT (); + + return 0; +} + +/*********************************************************************** + Secondary compression tests + ***********************************************************************/ + +#if SECONDARY_ANY +typedef int (*sec_dist_func) (xd3_stream *stream, xd3_output *data); + +static int sec_dist_func1 (xd3_stream *stream, xd3_output *data); +static int sec_dist_func2 (xd3_stream *stream, xd3_output *data); +static int sec_dist_func3 (xd3_stream *stream, xd3_output *data); +static int sec_dist_func4 (xd3_stream *stream, xd3_output *data); +static int sec_dist_func5 (xd3_stream *stream, xd3_output *data); +static int sec_dist_func6 (xd3_stream *stream, xd3_output *data); +static int sec_dist_func7 (xd3_stream *stream, xd3_output *data); +static int sec_dist_func8 (xd3_stream *stream, xd3_output *data); +static int sec_dist_func9 (xd3_stream *stream, xd3_output *data); +static int sec_dist_func10 (xd3_stream *stream, xd3_output *data); +static int sec_dist_func11 (xd3_stream *stream, xd3_output *data); + +static sec_dist_func sec_dists[] = +{ + sec_dist_func1, + sec_dist_func2, + sec_dist_func3, + sec_dist_func4, + sec_dist_func5, + sec_dist_func6, + sec_dist_func7, + sec_dist_func8, + sec_dist_func9, + sec_dist_func10, + sec_dist_func11, +}; + +/* Test ditsribution: 100 bytes of the same character (13). */ +static int +sec_dist_func1 (xd3_stream *stream, xd3_output *data) +{ + int i, ret; + for (i = 0; i < 100; i += 1) + { + if ((ret = xd3_emit_byte (stream, & data, 13))) { return ret; } + } + return 0; +} + +/* Test ditsribution: uniform covering half the alphabet. */ +static int +sec_dist_func2 (xd3_stream *stream, xd3_output *data) +{ + int i, ret; + for (i = 0; i < ALPHABET_SIZE; i += 1) + { + if ((ret = xd3_emit_byte (stream, & data, i%(ALPHABET_SIZE/2)))) { return ret; } + } + return 0; +} + +/* Test ditsribution: uniform covering the entire alphabet. */ +static int +sec_dist_func3 (xd3_stream *stream, xd3_output *data) +{ + int i, ret; + for (i = 0; i < ALPHABET_SIZE; i += 1) + { + if ((ret = xd3_emit_byte (stream, & data, i%ALPHABET_SIZE))) { return ret; } + } + return 0; +} + +/* Test distribution: An exponential distribution covering half the alphabet */ +static int +sec_dist_func4 (xd3_stream *stream, xd3_output *data) +{ + int i, ret, x; + for (i = 0; i < ALPHABET_SIZE*20; i += 1) + { + x = mt_exp_rand (10, ALPHABET_SIZE/2); + if ((ret = xd3_emit_byte (stream, & data, x))) { return ret; } + } + return 0; +} + +/* Test distribution: An exponential distribution covering the entire alphabet */ +static int +sec_dist_func5 (xd3_stream *stream, xd3_output *data) +{ + int i, ret, x; + for (i = 0; i < ALPHABET_SIZE*20; i += 1) + { + x = mt_exp_rand (10, ALPHABET_SIZE-1); + if ((ret = xd3_emit_byte (stream, & data, x))) { return ret; } + } + return 0; +} + +/* Test distribution: An uniform random distribution covering half the alphabet */ +static int +sec_dist_func6 (xd3_stream *stream, xd3_output *data) +{ + int i, ret, x; + for (i = 0; i < ALPHABET_SIZE*20; i += 1) + { + x = mt_random (&static_mtrand) % (ALPHABET_SIZE/2); + if ((ret = xd3_emit_byte (stream, & data, x))) { return ret; } + } + return 0; +} + +/* Test distribution: An uniform random distribution covering the entire alphabet */ +static int +sec_dist_func7 (xd3_stream *stream, xd3_output *data) +{ + int i, ret, x; + for (i = 0; i < ALPHABET_SIZE*200; i += 1) + { + x = mt_random (&static_mtrand) % ALPHABET_SIZE; + if ((ret = xd3_emit_byte (stream, & data, x))) { return ret; } + } + return 0; +} + +/* Test distribution: A small number of frequent characters, difficult + * to divide into many groups */ +static int +sec_dist_func8 (xd3_stream *stream, xd3_output *data) +{ + int i, ret; + for (i = 0; i < ALPHABET_SIZE*5; i += 1) + { + if ((ret = xd3_emit_byte (stream, & data, 0))) { return ret; } + if ((ret = xd3_emit_byte (stream, & data, 64))) { return ret; } + if ((ret = xd3_emit_byte (stream, & data, 128))) { return ret; } + if ((ret = xd3_emit_byte (stream, & data, 255))) { return ret; } + } + return 0; +} + +/* Test distribution: One that causes many FGK block promotions (found a bug) */ +static int +sec_dist_func9 (xd3_stream *stream, xd3_output *data) +{ + int i, ret; + + int ramp = 0; + int rcount = 0; + int prom = 0; + int pcount = 0; + + /* 200 was long enough to trigger it--only when stricter checking + * that counted all blocks was turned on, but it seems I deleted + * this code. (missing fgk_free_block on line 398). */ + for (i = 0; i < ALPHABET_SIZE*200; i += 1) + { + repeat: + if (ramp < ALPHABET_SIZE) + { + /* Initially Nth symbol has (N+1) frequency */ + if (rcount <= ramp) + { + rcount += 1; + if ((ret = xd3_emit_byte (stream, & data, ramp))) { return ret; } + continue; + } + + ramp += 1; + rcount = 0; + goto repeat; + } + + /* Thereafter, promote least freq to max freq */ + if (pcount == ALPHABET_SIZE) + { + pcount = 0; + prom = (prom + 1) % ALPHABET_SIZE; + } + + pcount += 1; + if ((ret = xd3_emit_byte (stream, & data, prom))) { return ret; } + } + + return 0; +} + +/* Test distribution: freq[i] == i*i, creates a 21-bit code length, fixed in 3.0r. */ +static int +sec_dist_func10 (xd3_stream *stream, xd3_output *data) +{ + int i, j, ret; + for (i = 0; i < ALPHABET_SIZE; i += 1) + { + for (j = 0; j <= (i*i); j += 1) + { + if ((ret = xd3_emit_byte (stream, & data, i))) { return ret; } + } + } + return 0; +} + +/* Test distribution: fibonacci */ +static int +sec_dist_func11 (xd3_stream *stream, xd3_output *data) +{ + int sum0 = 0; + int sum1 = 1; + int i, j, ret; + for (i = 0; i < 33; ++i) + { + for (j = 0; j < (sum0 + sum1); ++j) + { + if ((ret = xd3_emit_byte (stream, & data, i))) { return ret; } + } + sum0 = sum1; + sum1 = j; + } + return 0; +} + +static int +test_secondary_decode (xd3_stream *stream, + const xd3_sec_type *sec, + usize_t input_size, + usize_t compress_size, + const uint8_t *dec_input, + const uint8_t *dec_correct, + uint8_t *dec_output) +{ + int ret; + xd3_sec_stream *dec_stream; + const uint8_t *dec_input_used, *dec_input_end; + uint8_t *dec_output_used, *dec_output_end; + + if ((dec_stream = sec->alloc (stream)) == NULL) { return ENOMEM; } + + if ((ret = sec->init (stream, dec_stream, 0)) != 0) { goto fail; } + + dec_input_used = dec_input; + dec_input_end = dec_input + compress_size; + + dec_output_used = dec_output; + dec_output_end = dec_output + input_size; + + if ((ret = sec->decode (stream, dec_stream, + & dec_input_used, dec_input_end, + & dec_output_used, dec_output_end))) + { + goto fail; + } + + if (dec_input_used != dec_input_end) + { + stream->msg = "unused input"; + ret = XD3_INTERNAL; + goto fail; + } + + if (dec_output_used != dec_output_end) + { + stream->msg = "unfinished output"; + ret = XD3_INTERNAL; + goto fail; + } + + if (memcmp (dec_output, dec_correct, input_size) != 0) + { + stream->msg = "incorrect output"; + ret = XD3_INTERNAL; + goto fail; + } + + fail: + sec->destroy (stream, dec_stream); + return ret; +} + +static int +test_secondary (xd3_stream *stream, const xd3_sec_type *sec, usize_t groups) +{ + usize_t test_i; + int ret; + xd3_output *in_head, *out_head, *p; + usize_t p_off, input_size, compress_size; + uint8_t *dec_input = NULL, *dec_output = NULL, *dec_correct = NULL; + xd3_sec_stream *enc_stream; + xd3_sec_cfg cfg; + + memset (& cfg, 0, sizeof (cfg)); + + cfg.inefficient = 1; + + for (cfg.ngroups = 1; cfg.ngroups <= groups; cfg.ngroups += 1) + { + XPR(NTR "\n..."); + for (test_i = 0; test_i < SIZEOF_ARRAY (sec_dists); test_i += 1) + { + mt_init (& static_mtrand, 0x9f73f7fc); + + in_head = xd3_alloc_output (stream, NULL); + out_head = xd3_alloc_output (stream, NULL); + enc_stream = sec->alloc (stream); + dec_input = NULL; + dec_output = NULL; + dec_correct = NULL; + + if (in_head == NULL || out_head == NULL || enc_stream == NULL) + { + goto nomem; + } + + if ((ret = sec_dists[test_i] (stream, in_head))) { goto fail; } + + if ((ret = sec->init (stream, enc_stream, 1)) != 0) { goto fail; } + + /* Encode data */ + if ((ret = sec->encode (stream, enc_stream, + in_head, out_head, & cfg))) + { + XPR(NT "test %u: encode: %s", test_i, stream->msg); + goto fail; + } + + /* Calculate sizes, allocate contiguous arrays for decoding */ + input_size = xd3_sizeof_output (in_head); + compress_size = xd3_sizeof_output (out_head); + + XPR(NTR "%.3f", 8.0 * (double) compress_size / (double) input_size); + + if ((dec_input = (uint8_t*) xd3_alloc (stream, compress_size, 1)) == NULL || + (dec_output = (uint8_t*) xd3_alloc (stream, input_size, 1)) == NULL || + (dec_correct = (uint8_t*) xd3_alloc (stream, input_size, 1)) == NULL) + { + goto nomem; + } + + /* Fill the compressed data array */ + for (p_off = 0, p = out_head; p != NULL; + p_off += p->next, p = p->next_page) + { + memcpy (dec_input + p_off, p->base, p->next); + } + + CHECK(p_off == compress_size); + + /* Fill the input data array */ + for (p_off = 0, p = in_head; p != NULL; + p_off += p->next, p = p->next_page) + { + memcpy (dec_correct + p_off, p->base, p->next); + } + + CHECK(p_off == input_size); + + if ((ret = test_secondary_decode (stream, sec, input_size, + compress_size, dec_input, + dec_correct, dec_output))) + { + XPR(NT "test %u: decode: %s", test_i, stream->msg); + goto fail; + } + + /* Single-bit error test, only cover the first 10 bytes. + * Some non-failures are expected in the Huffman case: + * Changing the clclen array, for example, may not harm the + * decoding. Really looking for faults here. */ + { + int i; + int bytes = min (compress_size, 10U); + for (i = 0; i < bytes * 8; i += 1) + { + dec_input[i/8] ^= 1 << (i%8); + + if ((ret = test_secondary_decode (stream, sec, input_size, + compress_size, dec_input, + dec_correct, dec_output)) + == 0) + { + /*XPR(NT "test %u: decode single-bit [%u/%u] + error non-failure", test_i, i/8, i%8);*/ + } + + dec_input[i/8] ^= 1 << (i%8); + + if ((i % (2*bytes)) == (2*bytes)-1) + { + DOT (); + } + } + ret = 0; + } + + if (0) { nomem: ret = ENOMEM; } + + fail: + sec->destroy (stream, enc_stream); + xd3_free_output (stream, in_head); + xd3_free_output (stream, out_head); + xd3_free (stream, dec_input); + xd3_free (stream, dec_output); + xd3_free (stream, dec_correct); + + if (ret != 0) { return ret; } + } + } + + return 0; +} + +IF_FGK (static int test_secondary_fgk (xd3_stream *stream, usize_t gp) + { return test_secondary (stream, & fgk_sec_type, gp); }) +IF_DJW (static int test_secondary_huff (xd3_stream *stream, usize_t gp) + { return test_secondary (stream, & djw_sec_type, gp); }) +IF_LZMA (static int test_secondary_lzma (xd3_stream *stream, usize_t gp) + { return test_secondary (stream, & lzma_sec_type, gp); }) +#endif + +/*********************************************************************** + TEST INSTRUCTION TABLE + ***********************************************************************/ + +/* Test that xd3_choose_instruction() does the right thing for its code + * table. */ +static int +test_choose_instruction (xd3_stream *stream, int ignore) +{ + int i; + + stream->code_table = (*stream->code_table_func) (); + + for (i = 0; i < 256; i += 1) + { + const xd3_dinst *d = stream->code_table + i; + xd3_rinst prev, inst; + + CHECK(d->type1 > 0); + + memset (& prev, 0, sizeof (prev)); + memset (& inst, 0, sizeof (inst)); + + if (d->type2 == 0) + { + inst.type = d->type1; + + if ((inst.size = d->size1) == 0) + { + inst.size = TESTBUFSIZE; + } + + XD3_CHOOSE_INSTRUCTION (stream, NULL, & inst); + + if (inst.code2 != 0 || inst.code1 != i) + { + stream->msg = "wrong single instruction"; + return XD3_INTERNAL; + } + } + else + { + prev.type = d->type1; + prev.size = d->size1; + inst.type = d->type2; + inst.size = d->size2; + + XD3_CHOOSE_INSTRUCTION (stream, & prev, & inst); + + if (prev.code2 != i) + { + stream->msg = "wrong double instruction"; + return XD3_INTERNAL; + } + } + } + + return 0; +} + +/*********************************************************************** + TEST INSTRUCTION TABLE CODING + ***********************************************************************/ + +#if GENERIC_ENCODE_TABLES +/* Test that encoding and decoding a code table works */ +static int +test_encode_code_table (xd3_stream *stream, int ignore) +{ + int ret; + const uint8_t *comp_data; + usize_t comp_size; + + if ((ret = xd3_compute_alternate_table_encoding (stream, & comp_data, & comp_size))) + { + return ret; + } + + stream->acache.s_near = __alternate_code_table_desc.near_modes; + stream->acache.s_same = __alternate_code_table_desc.same_modes; + + if ((ret = xd3_apply_table_encoding (stream, comp_data, comp_size))) + { + return ret; + } + + if (memcmp (stream->code_table, xd3_alternate_code_table (), sizeof (xd3_dinst) * 256) != 0) + { + stream->msg = "wrong code table reconstruction"; + return XD3_INTERNAL; + } + + return 0; +} +#endif + +/*********************************************************************** + 64BIT STREAMING + ***********************************************************************/ + +/* This test encodes and decodes a series of 1 megabyte windows, each + * containing a long run of zeros along with a single xoff_t size + * record to indicate the sequence. */ +static int +test_streaming (xd3_stream *in_stream, uint8_t *encbuf, uint8_t *decbuf, uint8_t *delbuf, usize_t megs) +{ + xd3_stream estream, dstream; + int ret; + usize_t i, delsize, decsize; + xd3_config cfg; + xd3_init_config (& cfg, in_stream->flags); + cfg.flags |= XD3_COMPLEVEL_6; + + if ((ret = xd3_config_stream (& estream, & cfg)) || + (ret = xd3_config_stream (& dstream, & cfg))) + { + goto fail; + } + + for (i = 0; i < megs; i += 1) + { + ((usize_t*) encbuf)[0] = i; + + if ((i % 200) == 199) { DOT (); } + + if ((ret = xd3_process_stream (1, & estream, xd3_encode_input, 0, + encbuf, 1 << 20, + delbuf, & delsize, 1 << 20))) + { + in_stream->msg = estream.msg; + goto fail; + } + + if ((ret = xd3_process_stream (0, & dstream, xd3_decode_input, 0, + delbuf, delsize, + decbuf, & decsize, 1 << 20))) + { + in_stream->msg = dstream.msg; + goto fail; + } + + if (decsize != 1 << 20 || + memcmp (encbuf, decbuf, 1 << 20) != 0) + { + in_stream->msg = "wrong result"; + ret = XD3_INTERNAL; + goto fail; + } + } + + if ((ret = xd3_close_stream (& estream)) || + (ret = xd3_close_stream (& dstream))) + { + goto fail; + } + + fail: + xd3_free_stream (& estream); + xd3_free_stream (& dstream); + return ret; +} + +/* Run tests of data streaming of over and around 4GB of data. */ +static int +test_compressed_stream_overflow (xd3_stream *stream, int ignore) +{ + int ret; + int i; + uint8_t *buf; + + if ((buf = (uint8_t*) malloc (TWO_MEGS_AND_DELTA)) == NULL) { return ENOMEM; } + + memset (buf, 0, TWO_MEGS_AND_DELTA); + for (i = 0; i < (2 << 20); i += 256) + { + int j; + int off = mt_random(& static_mtrand) % 10; + for (j = 0; j < 256; j++) + { + buf[i + j] = j + off; + } + } + + /* Test overflow of a 32-bit file offset. */ + if (SIZEOF_XOFF_T == 4) + { + ret = test_streaming (stream, buf, buf + (1 << 20), buf + (2 << 20), (1 << 12) + 1); + + if (ret == XD3_INVALID_INPUT && MSG_IS ("decoder file offset overflow")) + { + ret = 0; + } + else + { + XPR(NT XD3_LIB_ERRMSG (stream, ret)); + stream->msg = "expected overflow condition"; + ret = XD3_INTERNAL; + goto fail; + } + } + + /* Test transfer of exactly 32bits worth of data. */ + if ((ret = test_streaming (stream, + buf, + buf + (1 << 20), + buf + (2 << 20), + 1 << 12))) + { + goto fail; + } + fail: + free (buf); + return ret; +} + +/*********************************************************************** + COMMAND LINE + ***********************************************************************/ + +#if SHELL_TESTS + +/* For each pair of command templates in the array below, test that + * encoding and decoding commands work. Also check for the expected + * size delta, which should be approximately TEST_ADD_RATIO times the + * file size created by test_make_inputs. Due to differences in the + * application header, it is suppressed (-A) so that all delta files + * are the same. */ +static int +test_command_line_arguments (xd3_stream *stream, int ignore) +{ + int i, ret; + + static const char* cmdpairs[] = + { + /* standard input, output */ + "%s %s -A < %s > %s", "%s -d < %s > %s", + "%s %s -A -e < %s > %s", "%s -d < %s > %s", + "%s %s -A= encode < %s > %s", "%s decode < %s > %s", + "%s %s -A -q encode < %s > %s", "%s -qdq < %s > %s", + + /* file input, standard output */ + "%s %s -A= %s > %s", "%s -d %s > %s", + "%s %s -A -e %s > %s", "%s -d %s > %s", + "%s %s encode -A= %s > %s", "%s decode %s > %s", + + /* file input, output */ + "%s %s -A= %s %s", "%s -d %s %s", + "%s %s -A -e %s %s", "%s -d %s %s", + "%s %s -A= encode %s %s", "%s decode %s %s", + + /* option placement */ + "%s %s -A -f %s %s", "%s -f -d %s %s", + "%s %s -e -A= %s %s", "%s -d -f %s %s", + "%s %s -f encode -A= %s %s", "%s -f decode -f %s %s", + }; + + char ecmd[TESTBUFSIZE], dcmd[TESTBUFSIZE]; + int pairs = SIZEOF_ARRAY (cmdpairs) / 2; + xoff_t tsize; + xoff_t dsize; + double ratio; + + mt_init (& static_mtrand, 0x9f73f7fc); + + for (i = 0; i < pairs; i += 1) + { + test_setup (); + if ((ret = test_make_inputs (stream, NULL, & tsize))) { return ret; } + + snprintf_func (ecmd, TESTBUFSIZE, cmdpairs[2*i], program_name, + test_softcfg_str, TEST_TARGET_FILE, TEST_DELTA_FILE); + snprintf_func (dcmd, TESTBUFSIZE, cmdpairs[2*i+1], program_name, + TEST_DELTA_FILE, TEST_RECON_FILE); + + /* Encode and decode. */ + if ((ret = system (ecmd)) != 0) + { + XPR(NT "encode command: %s\n", ecmd); + stream->msg = "encode cmd failed"; + return XD3_INTERNAL; + } + + if ((ret = system (dcmd)) != 0) + { + XPR(NT "decode command: %s\n", dcmd); + stream->msg = "decode cmd failed"; + return XD3_INTERNAL; + } + + /* Compare the target file. */ + if ((ret = test_compare_files (TEST_TARGET_FILE, TEST_RECON_FILE))) + { + return ret; + } + + if ((ret = test_file_size (TEST_DELTA_FILE, & dsize))) + { + return ret; + } + + ratio = (double) dsize / (double) tsize; + + /* Check that it is not too small, not too large. */ + if (ratio >= TEST_ADD_RATIO + TEST_EPSILON) + { + XPR(NT "test encode with size ratio %.4f, " + "expected < %.4f (%"Q"u, %"Q"u)\n", + ratio, TEST_ADD_RATIO + TEST_EPSILON, dsize, tsize); + stream->msg = "strange encoding"; + return XD3_INTERNAL; + } + + if (ratio <= TEST_ADD_RATIO * (1.0 - 2 * TEST_EPSILON)) + { + XPR(NT "test encode with size ratio %.4f, " + "expected > %.4f\n", + ratio, TEST_ADD_RATIO - TEST_EPSILON); + stream->msg = "strange encoding"; + return XD3_INTERNAL; + } + + /* Also check that test_compare_files works. The delta and original should + * not be identical. */ + if ((ret = test_compare_files (TEST_DELTA_FILE, + TEST_TARGET_FILE)) == 0) + { + stream->msg = "broken test_compare_files"; + return XD3_INTERNAL; + } + + test_cleanup (); + DOT (); + } + + return 0; +} + +static int +check_vcdiff_header (xd3_stream *stream, + const char *input, + const char *line_start, + const char *matches, + int yes_or_no) +{ + int ret; + char vcmd[TESTBUFSIZE], gcmd[TESTBUFSIZE]; + + snprintf_func (vcmd, TESTBUFSIZE, "%s printhdr -f %s %s", + program_name, input, TEST_RECON2_FILE); + + if ((ret = system (vcmd)) != 0) + { + XPR(NT "printhdr command: %s\n", vcmd); + stream->msg = "printhdr cmd failed"; + return XD3_INTERNAL; + } + + snprintf_func (gcmd, TESTBUFSIZE, "grep \"%s.*%s.*\" %s > /dev/null", + line_start, matches, TEST_RECON2_FILE); + + if (yes_or_no) + { + if ((ret = do_cmd (stream, gcmd))) + { + XPR(NT "%s\n", gcmd); + return ret; + } + } + else + { + if ((ret = do_fail (stream, gcmd))) + { + XPR(NT "%s\n", gcmd); + return ret; + } + } + + return 0; +} + +static int +test_recode_command2 (xd3_stream *stream, int has_source, + int variant, int change) +{ + int has_adler32 = (variant & 0x1) != 0; + int has_apphead = (variant & 0x2) != 0; + int has_secondary = (variant & 0x4) != 0; + + int change_adler32 = (change & 0x1) != 0; + int change_apphead = (change & 0x2) != 0; + int change_secondary = (change & 0x4) != 0; + + int recoded_adler32 = change_adler32 ? !has_adler32 : has_adler32; + int recoded_apphead = change_apphead ? !has_apphead : has_apphead; + int recoded_secondary = change_secondary ? !has_secondary : has_secondary; + + char ecmd[TESTBUFSIZE], recmd[TESTBUFSIZE], dcmd[TESTBUFSIZE]; + xoff_t tsize, ssize; + int ret; + + test_setup (); + + if ((ret = test_make_inputs (stream, has_source ? & ssize : NULL, & tsize))) + { + return ret; + } + + /* First encode */ + snprintf_func (ecmd, TESTBUFSIZE, "%s %s -f %s %s %s %s %s %s %s", + program_name, test_softcfg_str, + has_adler32 ? "" : "-n ", + has_apphead ? "-A=encode_apphead " : "-A= ", + has_secondary ? "-S djw " : "-S none ", + has_source ? "-s " : "", + has_source ? TEST_SOURCE_FILE : "", + TEST_TARGET_FILE, + TEST_DELTA_FILE); + + if ((ret = system (ecmd)) != 0) + { + XPR(NT "encode command: %s\n", ecmd); + stream->msg = "encode cmd failed"; + return XD3_INTERNAL; + } + + /* Now recode */ + snprintf_func (recmd, TESTBUFSIZE, + "%s recode %s -f %s %s %s %s %s", program_name, test_softcfg_str, + recoded_adler32 ? "" : "-n ", + !change_apphead ? "" : + (recoded_apphead ? "-A=recode_apphead " : "-A= "), + recoded_secondary ? "-S djw " : "-S none ", + TEST_DELTA_FILE, + TEST_COPY_FILE); + + if ((ret = system (recmd)) != 0) + { + XPR(NT "recode command: %s\n", recmd); + stream->msg = "recode cmd failed"; + return XD3_INTERNAL; + } + + /* Check recode changes. */ + + if ((ret = check_vcdiff_header (stream, + TEST_COPY_FILE, + "VCDIFF window indicator", + "VCD_SOURCE", + has_source))) { return ret; } + + if ((ret = check_vcdiff_header (stream, + TEST_COPY_FILE, + "VCDIFF header indicator", + "VCD_SECONDARY", + recoded_secondary))) { return ret; } + + if ((ret = check_vcdiff_header (stream, + TEST_COPY_FILE, + "VCDIFF window indicator", + "VCD_ADLER32", + /* Recode can't generate an adler32 + * checksum, it can only preserve it or + * remove it. */ + has_adler32 && recoded_adler32))) + { + return ret; + } + + if (!change_apphead) + { + if ((ret = check_vcdiff_header (stream, + TEST_COPY_FILE, + "VCDIFF header indicator", + "VCD_APPHEADER", + has_apphead))) + { + return ret; + } + if ((ret = check_vcdiff_header (stream, + TEST_COPY_FILE, + "VCDIFF application header", + "encode_apphead", + has_apphead))) + { + return ret; + } + } + else + { + if ((ret = check_vcdiff_header (stream, + TEST_COPY_FILE, + "VCDIFF header indicator", + "VCD_APPHEADER", + recoded_apphead))) + { + return ret; + } + if (recoded_apphead && + (ret = check_vcdiff_header (stream, + TEST_COPY_FILE, + "VCDIFF application header", + "recode_apphead", + 1))) + { + return ret; + } + } + + /* Now decode */ + snprintf_func (dcmd, TESTBUFSIZE, "%s -fd %s %s %s %s ", program_name, + has_source ? "-s " : "", + has_source ? TEST_SOURCE_FILE : "", + TEST_COPY_FILE, + TEST_RECON_FILE); + + if ((ret = system (dcmd)) != 0) + { + XPR(NT "decode command: %s\n", dcmd); + stream->msg = "decode cmd failed"; + return XD3_INTERNAL; + } + + /* Now compare. */ + if ((ret = test_compare_files (TEST_TARGET_FILE, TEST_RECON_FILE))) + { + return ret; + } + + return 0; +} + +static int +test_recode_command (xd3_stream *stream, int ignore) +{ + /* Things to test: + * - with and without a source file (recode does not change) + * + * (recode may or may not change -- 8 variations) + * - with and without adler32 + * - with and without app header + * - with and without secondary + */ + int has_source; + int variant; + int change; + int ret; + + for (has_source = 0; has_source < 2; has_source++) + { + for (variant = 0; variant < 8; variant++) + { + for (change = 0; change < 8; change++) + { + if ((ret = test_recode_command2 (stream, has_source, + variant, change))) + { + return ret; + } + } + DOT (); + } + } + + return 0; +} +#endif + +/*********************************************************************** + EXTERNAL I/O DECOMPRESSION/RECOMPRESSION + ***********************************************************************/ + +#if EXTERNAL_COMPRESSION +/* This performs one step of the test_externally_compressed_io + * function described below. It builds a pipe containing both Xdelta + * and external compression/decompression that should not modify the + * data passing through. */ +static int +test_compressed_pipe (xd3_stream *stream, main_extcomp *ext, char* buf, + const char* comp_options, const char* decomp_options, + int do_ext_recomp, const char* msg) +{ + int ret; + char decomp_buf[TESTBUFSIZE]; + + if (do_ext_recomp) + { + snprintf_func (decomp_buf, TESTBUFSIZE, + " | %s %s", ext->decomp_cmdname, ext->decomp_options); + } + else + { + decomp_buf[0] = 0; + } + + snprintf_func (buf, TESTBUFSIZE, "%s %s < %s | %s %s | %s %s%s > %s", + ext->recomp_cmdname, ext->recomp_options, + TEST_TARGET_FILE, + program_name, comp_options, + program_name, decomp_options, + decomp_buf, + TEST_RECON_FILE); + + if ((ret = system (buf)) != 0) + { + stream->msg = msg; + return XD3_INTERNAL; + } + + if ((ret = test_compare_files (TEST_TARGET_FILE, TEST_RECON_FILE))) + { + return XD3_INTERNAL; + } + + DOT (); + return 0; +} + +/* We want to test that a pipe such as: + * + * --> | gzip -cf | xdelta3 -cf | xdelta3 -dcf | gzip -dcf | --> + * + * is transparent, i.e., does not modify the stream of data. However, + * we also want to verify that at the center the data is properly + * compressed, i.e., that we do not just have a re-compressed gzip + * format, that we have an VCDIFF format. We do this in two steps. + * First test the above pipe, then test with suppressed output + * recompression (-D). The result should be the original input: + * + * --> | gzip -cf | xdelta3 -cf | xdelta3 -Ddcf | --> + * + * Finally we want to test that -D also disables input decompression: + * + * --> | gzip -cf | xdelta3 -Dcf | xdelta3 -Ddcf | gzip -dcf | --> + */ +static int +test_externally_compressed_io (xd3_stream *stream, int ignore) +{ + usize_t i; + int ret; + char buf[TESTBUFSIZE]; + + mt_init (& static_mtrand, 0x9f73f7fc); + + if ((ret = test_make_inputs (stream, NULL, NULL))) { return ret; } + + for (i = 0; i < SIZEOF_ARRAY (extcomp_types); i += 1) + { + main_extcomp *ext = & extcomp_types[i]; + + /* Test for the existence of the external command first, if not skip. */ + snprintf_func (buf, TESTBUFSIZE, "%s %s < /dev/null > /dev/null", ext->recomp_cmdname, ext->recomp_options); + + if ((ret = system (buf)) != 0) + { + XPR(NT "%s=0", ext->recomp_cmdname); + continue; + } + + if ((ret = test_compressed_pipe (stream, ext, buf, "-cfq", "-dcfq", 1, + "compression failed: identity pipe")) || + (ret = test_compressed_pipe (stream, ext, buf, "-cfq", "-Rdcfq", 0, + "compression failed: without recompression")) || + (ret = test_compressed_pipe (stream, ext, buf, "-Dcfq", "-Rdcfq", 1, + "compression failed: without decompression"))) + { + return ret; + } + } + + return 0; +} + +/* This tests the proper functioning of external decompression for + * source files. The source and target files are identical and + * compressed by gzip. Decoding such a delta with recompression + * disbaled (-R) should produce the original, uncompressed + * source/target file. Then it checks with output recompression + * enabled--in this case the output should be a compressed copy of the + * original source/target file. Then it checks that encoding with + * decompression disabled works--the compressed files are identical + * and decoding them should always produce a compressed output, + * regardless of -R since the encoded delta file had decompression + * disabled.. + */ +static int +test_source_decompression (xd3_stream *stream, int ignore) +{ + int ret; + char buf[TESTBUFSIZE]; + const main_extcomp *ext; + xoff_t dsize; + + mt_init (& static_mtrand, 0x9f73f7fc); + + test_setup (); + if ((ret = test_make_inputs (stream, NULL, NULL))) { return ret; } + + /* Use gzip. */ + if ((ext = main_get_compressor ("G")) == NULL) + { + XPR(NT "skipped"); + return 0; + } + + /* Save an uncompressed copy. */ + if ((ret = test_save_copy (TEST_TARGET_FILE))) { return ret; } + + /* Compress the source. */ + snprintf_func (buf, TESTBUFSIZE, "%s -1 %s < %s > %s", ext->recomp_cmdname, + ext->recomp_options, TEST_COPY_FILE, TEST_SOURCE_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + /* Compress the target. */ + snprintf_func (buf, TESTBUFSIZE, "%s -9 %s < %s > %s", ext->recomp_cmdname, + ext->recomp_options, TEST_COPY_FILE, TEST_TARGET_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + + /* Now the two identical files are compressed. Delta-encode the target, + * with decompression. */ + snprintf_func (buf, TESTBUFSIZE, "%s -e -vfq -s%s %s %s", program_name, TEST_SOURCE_FILE, + TEST_TARGET_FILE, TEST_DELTA_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + + /* Check that the compressed file is small (b/c inputs are + * identical). */ + if ((ret = test_file_size (TEST_DELTA_FILE, & dsize))) { return ret; } + /* Deltas for identical files should be very small. */ + if (dsize > 200) + { + XPR(NT "external compression did not happen\n"); + stream->msg = "external compression did not happen"; + return XD3_INTERNAL; + } + + /* Decode the delta file with recompression disabled, should get an + * uncompressed file out. */ + snprintf_func (buf, TESTBUFSIZE, "%s -v -dq -R -s%s %s %s", program_name, + TEST_SOURCE_FILE, TEST_DELTA_FILE, TEST_RECON_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + if ((ret = test_compare_files (TEST_COPY_FILE, + TEST_RECON_FILE))) { return ret; } + + /* Decode the delta file with recompression, should get a compressed file + * out. But we can't compare compressed files directly. */ + snprintf_func (buf, TESTBUFSIZE, "%s -v -dqf -s%s %s %s", program_name, + TEST_SOURCE_FILE, TEST_DELTA_FILE, TEST_RECON_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + snprintf_func (buf, TESTBUFSIZE, "%s %s < %s > %s", ext->decomp_cmdname, ext->decomp_options, + TEST_RECON_FILE, TEST_RECON2_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + if ((ret = test_compare_files (TEST_COPY_FILE, + TEST_RECON2_FILE))) { return ret; } + + /* Encode with decompression disabled */ + snprintf_func (buf, TESTBUFSIZE, "%s -e -D -vfq -s%s %s %s", program_name, + TEST_SOURCE_FILE, TEST_TARGET_FILE, TEST_DELTA_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + + /* Decode the delta file with decompression disabled, should get the + * identical compressed file out. */ + snprintf_func (buf, TESTBUFSIZE, "%s -d -D -vfq -s%s %s %s", program_name, + TEST_SOURCE_FILE, TEST_DELTA_FILE, TEST_RECON_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + if ((ret = test_compare_files (TEST_TARGET_FILE, + TEST_RECON_FILE))) { return ret; } + + test_cleanup(); + return 0; +} +#endif + +/*********************************************************************** + FORCE, STDOUT + ***********************************************************************/ + +/* This tests that output will not overwrite an existing file unless + * -f was specified. The test is for encoding (the same code handles + * it for decoding). */ +static int +test_force_behavior (xd3_stream *stream, int ignore) +{ + int ret; + char buf[TESTBUFSIZE]; + + /* Create empty target file */ + test_setup (); + snprintf_func (buf, TESTBUFSIZE, "cp /dev/null %s", TEST_TARGET_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + + /* Encode to delta file */ + snprintf_func (buf, TESTBUFSIZE, "%s -e %s %s", program_name, + TEST_TARGET_FILE, TEST_DELTA_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + + /* Encode again, should fail. */ + snprintf_func (buf, TESTBUFSIZE, "%s -q -e %s %s ", program_name, + TEST_TARGET_FILE, TEST_DELTA_FILE); + if ((ret = do_fail (stream, buf))) { return ret; } + + /* Force it, should succeed. */ + snprintf_func (buf, TESTBUFSIZE, "%s -f -e %s %s", program_name, + TEST_TARGET_FILE, TEST_DELTA_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + test_cleanup(); + return 0; +} + +/* This checks the proper operation of the -c flag. When specified + * the default output becomes stdout, otherwise the input must be + * provided (encode) or it may be defaulted (decode w/ app header). */ +static int +test_stdout_behavior (xd3_stream *stream, int ignore) +{ + int ret; + char buf[TESTBUFSIZE]; + + test_setup(); + snprintf_func (buf, TESTBUFSIZE, "cp /dev/null %s", TEST_TARGET_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + + /* Without -c, encode writes to delta file */ + snprintf_func (buf, TESTBUFSIZE, "%s -e %s %s", program_name, + TEST_TARGET_FILE, TEST_DELTA_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + + /* With -c, encode writes to stdout */ + snprintf_func (buf, TESTBUFSIZE, "%s -e -c %s > %s", program_name, + TEST_TARGET_FILE, TEST_DELTA_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + + /* Without -c, decode writes to target file name, but it fails because the + * file exists. */ + snprintf_func (buf, TESTBUFSIZE, "%s -q -d %s ", program_name, TEST_DELTA_FILE); + if ((ret = do_fail (stream, buf))) { return ret; } + + /* With -c, decode writes to stdout */ + snprintf_func (buf, TESTBUFSIZE, "%s -d -c %s > /dev/null", program_name, TEST_DELTA_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + test_cleanup(); + + return 0; +} + +/* This tests that the no-output flag (-J) works. */ +static int +test_no_output (xd3_stream *stream, int ignore) +{ + int ret; + char buf[TESTBUFSIZE]; + + test_setup (); + + snprintf_func (buf, TESTBUFSIZE, "touch %s && chmod 0000 %s", + TEST_NOPERM_FILE, TEST_NOPERM_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + + if ((ret = test_make_inputs (stream, NULL, NULL))) { return ret; } + + /* Try no_output encode w/out unwritable output file */ + snprintf_func (buf, TESTBUFSIZE, "%s -q -f -e %s %s", program_name, + TEST_TARGET_FILE, TEST_NOPERM_FILE); + if ((ret = do_fail (stream, buf))) { return ret; } + snprintf_func (buf, TESTBUFSIZE, "%s -J -e %s %s", program_name, + TEST_TARGET_FILE, TEST_NOPERM_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + + /* Now really write the delta to test decode no-output */ + snprintf_func (buf, TESTBUFSIZE, "%s -e %s %s", program_name, + TEST_TARGET_FILE, TEST_DELTA_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + + snprintf_func (buf, TESTBUFSIZE, "%s -q -f -d %s %s", program_name, + TEST_DELTA_FILE, TEST_NOPERM_FILE); + if ((ret = do_fail (stream, buf))) { return ret; } + snprintf_func (buf, TESTBUFSIZE, "%s -J -d %s %s", program_name, + TEST_DELTA_FILE, TEST_NOPERM_FILE); + if ((ret = do_cmd (stream, buf))) { return ret; } + test_cleanup (); + return 0; +} + +/*********************************************************************** + Source identical optimization + ***********************************************************************/ + +/* Computing a delta should be fastest when the two inputs are + * identical, this checks it. The library is called to compute a + * delta between a 10000 byte file, 1000 byte winsize, 500 byte source + * blocksize. The same buffer is used for both source and target. */ +static int +test_identical_behavior (xd3_stream *stream, int ignore) +{ +#define IDB_TGTSZ 10000 /* Not a power of two b/c of hard-coded expectations below. */ +#define IDB_BLKSZ 512 +#define IDB_WINSZ 1000 +#define IDB_DELSZ 1000 +#define IDB_WINCNT (IDB_TGTSZ / IDB_WINSZ) + + int ret, i; + uint8_t buf[IDB_TGTSZ]; + uint8_t del[IDB_DELSZ]; + uint8_t rec[IDB_TGTSZ]; + xd3_source source; + int nextencwin = 0; + int winstarts = 0, winfinishes = 0; + usize_t delpos = 0, recsize; + xd3_config config; + memset(&source, 0, sizeof(source)); + + for (i = 0; i < IDB_TGTSZ; i += 1) + { + buf[i] = (uint8_t) mt_random (&static_mtrand); + } + + stream->winsize = IDB_WINSZ; + + source.blksize = IDB_BLKSZ; + source.name = ""; + source.curblk = NULL; + source.curblkno = 0; + + if ((ret = xd3_set_source (stream, & source))) { goto fail; } + + /* Compute an delta between identical source and targets. */ + for (;;) + { + ret = xd3_encode_input (stream); + + if (ret == XD3_INPUT) + { + xd3_avail_input (stream, buf + (IDB_WINSZ * nextencwin), IDB_WINSZ); + nextencwin += 1; + continue; + } + + if (ret == XD3_GETSRCBLK) + { + source.curblkno = source.getblkno; + source.onblk = IDB_BLKSZ; + source.curblk = buf + source.getblkno * IDB_BLKSZ; + continue; + } + + if (ret == XD3_WINSTART) + { + winstarts++; + continue; + } + if (ret == XD3_WINFINISH) + { + winfinishes++; + if (winfinishes == IDB_WINCNT) + { + break; + } + continue; + } + + if (ret != XD3_OUTPUT) { goto fail; } + + CHECK(delpos + stream->avail_out <= IDB_DELSZ); + + memcpy (del + delpos, stream->next_out, stream->avail_out); + + delpos += stream->avail_out; + + xd3_consume_output (stream); + } + + CHECK(winfinishes == IDB_WINCNT); + CHECK(winstarts == IDB_WINCNT); + CHECK(nextencwin == IDB_WINCNT); + + /* Reset. */ + memset(&source, 0, sizeof(source)); + source.blksize = IDB_TGTSZ; + source.onblk = IDB_TGTSZ; + source.curblk = buf; + source.curblkno = 0; + + if ((ret = xd3_close_stream (stream))) { goto fail; } + xd3_free_stream (stream); + xd3_init_config (& config, 0); + if ((ret = xd3_config_stream (stream, & config))) { goto fail; } + if ((ret = xd3_set_source_and_size (stream, & source, IDB_TGTSZ))) { goto fail; } + + /* Decode. */ + if ((ret = xd3_decode_stream (stream, del, delpos, rec, & recsize, IDB_TGTSZ))) { goto fail; } + + /* Check result size and data. */ + if (recsize != IDB_TGTSZ) { stream->msg = "wrong size reconstruction"; goto fail; } + if (memcmp (rec, buf, IDB_TGTSZ) != 0) { stream->msg = "wrong data reconstruction"; goto fail; } + + /* Check that there was one copy per window. */ + IF_DEBUG (if (stream->n_scpy != IDB_WINCNT || + stream->n_add != 0 || + stream->n_run != 0) { stream->msg = "wrong copy count"; goto fail; }); + + /* Check that no checksums were computed because the initial match + was presumed. */ + IF_DEBUG (if (stream->large_ckcnt != 0) { stream->msg = "wrong checksum behavior"; goto fail; }); + + ret = 0; + fail: + return ret; +} + +/*********************************************************************** + String matching test + ***********************************************************************/ + +/* Check particular matching behaviors by calling + * xd3_string_match_soft directly with specific arguments. */ +typedef struct _string_match_test string_match_test; + +typedef enum +{ + SM_NONE = 0, + SM_LAZY = (1 << 1), +} string_match_flags; + +struct _string_match_test +{ + const char *input; + int flags; + const char *result; +}; + +static const string_match_test match_tests[] = +{ + /* nothing */ + { "1234567890", SM_NONE, "" }, + + /* basic run, copy */ + { "11111111112323232323", SM_NONE, "R0/10 C12/8@10" }, + + /* no run smaller than MIN_RUN=8 */ + { "1111111", SM_NONE, "C1/6@0" }, + { "11111111", SM_NONE, "R0/8" }, + + /* simple promotion: the third copy address depends on promotion */ + { "ABCDEF_ABCDEF^ABCDEF", SM_NONE, "C7/6@0 C14/6@7" }, + /* { "ABCDEF_ABCDEF^ABCDEF", SM_PROMOTE, "C7/6@0 C14/6@0" }, forgotten */ + + /* simple lazy: there is a better copy starting with "23 X" than "123 " */ + { "123 23 XYZ 123 XYZ", SM_NONE, "C11/4@0" }, + { "123 23 XYZ 123 XYZ", SM_LAZY, "C11/4@0 C12/6@4" }, + + /* trylazy: no lazy matches unless there are at least two characters beyond + * the first match */ + { "2123_121212", SM_LAZY, "C7/4@5" }, + { "2123_1212123", SM_LAZY, "C7/4@5" }, + { "2123_1212123_", SM_LAZY, "C7/4@5 C8/5@0" }, + + /* trylazy: no lazy matches if the copy is >= MAXLAZY=10 */ + { "2123_121212123_", SM_LAZY, "C7/6@5 C10/5@0" }, + { "2123_12121212123_", SM_LAZY, "C7/8@5 C12/5@0" }, + { "2123_1212121212123_", SM_LAZY, "C7/10@5" }, + + /* lazy run: check a run overlapped by a longer copy */ + { "11111112 111111112 1", SM_LAZY, "C1/6@0 R9/8 C10/10@0" }, + + /* lazy match: match_length,run_l >= min_match tests, shouldn't get any + * copies within the run, no run within the copy */ + { "^________^________ ", SM_LAZY, "R1/8 C9/9@0" }, + + /* chain depth: it only goes back 10. this checks that the 10th match hits + * and the 11th misses. */ + { "1234 1234_1234-1234=1234+1234[1234]1234{1234}1234<1234 ", SM_NONE, + "C5/4@0 C10/4@5 C15/4@10 C20/4@15 C25/4@20 C30/4@25 C35/4@30 C40/4@35 C45/4@40 C50/5@0" }, + { "1234 1234_1234-1234=1234+1234[1234]1234{1234}1234<1234>1234 ", SM_NONE, + "C5/4@0 C10/4@5 C15/4@10 C20/4@15 C25/4@20 C30/4@25 C35/4@30 C40/4@35 C45/4@40 C50/4@45 C55/4@50" }, + + /* ssmatch test */ + { "ABCDE___ABCDE*** BCDE***", SM_NONE, "C8/5@0 C17/4@1" }, + /*{ "ABCDE___ABCDE*** BCDE***", SM_SSMATCH, "C8/5@0 C17/7@9" }, forgotten */ +}; + +static int +test_string_matching (xd3_stream *stream, int ignore) +{ + usize_t i; + int ret; + xd3_config config; + char rbuf[TESTBUFSIZE]; + + for (i = 0; i < SIZEOF_ARRAY (match_tests); i += 1) + { + const string_match_test *test = & match_tests[i]; + char *rptr = rbuf; + usize_t len = (usize_t) strlen (test->input); + + xd3_free_stream (stream); + xd3_init_config (& config, 0); + + config.smatch_cfg = XD3_SMATCH_SOFT; + config.smatcher_soft.large_look = 4; + config.smatcher_soft.large_step = 4; + config.smatcher_soft.small_look = 4; + config.smatcher_soft.small_chain = 10; + config.smatcher_soft.small_lchain = 10; + config.smatcher_soft.max_lazy = (test->flags & SM_LAZY) ? 10 : 0; + config.smatcher_soft.long_enough = 10; + + if ((ret = xd3_config_stream (stream, & config))) { return ret; } + if ((ret = xd3_encode_init_full (stream))) { return ret; } + + xd3_avail_input (stream, (uint8_t*)test->input, len); + + if ((ret = stream->smatcher.string_match (stream))) { return ret; } + + *rptr = 0; + while (! xd3_rlist_empty (& stream->iopt_used)) + { + xd3_rinst *inst = xd3_rlist_pop_front (& stream->iopt_used); + + switch (inst->type) + { + case XD3_RUN: *rptr++ = 'R'; break; + case XD3_CPY: *rptr++ = 'C'; break; + default: CHECK(0); + } + + snprintf_func (rptr, rbuf+TESTBUFSIZE-rptr, "%d/%d", + inst->pos, inst->size); + rptr += strlen (rptr); + + if (inst->type == XD3_CPY) + { + *rptr++ = '@'; + snprintf_func (rptr, rbuf+TESTBUFSIZE-rptr, "%"Q"d", inst->addr); + rptr += strlen (rptr); + } + + *rptr++ = ' '; + + xd3_rlist_push_back (& stream->iopt_free, inst); + } + + if (rptr != rbuf) + { + rptr -= 1; *rptr = 0; + } + + if (strcmp (rbuf, test->result) != 0) + { + XPR(NT "test %u: expected %s: got %s", i, test->result, rbuf); + stream->msg = "wrong result"; + return XD3_INTERNAL; + } + } + + return 0; +} + +/* + * This is a test for many overlapping instructions. It must be a lazy + * matcher. + */ +static int +test_iopt_flush_instructions (xd3_stream *stream, int ignore) +{ + int ret, i; + usize_t tpos = 0; + usize_t delta_size, recon_size; + xd3_config config; + uint8_t target[TESTBUFSIZE]; + uint8_t delta[TESTBUFSIZE]; + uint8_t recon[TESTBUFSIZE]; + + xd3_free_stream (stream); + xd3_init_config (& config, 0); + + config.smatch_cfg = XD3_SMATCH_SOFT; + config.smatcher_soft.large_look = 16; + config.smatcher_soft.large_step = 16; + config.smatcher_soft.small_look = 4; + config.smatcher_soft.small_chain = 128; + config.smatcher_soft.small_lchain = 16; + config.smatcher_soft.max_lazy = 8; + config.smatcher_soft.long_enough = 128; + + if ((ret = xd3_config_stream (stream, & config))) { return ret; } + + for (i = 1; i < 250; i++) + { + target[tpos++] = i; + target[tpos++] = i+1; + target[tpos++] = i+2; + target[tpos++] = i+3; + target[tpos++] = 0; + } + for (i = 1; i < 253; i++) + { + target[tpos++] = i; + } + + if ((ret = xd3_encode_stream (stream, target, tpos, + delta, & delta_size, sizeof (delta)))) + { + return ret; + } + + xd3_free_stream(stream); + if ((ret = xd3_config_stream (stream, & config))) { return ret; } + + if ((ret = xd3_decode_stream (stream, delta, delta_size, + recon, & recon_size, sizeof (recon)))) + { + return ret; + } + + CHECK(tpos == recon_size); + CHECK(memcmp(target, recon, recon_size) == 0); + + return 0; +} + +/* + * This tests the 32/64bit ambiguity for source-window matching. + */ +static int +test_source_cksum_offset (xd3_stream *stream, int ignore) +{ + xd3_source source; + + // Inputs are: + struct { + xoff_t cpos; // stream->srcwin_cksum_pos; + xoff_t ipos; // stream->total_in; + xoff_t size; // stream->src->size; + + usize_t input; // input 32-bit offset + xoff_t output; // output 64-bit offset + + } cksum_test[] = { + // If cpos is <= 2^32 + { 1, 1, 1, 1, 1 }, + +#if XD3_USE_LARGEFILE64 +// cpos ipos size input output +// 0x____xxxxxULL, 0x____xxxxxULL, 0x____xxxxxULL, 0x___xxxxxUL, 0x____xxxxxULL + { 0x100100000ULL, 0x100000000ULL, 0x100200000ULL, 0x00000000UL, 0x100000000ULL }, + { 0x100100000ULL, 0x100000000ULL, 0x100200000ULL, 0xF0000000UL, 0x0F0000000ULL }, + + { 0x100200000ULL, 0x100100000ULL, 0x100200000ULL, 0x00300000UL, 0x000300000ULL }, + + { 25771983104ULL, 25770000000ULL, 26414808769ULL, 2139216707UL, 23614053187ULL }, + +#endif + + { 0, 0, 0, 0, 0 }, + }, *test_ptr; + + stream->src = &source; + + for (test_ptr = cksum_test; test_ptr->cpos; test_ptr++) { + xoff_t r; + stream->srcwin_cksum_pos = test_ptr->cpos; + stream->total_in = test_ptr->ipos; + + r = xd3_source_cksum_offset(stream, test_ptr->input); + CHECK(r == test_ptr->output); + } + return 0; +} + +static int +test_in_memory (xd3_stream *stream, int ignore) +{ + // test_text is 256 bytes + uint8_t ibuf[sizeof(test_text)]; + uint8_t dbuf[sizeof(test_text)]; + uint8_t obuf[sizeof(test_text)]; + usize_t size = sizeof(test_text); + usize_t dsize, osize; + int r1, r2; + int eflags = SECONDARY_DJW ? XD3_SEC_DJW : 0; + + memcpy(ibuf, test_text, size); + memset(ibuf + 128, 0, 16); + + r1 = xd3_encode_memory(ibuf, size, + test_text, size, + dbuf, &dsize, size, eflags); + + r2 = xd3_decode_memory(dbuf, dsize, + test_text, size, + obuf, &osize, size, 0); + + if (r1 != 0 || r2 != 0 || dsize >= (size/2) || dsize < 1 || + osize != size) { + stream->msg = "encode/decode size error"; + return XD3_INTERNAL; + } + + if (memcmp(obuf, ibuf, size) != 0) { + stream->msg = "encode/decode data error"; + return XD3_INTERNAL; + } + + return 0; +} + +/*********************************************************************** + TEST MAIN + ***********************************************************************/ + +static int +xd3_selftest (void) +{ +#define DO_TEST(fn,flags,arg) \ + do { \ + xd3_stream stream; \ + xd3_config config; \ + xd3_init_config (& config, flags); \ + XPR(NT "testing " #fn "%s...", \ + flags ? (" (" #flags ")") : ""); \ + if ((ret = xd3_config_stream (& stream, & config) == 0) && \ + (ret = test_ ## fn (& stream, arg)) == 0) { \ + XPR(NTR " success\n"); \ + } else { \ + XPR(NTR " failed: %s: %s\n", xd3_errstring (& stream), \ + xd3_mainerror (ret)); } \ + xd3_free_stream (& stream); \ + if (ret != 0) { goto failure; } \ + } while (0) + + int ret; + DO_TEST (random_numbers, 0, 0); + + DO_TEST (decode_integer_end_of_input, 0, 0); + DO_TEST (decode_integer_overflow, 0, 0); + DO_TEST (encode_decode_uint32_t, 0, 0); + DO_TEST (encode_decode_uint64_t, 0, 0); + DO_TEST (usize_t_overflow, 0, 0); + DO_TEST (forward_match, 0, 0); + + DO_TEST (address_cache, 0, 0); + IF_GENCODETBL (DO_TEST (address_cache, XD3_ALT_CODE_TABLE, 0)); + + DO_TEST (string_matching, 0, 0); + DO_TEST (choose_instruction, 0, 0); + DO_TEST (identical_behavior, 0, 0); + DO_TEST (in_memory, 0, 0); + + IF_GENCODETBL (DO_TEST (choose_instruction, XD3_ALT_CODE_TABLE, 0)); + IF_GENCODETBL (DO_TEST (encode_code_table, 0, 0)); + + DO_TEST (iopt_flush_instructions, 0, 0); + DO_TEST (source_cksum_offset, 0, 0); + + DO_TEST (decompress_single_bit_error, 0, 3); + DO_TEST (decompress_single_bit_error, XD3_ADLER32, 3); + + IF_LZMA (DO_TEST (decompress_single_bit_error, XD3_SEC_LZMA, 54)); + IF_FGK (DO_TEST (decompress_single_bit_error, XD3_SEC_FGK, 3)); + IF_DJW (DO_TEST (decompress_single_bit_error, XD3_SEC_DJW, 8)); + + /* There are many expected non-failures for ALT_CODE_TABLE because + * not all of the instruction codes are used. */ + IF_GENCODETBL ( + DO_TEST (decompress_single_bit_error, XD3_ALT_CODE_TABLE, 224)); + +#if SHELL_TESTS + DO_TEST (force_behavior, 0, 0); + DO_TEST (stdout_behavior, 0, 0); + DO_TEST (no_output, 0, 0); + DO_TEST (command_line_arguments, 0, 0); + +#if EXTERNAL_COMPRESSION + DO_TEST (source_decompression, 0, 0); + DO_TEST (externally_compressed_io, 0, 0); +#endif + + DO_TEST (recode_command, 0, 0); +#endif + + IF_LZMA (DO_TEST (secondary_lzma, 0, 1)); + IF_DJW (DO_TEST (secondary_huff, 0, DJW_MAX_GROUPS)); + IF_FGK (DO_TEST (secondary_fgk, 0, 1)); + + DO_TEST (compressed_stream_overflow, 0, 0); + IF_LZMA (DO_TEST (compressed_stream_overflow, XD3_SEC_LZMA, 0)); + +failure: + test_cleanup (); + return ret == 0 ? EXIT_SUCCESS : EXIT_FAILURE; +#undef DO_TEST +} diff --git a/3rdparty/xdelta3-3.0.8/xdelta3.c b/3rdparty/xdelta3-3.0.8/xdelta3.c new file mode 100644 index 0000000..dffdf56 --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3.c @@ -0,0 +1,5497 @@ +/* xdelta 3 - delta compression tools and library + * Copyright (C) 2001, 2003, 2004, 2005, 2006, 2007, + * 2008, 2009, 2010, 2011, 2012, 2013. Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + + ------------------------------------------------------------------- + + Xdelta 3 + + The goal of this library is to to implement both the (stand-alone) + data-compression and delta-compression aspects of VCDIFF encoding, and + to support a programming interface that works like Zlib + (http://www.gzip.org/zlib.html). See RFC3284: The VCDIFF Generic + Differencing and Compression Data Format. + + VCDIFF is a unified encoding that combines data-compression and + delta-encoding ("differencing"). + + VCDIFF has a detailed byte-code instruction set with many features. + The instruction format supports an immediate size operand for small + COPYs and ADDs (e.g., under 18 bytes). There are also instruction + "modes", which are used to compress COPY addresses by using two + address caches. An instruction mode refers to slots in the NEAR + and SAME caches for recent addresses. NEAR remembers the + previous 4 (by default) COPY addresses, and SAME catches + frequent re-uses of the same address using a 3-way (by default) + 256-entry associative cache of [ADDR mod 256], the encoded byte. + A hit in the NEAR/SAME cache requires 0/1 ADDR bytes. + + VCDIFF has a default instruction table, but an alternate + instruction tables may themselves be be delta-compressed and + included in the encoding header. This allows even more freedom. + There are 9 instruction modes in the default code table, 4 near, 3 + same, VCD_SELF (absolute encoding) and VCD_HERE (relative to the + current position). + + ---------------------------------------------------------------------- + + Algorithms + + Aside from the details of encoding and decoding, there are a bunch + of algorithms needed. + + 1. STRING-MATCH. A two-level fingerprinting approach is used. A + single loop computes the two checksums -- small and large -- at + successive offsets in the TARGET file. The large checksum is more + accurate and is used to discover SOURCE matches, which are + potentially very long. The small checksum is used to discover + copies within the TARGET. Small matching, which is more expensive, + usually dominates the large STRING-MATCH costs in this code - the + more exhaustive the search, the better the results. Either of the + two string-matching mechanisms may be disabled. + + 2. INSTRUCTION SELECTION. The IOPT buffer here represents a queue + used to store overlapping copy instructions. There are two possible + optimizations that go beyond a greedy search. Both of these fall + into the category of "non-greedy matching" optimizations. + + The first optimization stems from backward SOURCE-COPY matching. + When a new SOURCE-COPY instruction covers a previous instruction in + the target completely, it is erased from the queue. Randal Burns + originally analyzed these algorithms and did a lot of related work + (\cite the 1.5-pass algorithm). + + The second optimization comes by the encoding of common very-small + COPY and ADD instructions, for which there are special DOUBLE-code + instructions, which code two instructions in a single byte. + + The cost of bad instruction-selection overhead is relatively high + for data-compression, relative to delta-compression, so this second + optimization is fairly important. With "lazy" matching (the name + used in Zlib for a similar optimization), the string-match + algorithm searches after a match for potential overlapping copy + instructions. In Xdelta and by default, VCDIFF, the minimum match + size is 4 bytes, whereas Zlib searches with a 3-byte minimum. This + feature, combined with double instructions, provides a nice + challenge. Search in this file for "black magic", a heuristic. + + 3. STREAM ALIGNMENT. Stream alignment is needed to compress large + inputs in constant space. See xd3_srcwin_move_point(). + + 4. WINDOW SELECTION. When the IOPT buffer flushes, in the first call + to xd3_iopt_finish_encoding containing any kind of copy instruction, + the parameters of the source window must be decided: the offset into + the source and the length of the window. Since the IOPT buffer is + finite, the program may be forced to fix these values before knowing + the best offset/length. + + 5. SECONDARY COMPRESSION. VCDIFF supports a secondary encoding to + be applied to the individual sections of the data format, which are + ADDRess, INSTruction, and DATA. Several secondary compressor + variations are implemented here, although none is standardized yet. + + One is an adaptive huffman algorithm -- the FGK algorithm (Faller, + Gallager, and Knuth, 1985). This compressor is extremely slow. + + The other is a simple static Huffman routine, which is the base + case of a semi-adaptive scheme published by D.J. Wheeler and first + widely used in bzip2 (by Julian Seward). This is a very + interesting algorithm, originally published in nearly cryptic form + by D.J. Wheeler. !!!NOTE!!! Because these are not standardized, + secondary compression remains off by default. + ftp://ftp.cl.cam.ac.uk/users/djw3/bred3.{c,ps} + -------------------------------------------------------------------- + + Other Features + + 1. USER CONVENIENCE + + For user convenience, it is essential to recognize Gzip-compressed + files and automatically Gzip-decompress them prior to + delta-compression (or else no delta-compression will be achieved + unless the user manually decompresses the inputs). The compressed + represention competes with Xdelta, and this must be hidden from the + command-line user interface. The Xdelta-1.x encoding was simple, not + compressed itself, so Xdelta-1.x uses Zlib internally to compress the + representation. + + This implementation supports external compression, which implements + the necessary fork() and pipe() mechanics. There is a tricky step + involved to support automatic detection of a compressed input in a + non-seekable input. First you read a bit of the input to detect + magic headers. When a compressed format is recognized, exec() the + external compression program and create a second child process to + copy the original input stream. [Footnote: There is a difficulty + related to using Gzip externally. It is not possible to decompress + and recompress a Gzip file transparently. If FILE.GZ had a + cryptographic signature, then, after: (1) Gzip-decompression, (2) + Xdelta-encoding, (3) Gzip-compression the signature could be + broken. The only way to solve this problem is to guess at Gzip's + compression level or control it by other means. I recommend that + specific implementations of any compression scheme store + information needed to exactly re-compress the input, that way + external compression is transparent - however, this won't happen + here until it has stabilized.] + + 2. APPLICATION-HEADER + + This feature was introduced in RFC3284. It allows any application + to include a header within the VCDIFF file format. This allows + general inter-application data exchange with support for + application-specific extensions to communicate metadata. + + 3. VCDIFF CHECKSUM + + An optional checksum value is included with each window, which can + be used to validate the final result. This verifies the correct source + file was used for decompression as well as the obvious advantage: + checking the implementation (and underlying) correctness. + + 4. LIGHT WEIGHT + + The code makes efforts to avoid copying data more than necessary. + The code delays many initialization tasks until the first use, it + optimizes for identical (perfectly matching) inputs. It does not + compute any checksums until the first lookup misses. Memory usage + is reduced. String-matching is templatized (by slightly gross use + of CPP) to hard-code alternative compile-time defaults. The code + has few outside dependencies. + ---------------------------------------------------------------------- + + The default rfc3284 instruction table: + (see RFC for the explanation) + + TYPE SIZE MODE TYPE SIZE MODE INDEX + -------------------------------------------------------------------- + 1. Run 0 0 Noop 0 0 0 + 2. Add 0, [1,17] 0 Noop 0 0 [1,18] + 3. Copy 0, [4,18] 0 Noop 0 0 [19,34] + 4. Copy 0, [4,18] 1 Noop 0 0 [35,50] + 5. Copy 0, [4,18] 2 Noop 0 0 [51,66] + 6. Copy 0, [4,18] 3 Noop 0 0 [67,82] + 7. Copy 0, [4,18] 4 Noop 0 0 [83,98] + 8. Copy 0, [4,18] 5 Noop 0 0 [99,114] + 9. Copy 0, [4,18] 6 Noop 0 0 [115,130] + 10. Copy 0, [4,18] 7 Noop 0 0 [131,146] + 11. Copy 0, [4,18] 8 Noop 0 0 [147,162] + 12. Add [1,4] 0 Copy [4,6] 0 [163,174] + 13. Add [1,4] 0 Copy [4,6] 1 [175,186] + 14. Add [1,4] 0 Copy [4,6] 2 [187,198] + 15. Add [1,4] 0 Copy [4,6] 3 [199,210] + 16. Add [1,4] 0 Copy [4,6] 4 [211,222] + 17. Add [1,4] 0 Copy [4,6] 5 [223,234] + 18. Add [1,4] 0 Copy 4 6 [235,238] + 19. Add [1,4] 0 Copy 4 7 [239,242] + 20. Add [1,4] 0 Copy 4 8 [243,246] + 21. Copy 4 [0,8] Add 1 0 [247,255] + -------------------------------------------------------------------- + + Reading the source: Overview + + This file includes itself in several passes to macro-expand certain + sections with variable forms. Just read ahead, there's only a + little confusion. I know this sounds ugly, but hard-coding some of + the string-matching parameters results in a 10-15% increase in + string-match performance. The only time this hurts is when you have + unbalanced #if/endifs. + + A single compilation unit tames the Makefile. In short, this is to + allow the above-described hack without an explodingMakefile. The + single compilation unit includes the core library features, + configurable string-match templates, optional main() command-line + tool, misc optional features, and a regression test. Features are + controled with CPP #defines, see Makefile.am. + + The initial __XDELTA3_C_HEADER_PASS__ starts first, the _INLINE_ and + _TEMPLATE_ sections follow. Easy stuff first, hard stuff last. + + Optional features include: + + xdelta3-main.h The command-line interface, external compression + support, POSIX-specific, info & VCDIFF-debug tools. + xdelta3-second.h The common secondary compression routines. + xdelta3-decoder.h All decoding routines. + xdelta3-djw.h The semi-adaptive huffman secondary encoder. + xdelta3-fgk.h The adaptive huffman secondary encoder. + xdelta3-test.h The unit test covers major algorithms, + encoding and decoding. There are single-bit + error decoding tests. There are 32/64-bit file size + boundary tests. There are command-line tests. + There are compression tests. There are external + compression tests. There are string-matching tests. + There should be more tests... + + Additional headers include: + + xdelta3.h The public header file. + xdelta3-cfgs.h The default settings for default, built-in + encoders. These are hard-coded at + compile-time. There is also a single + soft-coded string matcher for experimenting + with arbitrary values. + xdelta3-list.h A cyclic list template + + Misc little debug utilities: + + badcopy.c Randomly modifies an input file based on two + parameters: (1) the probability that a byte in + the file is replaced with a pseudo-random value, + and (2) the mean change size. Changes are + generated using an expoential distribution + which approximates the expected error_prob + distribution. + -------------------------------------------------------------------- + + This file itself is unusually large. I hope to defend this layout + with lots of comments. Everything in this file is related to + encoding and decoding. I like it all together - the template stuff + is just a hack. */ + +#ifndef __XDELTA3_C_HEADER_PASS__ +#define __XDELTA3_C_HEADER_PASS__ + +#include "xdelta3.h" + +/*********************************************************************** + STATIC CONFIGURATION + ***********************************************************************/ + +#ifndef XD3_MAIN /* the main application */ +#define XD3_MAIN 0 +#endif + +#ifndef VCDIFF_TOOLS +#define VCDIFF_TOOLS XD3_MAIN +#endif + +#ifndef SECONDARY_FGK /* one from the algorithm preservation department: */ +#define SECONDARY_FGK 0 /* adaptive Huffman routines */ +#endif + +#ifndef SECONDARY_DJW /* semi-adaptive/static Huffman for the eventual */ +#define SECONDARY_DJW 0 /* standardization, off by default until such time. */ +#endif + +#ifndef SECONDARY_LZMA +#ifdef HAVE_LZMA_H +#define SECONDARY_LZMA 1 +#else +#define SECONDARY_LZMA 0 +#endif +#endif + +#ifndef GENERIC_ENCODE_TABLES /* These three are the RFC-spec app-specific */ +#define GENERIC_ENCODE_TABLES 0 /* code features. This is tested but not */ +#endif /* recommended unless there's a real use. */ +#ifndef GENERIC_ENCODE_TABLES_COMPUTE +#define GENERIC_ENCODE_TABLES_COMPUTE 0 +#endif +#ifndef GENERIC_ENCODE_TABLES_COMPUTE_PRINT +#define GENERIC_ENCODE_TABLES_COMPUTE_PRINT 0 +#endif + +#if XD3_ENCODER +#define IF_ENCODER(x) x +#else +#define IF_ENCODER(x) +#endif + +/***********************************************************************/ + + /* header indicator bits */ +#define VCD_SECONDARY (1U << 0) /* uses secondary compressor */ +#define VCD_CODETABLE (1U << 1) /* supplies code table data */ +#define VCD_APPHEADER (1U << 2) /* supplies application data */ +#define VCD_INVHDR (~0x7U) + + /* window indicator bits */ +#define VCD_SOURCE (1U << 0) /* copy window in source file */ +#define VCD_TARGET (1U << 1) /* copy window in target file */ +#define VCD_ADLER32 (1U << 2) /* has adler32 checksum */ +#define VCD_INVWIN (~0x7U) + +#define VCD_SRCORTGT (VCD_SOURCE | VCD_TARGET) + + /* delta indicator bits */ +#define VCD_DATACOMP (1U << 0) +#define VCD_INSTCOMP (1U << 1) +#define VCD_ADDRCOMP (1U << 2) +#define VCD_INVDEL (~0x7U) + +typedef enum { + VCD_DJW_ID = 1, + VCD_LZMA_ID = 2, + VCD_FGK_ID = 16 /* Note: these are not standard IANA-allocated IDs! */ +} xd3_secondary_ids; + +typedef enum { + SEC_NOFLAGS = 0, + + /* Note: SEC_COUNT_FREQS Not implemented (to eliminate 1st Huffman pass) */ + SEC_COUNT_FREQS = (1 << 0) +} xd3_secondary_flags; + +typedef enum { + DATA_SECTION, /* These indicate which section to the secondary + * compressor. */ + INST_SECTION, /* The header section is not compressed, therefore not + * listed here. */ + ADDR_SECTION +} xd3_section_type; + +typedef unsigned int xd3_rtype; + +/***********************************************************************/ + +#include "xdelta3-list.h" + +XD3_MAKELIST(xd3_rlist, xd3_rinst, link); + +/***********************************************************************/ + +#define SECONDARY_MIN_SAVINGS 2 /* Secondary compression has to save + at least this many bytes. */ +#define SECONDARY_MIN_INPUT 10 /* Secondary compression needs at + least this many bytes. */ + +#define VCDIFF_MAGIC1 0xd6 /* 1st file byte */ +#define VCDIFF_MAGIC2 0xc3 /* 2nd file byte */ +#define VCDIFF_MAGIC3 0xc4 /* 3rd file byte */ +#define VCDIFF_VERSION 0x00 /* 4th file byte */ + +#define VCD_SELF 0 /* 1st address mode */ +#define VCD_HERE 1 /* 2nd address mode */ + +#define CODE_TABLE_STRING_SIZE (6 * 256) /* Should fit a code table string. */ +#define CODE_TABLE_VCDIFF_SIZE (6 * 256) /* Should fit a compressed code + * table string */ + +#define SECONDARY_ANY (SECONDARY_DJW || SECONDARY_FGK || SECONDARY_LZMA) + +#define ALPHABET_SIZE 256 /* Used in test code--size of the secondary + * compressor alphabet. */ + +#define HASH_PERMUTE 1 /* The input is permuted by random nums */ +#define ADLER_LARGE_CKSUM 1 /* Adler checksum vs. RK checksum */ + +#define HASH_CKOFFSET 1U /* Table entries distinguish "no-entry" from + * offset 0 using this offset. */ + +#define MIN_SMALL_LOOK 2U /* Match-optimization stuff. */ +#define MIN_LARGE_LOOK 2U +#define MIN_MATCH_OFFSET 1U +#define MAX_MATCH_SPLIT 18U /* VCDIFF code table: 18 is the default limit + * for direct-coded ADD sizes */ + +#define LEAST_MATCH_INCR 0 /* The least number of bytes an overlapping + * match must beat the preceding match by. This + * is a bias for the lazy match optimization. A + * non-zero value means that an adjacent match + * has to be better by more than the step + * between them. 0. */ + +#define MIN_MATCH 4U /* VCDIFF code table: MIN_MATCH=4 */ +#define MIN_ADD 1U /* 1 */ +#define MIN_RUN 8U /* The shortest run, if it is shorter than this + * an immediate add/copy will be just as good. + * ADD1/COPY6 = 1I+1D+1A bytes, RUN18 = + * 1I+1D+1A. */ + +#define MAX_MODES 9 /* Maximum number of nodes used for + * compression--does not limit decompression. */ + +#define ENC_SECTS 4 /* Number of separate output sections. */ + +#define HDR_TAIL(s) ((s)->enc_tails[0]) +#define DATA_TAIL(s) ((s)->enc_tails[1]) +#define INST_TAIL(s) ((s)->enc_tails[2]) +#define ADDR_TAIL(s) ((s)->enc_tails[3]) + +#define HDR_HEAD(s) ((s)->enc_heads[0]) +#define DATA_HEAD(s) ((s)->enc_heads[1]) +#define INST_HEAD(s) ((s)->enc_heads[2]) +#define ADDR_HEAD(s) ((s)->enc_heads[3]) + +#define TOTAL_MODES(x) (2+(x)->acache.s_same+(x)->acache.s_near) + +/* Template instances. */ +#if XD3_BUILD_SLOW +#define IF_BUILD_SLOW(x) x +#else +#define IF_BUILD_SLOW(x) +#endif +#if XD3_BUILD_FAST +#define IF_BUILD_FAST(x) x +#else +#define IF_BUILD_FAST(x) +#endif +#if XD3_BUILD_FASTER +#define IF_BUILD_FASTER(x) x +#else +#define IF_BUILD_FASTER(x) +#endif +#if XD3_BUILD_FASTEST +#define IF_BUILD_FASTEST(x) x +#else +#define IF_BUILD_FASTEST(x) +#endif +#if XD3_BUILD_SOFT +#define IF_BUILD_SOFT(x) x +#else +#define IF_BUILD_SOFT(x) +#endif +#if XD3_BUILD_DEFAULT +#define IF_BUILD_DEFAULT(x) x +#else +#define IF_BUILD_DEFAULT(x) +#endif + +/* Consume N bytes of input, only used by the decoder. */ +#define DECODE_INPUT(n) \ + do { \ + stream->total_in += (xoff_t) (n); \ + stream->avail_in -= (n); \ + stream->next_in += (n); \ + } while (0) + +/* Update the run-length state */ +#define NEXTRUN(c) do { if ((c) == run_c) { run_l += 1; } \ + else { run_c = (c); run_l = 1; } } while (0) + +/* This CPP-conditional stuff can be cleaned up... */ +#if REGRESSION_TEST +#define IF_REGRESSION(x) x +#else +#define IF_REGRESSION(x) +#endif + +/***********************************************************************/ + +#if XD3_ENCODER +static void* xd3_alloc0 (xd3_stream *stream, + usize_t elts, + usize_t size); + + +static xd3_output* xd3_alloc_output (xd3_stream *stream, + xd3_output *old_output); + +static int xd3_alloc_iopt (xd3_stream *stream, usize_t elts); + +static void xd3_free_output (xd3_stream *stream, + xd3_output *output); + +static int xd3_emit_byte (xd3_stream *stream, + xd3_output **outputp, + uint8_t code); + +static int xd3_emit_bytes (xd3_stream *stream, + xd3_output **outputp, + const uint8_t *base, + usize_t size); + +static int xd3_emit_double (xd3_stream *stream, xd3_rinst *first, + xd3_rinst *second, usize_t code); +static int xd3_emit_single (xd3_stream *stream, xd3_rinst *single, + usize_t code); + +static usize_t xd3_sizeof_output (xd3_output *output); +static void xd3_encode_reset (xd3_stream *stream); + +static int xd3_source_match_setup (xd3_stream *stream, xoff_t srcpos); +static int xd3_source_extend_match (xd3_stream *stream); +static int xd3_srcwin_setup (xd3_stream *stream); +static usize_t xd3_iopt_last_matched (xd3_stream *stream); +static int xd3_emit_uint32_t (xd3_stream *stream, xd3_output **output, + uint32_t num); + +static usize_t xd3_smatch (xd3_stream *stream, + usize_t base, + usize_t scksum, + usize_t *match_offset); +static int xd3_string_match_init (xd3_stream *stream); +static uint32_t xd3_scksum (uint32_t *state, const uint8_t *seg, + const usize_t ln); +static usize_t xd3_comprun (const uint8_t *seg, usize_t slook, uint8_t *run_cp); +static int xd3_srcwin_move_point (xd3_stream *stream, + usize_t *next_move_point); + +static int xd3_emit_run (xd3_stream *stream, usize_t pos, + usize_t size, uint8_t *run_c); +static usize_t xd3_checksum_hash (const xd3_hash_cfg *cfg, + const usize_t cksum); +static xoff_t xd3_source_cksum_offset(xd3_stream *stream, usize_t low); +static void xd3_scksum_insert (xd3_stream *stream, + usize_t inx, + usize_t scksum, + usize_t pos); + + +#if XD3_DEBUG +static void xd3_verify_run_state (xd3_stream *stream, + const uint8_t *inp, + usize_t x_run_l, + uint8_t *x_run_c); +static void xd3_verify_large_state (xd3_stream *stream, + const uint8_t *inp, + uint32_t x_cksum); +static void xd3_verify_small_state (xd3_stream *stream, + const uint8_t *inp, + uint32_t x_cksum); + +#endif /* XD3_DEBUG */ +#endif /* XD3_ENCODER */ + +static int xd3_decode_allocate (xd3_stream *stream, usize_t size, + uint8_t **copied1, usize_t *alloc1); + +static void xd3_compute_code_table_string (const xd3_dinst *code_table, + uint8_t *str); +static void* xd3_alloc (xd3_stream *stream, usize_t elts, usize_t size); +static void xd3_free (xd3_stream *stream, void *ptr); + +static int xd3_read_uint32_t (xd3_stream *stream, const uint8_t **inpp, + const uint8_t *max, uint32_t *valp); + +#if REGRESSION_TEST +static int xd3_selftest (void); +#endif + +/***********************************************************************/ + +#define UINT32_OFLOW_MASK 0xfe000000U +#define UINT64_OFLOW_MASK 0xfe00000000000000ULL + +#if SIZEOF_USIZE_T == 4 +#define USIZE_T_MAX UINT32_MAX +#define xd3_decode_size xd3_decode_uint32_t +#define xd3_emit_size xd3_emit_uint32_t +#define xd3_sizeof_size xd3_sizeof_uint32_t +#define xd3_read_size xd3_read_uint32_t +#elif SIZEOF_USIZE_T == 8 +#define USIZE_T_MAX UINT64_MAX +#define xd3_decode_size xd3_decode_uint64_t +#define xd3_emit_size xd3_emit_uint64_t +#define xd3_sizeof_size xd3_sizeof_uint64_t +#define xd3_read_size xd3_read_uint64_t +#endif + +#if SIZEOF_XOFF_T == 4 +#define XOFF_T_MAX UINT32_MAX +#define xd3_decode_offset xd3_decode_uint32_t +#define xd3_emit_offset xd3_emit_uint32_t +#elif SIZEOF_XOFF_T == 8 +#define XOFF_T_MAX UINT64_MAX +#define xd3_decode_offset xd3_decode_uint64_t +#define xd3_emit_offset xd3_emit_uint64_t +#endif + +#define USIZE_T_OVERFLOW(a,b) ((USIZE_T_MAX - (usize_t) (a)) < (usize_t) (b)) +#define XOFF_T_OVERFLOW(a,b) ((XOFF_T_MAX - (xoff_t) (a)) < (xoff_t) (b)) + +const char* xd3_strerror (int ret) +{ + switch (ret) + { + case XD3_INPUT: return "XD3_INPUT"; + case XD3_OUTPUT: return "XD3_OUTPUT"; + case XD3_GETSRCBLK: return "XD3_GETSRCBLK"; + case XD3_GOTHEADER: return "XD3_GOTHEADER"; + case XD3_WINSTART: return "XD3_WINSTART"; + case XD3_WINFINISH: return "XD3_WINFINISH"; + case XD3_TOOFARBACK: return "XD3_TOOFARBACK"; + case XD3_INTERNAL: return "XD3_INTERNAL"; + case XD3_INVALID: return "XD3_INVALID"; + case XD3_INVALID_INPUT: return "XD3_INVALID_INPUT"; + case XD3_NOSECOND: return "XD3_NOSECOND"; + case XD3_UNIMPLEMENTED: return "XD3_UNIMPLEMENTED"; + } + return NULL; +} + +/***********************************************************************/ + +#define xd3_sec_data(s) ((s)->sec_stream_d) +#define xd3_sec_inst(s) ((s)->sec_stream_i) +#define xd3_sec_addr(s) ((s)->sec_stream_a) + +struct _xd3_sec_type +{ + int id; + const char *name; + xd3_secondary_flags flags; + + /* xd3_sec_stream is opaque to the generic code */ + xd3_sec_stream* (*alloc) (xd3_stream *stream); + void (*destroy) (xd3_stream *stream, + xd3_sec_stream *sec); + int (*init) (xd3_stream *stream, + xd3_sec_stream *sec_stream, + int is_encode); + int (*decode) (xd3_stream *stream, + xd3_sec_stream *sec_stream, + const uint8_t **input, + const uint8_t *input_end, + uint8_t **output, + const uint8_t *output_end); +#if XD3_ENCODER + int (*encode) (xd3_stream *stream, + xd3_sec_stream *sec_stream, + xd3_output *input, + xd3_output *output, + xd3_sec_cfg *cfg); +#endif +}; + +#define BIT_STATE_ENCODE_INIT { 0, 1 } +#define BIT_STATE_DECODE_INIT { 0, 0x100 } + +typedef struct _bit_state bit_state; +struct _bit_state +{ + usize_t cur_byte; + usize_t cur_mask; +}; + +#if SECONDARY_ANY == 0 +#define IF_SEC(x) +#define IF_NSEC(x) x +#else /* yuck */ +#define IF_SEC(x) x +#define IF_NSEC(x) +static int +xd3_decode_secondary (xd3_stream *stream, + xd3_desect *sect, + xd3_sec_stream **sec_streamp); +#if XD3_ENCODER +static int +xd3_encode_secondary (xd3_stream *stream, + xd3_output **head, + xd3_output **tail, + xd3_sec_stream **sec_streamp, + xd3_sec_cfg *cfg, + int *did_it); +#endif +#endif /* SECONDARY_ANY */ + +#if SECONDARY_FGK +extern const xd3_sec_type fgk_sec_type; +#define IF_FGK(x) x +#define FGK_CASE(s) \ + s->sec_type = & fgk_sec_type; \ + break; +#else +#define IF_FGK(x) +#define FGK_CASE(s) \ + s->msg = "unavailable secondary compressor: FGK Adaptive Huffman"; \ + return XD3_INTERNAL; +#endif + +#if SECONDARY_DJW +extern const xd3_sec_type djw_sec_type; +#define IF_DJW(x) x +#define DJW_CASE(s) \ + s->sec_type = & djw_sec_type; \ + break; +#else +#define IF_DJW(x) +#define DJW_CASE(s) \ + s->msg = "unavailable secondary compressor: DJW Static Huffman"; \ + return XD3_INTERNAL; +#endif + +#if SECONDARY_LZMA +extern const xd3_sec_type lzma_sec_type; +#define IF_LZMA(x) x +#define LZMA_CASE(s) \ + s->sec_type = & lzma_sec_type; \ + break; +#else +#define IF_LZMA(x) +#define LZMA_CASE(s) \ + s->msg = "unavailable secondary compressor: LZMA"; \ + return XD3_INTERNAL; +#endif + +/***********************************************************************/ + +#include "xdelta3-hash.h" + +/* Process template passes - this includes xdelta3.c several times. */ +#define __XDELTA3_C_TEMPLATE_PASS__ +#include "xdelta3-cfgs.h" +#undef __XDELTA3_C_TEMPLATE_PASS__ + +/* Process the inline pass. */ +#define __XDELTA3_C_INLINE_PASS__ +#include "xdelta3.c" +#undef __XDELTA3_C_INLINE_PASS__ + +/* Secondary compression */ +#if SECONDARY_ANY +#include "xdelta3-second.h" +#endif + +#if SECONDARY_FGK +#include "xdelta3-fgk.h" +const xd3_sec_type fgk_sec_type = +{ + VCD_FGK_ID, + "FGK Adaptive Huffman", + SEC_NOFLAGS, + (xd3_sec_stream* (*)(xd3_stream*)) fgk_alloc, + (void (*)(xd3_stream*, xd3_sec_stream*)) fgk_destroy, + (int (*)(xd3_stream*, xd3_sec_stream*, int)) fgk_init, + (int (*)(xd3_stream*, xd3_sec_stream*, const uint8_t**, const uint8_t*, + uint8_t**, const uint8_t*)) xd3_decode_fgk, + IF_ENCODER((int (*)(xd3_stream*, xd3_sec_stream*, xd3_output*, + xd3_output*, xd3_sec_cfg*)) xd3_encode_fgk) +}; +#endif + +#if SECONDARY_DJW +#include "xdelta3-djw.h" +const xd3_sec_type djw_sec_type = +{ + VCD_DJW_ID, + "Static Huffman", + SEC_COUNT_FREQS, + (xd3_sec_stream* (*)(xd3_stream*)) djw_alloc, + (void (*)(xd3_stream*, xd3_sec_stream*)) djw_destroy, + (int (*)(xd3_stream*, xd3_sec_stream*, int)) djw_init, + (int (*)(xd3_stream*, xd3_sec_stream*, const uint8_t**, const uint8_t*, + uint8_t**, const uint8_t*)) xd3_decode_huff, + IF_ENCODER((int (*)(xd3_stream*, xd3_sec_stream*, xd3_output*, + xd3_output*, xd3_sec_cfg*)) xd3_encode_huff) +}; +#endif + +#if SECONDARY_LZMA +#include "xdelta3-lzma.h" +const xd3_sec_type lzma_sec_type = +{ + VCD_LZMA_ID, + "lzma", + SEC_NOFLAGS, + (xd3_sec_stream* (*)(xd3_stream*)) xd3_lzma_alloc, + (void (*)(xd3_stream*, xd3_sec_stream*)) xd3_lzma_destroy, + (int (*)(xd3_stream*, xd3_sec_stream*, int)) xd3_lzma_init, + (int (*)(xd3_stream*, xd3_sec_stream*, const uint8_t**, const uint8_t*, + uint8_t**, const uint8_t*)) xd3_decode_lzma, + IF_ENCODER((int (*)(xd3_stream*, xd3_sec_stream*, xd3_output*, + xd3_output*, xd3_sec_cfg*)) xd3_encode_lzma) +}; +#endif + +#if XD3_MAIN || PYTHON_MODULE || SWIG_MODULE || NOT_MAIN +#include "xdelta3-main.h" +#endif + +#if REGRESSION_TEST +#include "xdelta3-test.h" +#endif + +#endif /* __XDELTA3_C_HEADER_PASS__ */ +#ifdef __XDELTA3_C_INLINE_PASS__ + +const uint16_t __single_hash[256] = +{ + /* Random numbers generated using SLIB's pseudo-random number generator. + * This hashes the input alphabet. */ + 0xbcd1, 0xbb65, 0x42c2, 0xdffe, 0x9666, 0x431b, 0x8504, 0xeb46, + 0x6379, 0xd460, 0xcf14, 0x53cf, 0xdb51, 0xdb08, 0x12c8, 0xf602, + 0xe766, 0x2394, 0x250d, 0xdcbb, 0xa678, 0x02af, 0xa5c6, 0x7ea6, + 0xb645, 0xcb4d, 0xc44b, 0xe5dc, 0x9fe6, 0x5b5c, 0x35f5, 0x701a, + 0x220f, 0x6c38, 0x1a56, 0x4ca3, 0xffc6, 0xb152, 0x8d61, 0x7a58, + 0x9025, 0x8b3d, 0xbf0f, 0x95a3, 0xe5f4, 0xc127, 0x3bed, 0x320b, + 0xb7f3, 0x6054, 0x333c, 0xd383, 0x8154, 0x5242, 0x4e0d, 0x0a94, + 0x7028, 0x8689, 0x3a22, 0x0980, 0x1847, 0xb0f1, 0x9b5c, 0x4176, + 0xb858, 0xd542, 0x1f6c, 0x2497, 0x6a5a, 0x9fa9, 0x8c5a, 0x7743, + 0xa8a9, 0x9a02, 0x4918, 0x438c, 0xc388, 0x9e2b, 0x4cad, 0x01b6, + 0xab19, 0xf777, 0x365f, 0x1eb2, 0x091e, 0x7bf8, 0x7a8e, 0x5227, + 0xeab1, 0x2074, 0x4523, 0xe781, 0x01a3, 0x163d, 0x3b2e, 0x287d, + 0x5e7f, 0xa063, 0xb134, 0x8fae, 0x5e8e, 0xb7b7, 0x4548, 0x1f5a, + 0xfa56, 0x7a24, 0x900f, 0x42dc, 0xcc69, 0x02a0, 0x0b22, 0xdb31, + 0x71fe, 0x0c7d, 0x1732, 0x1159, 0xcb09, 0xe1d2, 0x1351, 0x52e9, + 0xf536, 0x5a4f, 0xc316, 0x6bf9, 0x8994, 0xb774, 0x5f3e, 0xf6d6, + 0x3a61, 0xf82c, 0xcc22, 0x9d06, 0x299c, 0x09e5, 0x1eec, 0x514f, + 0x8d53, 0xa650, 0x5c6e, 0xc577, 0x7958, 0x71ac, 0x8916, 0x9b4f, + 0x2c09, 0x5211, 0xf6d8, 0xcaaa, 0xf7ef, 0x287f, 0x7a94, 0xab49, + 0xfa2c, 0x7222, 0xe457, 0xd71a, 0x00c3, 0x1a76, 0xe98c, 0xc037, + 0x8208, 0x5c2d, 0xdfda, 0xe5f5, 0x0b45, 0x15ce, 0x8a7e, 0xfcad, + 0xaa2d, 0x4b5c, 0xd42e, 0xb251, 0x907e, 0x9a47, 0xc9a6, 0xd93f, + 0x085e, 0x35ce, 0xa153, 0x7e7b, 0x9f0b, 0x25aa, 0x5d9f, 0xc04d, + 0x8a0e, 0x2875, 0x4a1c, 0x295f, 0x1393, 0xf760, 0x9178, 0x0f5b, + 0xfa7d, 0x83b4, 0x2082, 0x721d, 0x6462, 0x0368, 0x67e2, 0x8624, + 0x194d, 0x22f6, 0x78fb, 0x6791, 0xb238, 0xb332, 0x7276, 0xf272, + 0x47ec, 0x4504, 0xa961, 0x9fc8, 0x3fdc, 0xb413, 0x007a, 0x0806, + 0x7458, 0x95c6, 0xccaa, 0x18d6, 0xe2ae, 0x1b06, 0xf3f6, 0x5050, + 0xc8e8, 0xf4ac, 0xc04c, 0xf41c, 0x992f, 0xae44, 0x5f1b, 0x1113, + 0x1738, 0xd9a8, 0x19ea, 0x2d33, 0x9698, 0x2fe9, 0x323f, 0xcde2, + 0x6d71, 0xe37d, 0xb697, 0x2c4f, 0x4373, 0x9102, 0x075d, 0x8e25, + 0x1672, 0xec28, 0x6acb, 0x86cc, 0x186e, 0x9414, 0xd674, 0xd1a5 +}; + +/**************************************************************** + Instruction tables + *****************************************************************/ + +/* The following code implements a parametrized description of the + * code table given above for a few reasons. It is not necessary for + * implementing the standard, to support compression with variable + * tables, so an implementation is only required to know the default + * code table to begin decompression. (If the encoder uses an + * alternate table, the table is included in compressed form inside + * the VCDIFF file.) + * + * Before adding variable-table support there were two functions which + * were hard-coded to the default table above. + * xd3_compute_default_table() would create the default table by + * filling a 256-elt array of xd3_dinst values. The corresponding + * function, xd3_choose_instruction(), would choose an instruction + * based on the hard-coded parameters of the default code table. + * + * Notes: The parametrized code table description here only generates + * tables of a certain regularity similar to the default table by + * allowing to vary the distribution of single- and + * double-instructions and change the number of near and same copy + * modes. More exotic tables are only possible by extending this + * code. + * + * For performance reasons, both the parametrized and non-parametrized + * versions of xd3_choose_instruction remain. The parametrized + * version is only needed for testing multi-table decoding support. + * If ever multi-table encoding is required, this can be optimized by + * compiling static functions for each table. + */ + +/* The XD3_CHOOSE_INSTRUCTION calls xd3_choose_instruction with the + * table description when GENERIC_ENCODE_TABLES are in use. The + * IF_GENCODETBL macro enables generic-code-table specific code. */ +#if GENERIC_ENCODE_TABLES +#define XD3_CHOOSE_INSTRUCTION(stream,prev,inst) xd3_choose_instruction (stream->code_table_desc, prev, inst) +#define IF_GENCODETBL(x) x +#else +#define XD3_CHOOSE_INSTRUCTION(stream,prev,inst) xd3_choose_instruction (prev, inst) +#define IF_GENCODETBL(x) +#endif + +/* This structure maintains information needed by + * xd3_choose_instruction to compute the code for a double instruction + * by first indexing an array of code_table_sizes by copy mode, then + * using (offset + (muliplier * X)) */ +struct _xd3_code_table_sizes { + uint8_t cpy_max; + uint8_t offset; + uint8_t mult; +}; + +/* This contains a complete description of a code table. */ +struct _xd3_code_table_desc +{ + /* Assumes a single RUN instruction */ + /* Assumes that MIN_MATCH is 4 */ + + uint8_t add_sizes; /* Number of immediate-size single adds (default 17) */ + uint8_t near_modes; /* Number of near copy modes (default 4) */ + uint8_t same_modes; /* Number of same copy modes (default 3) */ + uint8_t cpy_sizes; /* Number of immediate-size single copies (default 15) */ + + uint8_t addcopy_add_max; /* Maximum add size for an add-copy double instruction, + all modes (default 4) */ + uint8_t addcopy_near_cpy_max; /* Maximum cpy size for an add-copy double instruction, + up through VCD_NEAR modes (default 6) */ + uint8_t addcopy_same_cpy_max; /* Maximum cpy size for an add-copy double instruction, + VCD_SAME modes (default 4) */ + + uint8_t copyadd_add_max; /* Maximum add size for a copy-add double instruction, + all modes (default 1) */ + uint8_t copyadd_near_cpy_max; /* Maximum cpy size for a copy-add double instruction, + up through VCD_NEAR modes (default 4) */ + uint8_t copyadd_same_cpy_max; /* Maximum cpy size for a copy-add double instruction, + VCD_SAME modes (default 4) */ + + xd3_code_table_sizes addcopy_max_sizes[MAX_MODES]; + xd3_code_table_sizes copyadd_max_sizes[MAX_MODES]; +}; + +/* The rfc3284 code table is represented: */ +static const xd3_code_table_desc __rfc3284_code_table_desc = { + 17, /* add sizes */ + 4, /* near modes */ + 3, /* same modes */ + 15, /* copy sizes */ + + 4, /* add-copy max add */ + 6, /* add-copy max cpy, near */ + 4, /* add-copy max cpy, same */ + + 1, /* copy-add max add */ + 4, /* copy-add max cpy, near */ + 4, /* copy-add max cpy, same */ + + /* addcopy */ + { {6,163,3},{6,175,3},{6,187,3},{6,199,3},{6,211,3},{6,223,3},{4,235,1},{4,239,1},{4,243,1} }, + /* copyadd */ + { {4,247,1},{4,248,1},{4,249,1},{4,250,1},{4,251,1},{4,252,1},{4,253,1},{4,254,1},{4,255,1} }, +}; + +#if GENERIC_ENCODE_TABLES +/* An alternate code table for testing (5 near, 0 same): + * + * TYPE SIZE MODE TYPE SIZE MODE INDEX + * --------------------------------------------------------------- + * 1. Run 0 0 Noop 0 0 0 + * 2. Add 0, [1,23] 0 Noop 0 0 [1,24] + * 3. Copy 0, [4,20] 0 Noop 0 0 [25,42] + * 4. Copy 0, [4,20] 1 Noop 0 0 [43,60] + * 5. Copy 0, [4,20] 2 Noop 0 0 [61,78] + * 6. Copy 0, [4,20] 3 Noop 0 0 [79,96] + * 7. Copy 0, [4,20] 4 Noop 0 0 [97,114] + * 8. Copy 0, [4,20] 5 Noop 0 0 [115,132] + * 9. Copy 0, [4,20] 6 Noop 0 0 [133,150] + * 10. Add [1,4] 0 Copy [4,6] 0 [151,162] + * 11. Add [1,4] 0 Copy [4,6] 1 [163,174] + * 12. Add [1,4] 0 Copy [4,6] 2 [175,186] + * 13. Add [1,4] 0 Copy [4,6] 3 [187,198] + * 14. Add [1,4] 0 Copy [4,6] 4 [199,210] + * 15. Add [1,4] 0 Copy [4,6] 5 [211,222] + * 16. Add [1,4] 0 Copy [4,6] 6 [223,234] + * 17. Copy 4 [0,6] Add [1,3] 0 [235,255] + * --------------------------------------------------------------- */ +static const xd3_code_table_desc __alternate_code_table_desc = { + 23, /* add sizes */ + 5, /* near modes */ + 0, /* same modes */ + 17, /* copy sizes */ + + 4, /* add-copy max add */ + 6, /* add-copy max cpy, near */ + 0, /* add-copy max cpy, same */ + + 3, /* copy-add max add */ + 4, /* copy-add max cpy, near */ + 0, /* copy-add max cpy, same */ + + /* addcopy */ + { {6,151,3},{6,163,3},{6,175,3},{6,187,3},{6,199,3},{6,211,3},{6,223,3},{0,0,0},{0,0,0} }, + /* copyadd */ + { {4,235,1},{4,238,1},{4,241,1},{4,244,1},{4,247,1},{4,250,1},{4,253,1},{0,0,0},{0,0,0} }, +}; +#endif + +/* Computes code table entries of TBL using the specified description. */ +static void +xd3_build_code_table (const xd3_code_table_desc *desc, xd3_dinst *tbl) +{ + usize_t size1, size2, mode; + usize_t cpy_modes = 2 + desc->near_modes + desc->same_modes; + xd3_dinst *d = tbl; + + (d++)->type1 = XD3_RUN; + (d++)->type1 = XD3_ADD; + + for (size1 = 1; size1 <= desc->add_sizes; size1 += 1, d += 1) + { + d->type1 = XD3_ADD; + d->size1 = size1; + } + + for (mode = 0; mode < cpy_modes; mode += 1) + { + (d++)->type1 = XD3_CPY + mode; + + for (size1 = MIN_MATCH; size1 < MIN_MATCH + desc->cpy_sizes; size1 += 1, d += 1) + { + d->type1 = XD3_CPY + mode; + d->size1 = size1; + } + } + + for (mode = 0; mode < cpy_modes; mode += 1) + { + for (size1 = 1; size1 <= desc->addcopy_add_max; size1 += 1) + { + usize_t max = (mode < 2U + desc->near_modes) ? + desc->addcopy_near_cpy_max : + desc->addcopy_same_cpy_max; + + for (size2 = MIN_MATCH; size2 <= max; size2 += 1, d += 1) + { + d->type1 = XD3_ADD; + d->size1 = size1; + d->type2 = XD3_CPY + mode; + d->size2 = size2; + } + } + } + + for (mode = 0; mode < cpy_modes; mode += 1) + { + usize_t max = (mode < 2U + desc->near_modes) ? + desc->copyadd_near_cpy_max : + desc->copyadd_same_cpy_max; + + for (size1 = MIN_MATCH; size1 <= max; size1 += 1) + { + for (size2 = 1; size2 <= desc->copyadd_add_max; size2 += 1, d += 1) + { + d->type1 = XD3_CPY + mode; + d->size1 = size1; + d->type2 = XD3_ADD; + d->size2 = size2; + } + } + } + + XD3_ASSERT (d - tbl == 256); +} + +/* This function generates the static default code table. */ +static const xd3_dinst* +xd3_rfc3284_code_table (void) +{ + static xd3_dinst __rfc3284_code_table[256]; + + if (__rfc3284_code_table[0].type1 != XD3_RUN) + { + xd3_build_code_table (& __rfc3284_code_table_desc, __rfc3284_code_table); + } + + return __rfc3284_code_table; +} + +#if XD3_ENCODER +#if GENERIC_ENCODE_TABLES +/* This function generates the alternate code table. */ +static const xd3_dinst* +xd3_alternate_code_table (void) +{ + static xd3_dinst __alternate_code_table[256]; + + if (__alternate_code_table[0].type1 != XD3_RUN) + { + xd3_build_code_table (& __alternate_code_table_desc, __alternate_code_table); + } + + return __alternate_code_table; +} + +/* This function computes the ideal second instruction INST based on + * preceding instruction PREV. If it is possible to issue a double + * instruction based on this pair it sets PREV->code2, otherwise it + * sets INST->code1. */ +static void +xd3_choose_instruction (const xd3_code_table_desc *desc, xd3_rinst *prev, xd3_rinst *inst) +{ + switch (inst->type) + { + case XD3_RUN: + /* The 0th instruction is RUN */ + inst->code1 = 0; + break; + + case XD3_ADD: + + if (inst->size > desc->add_sizes) + { + /* The first instruction is non-immediate ADD */ + inst->code1 = 1; + } + else + { + /* The following ADD_SIZES instructions are immediate ADDs */ + inst->code1 = 1 + inst->size; + + /* Now check for a possible COPY-ADD double instruction */ + if (prev != NULL) + { + int prev_mode = prev->type - XD3_CPY; + + /* If previous is a copy. Note: as long as the previous + * is not a RUN instruction, it should be a copy because + * it cannot be an add. This check is more clear. */ + if (prev_mode >= 0 && inst->size <= desc->copyadd_add_max) + { + const xd3_code_table_sizes *sizes = + & desc->copyadd_max_sizes[prev_mode]; + + /* This check and the inst->size-<= above are == in + the default table. */ + if (prev->size <= sizes->cpy_max) + { + /* The second and third exprs are 0 in the + default table. */ + prev->code2 = sizes->offset + + (sizes->mult * (prev->size - MIN_MATCH)) + + (inst->size - MIN_ADD); + } + } + } + } + break; + + default: + { + int mode = inst->type - XD3_CPY; + + /* The large copy instruction is offset by the run, large add, + * and immediate adds, then multipled by the number of + * immediate copies plus one (the large copy) (i.e., if there + * are 15 immediate copy instructions then there are 16 copy + * instructions per mode). */ + inst->code1 = 2 + desc->add_sizes + (1 + desc->cpy_sizes) * mode; + + /* Now if the copy is short enough for an immediate instruction. */ + if (inst->size < MIN_MATCH + desc->cpy_sizes && + /* TODO: there needs to be a more comprehensive test for this + * boundary condition, merge is now exercising code in which + * size < MIN_MATCH is possible and it's unclear if the above + * size < (MIN_MATCH + cpy_sizes) should be a <= from inspection + * of the default table version below. */ + inst->size >= MIN_MATCH) + { + inst->code1 += inst->size + 1 - MIN_MATCH; + + /* Now check for a possible ADD-COPY double instruction. */ + if ( (prev != NULL) && + (prev->type == XD3_ADD) && + (prev->size <= desc->addcopy_add_max) ) + { + const xd3_code_table_sizes *sizes = & desc->addcopy_max_sizes[mode]; + + if (inst->size <= sizes->cpy_max) + { + prev->code2 = sizes->offset + + (sizes->mult * (prev->size - MIN_ADD)) + + (inst->size - MIN_MATCH); + } + } + } + } + } +} +#else /* GENERIC_ENCODE_TABLES */ + +/* This version of xd3_choose_instruction is hard-coded for the default + table. */ +static void +xd3_choose_instruction (xd3_rinst *prev, xd3_rinst *inst) +{ + switch (inst->type) + { + case XD3_RUN: + inst->code1 = 0; + break; + + case XD3_ADD: + inst->code1 = 1; + + if (inst->size <= 17) + { + inst->code1 += inst->size; + + if ( (inst->size == 1) && + (prev != NULL) && + (prev->size == 4) && + (prev->type >= XD3_CPY) ) + { + prev->code2 = 247 + (prev->type - XD3_CPY); + } + } + + break; + + default: + { + int mode = inst->type - XD3_CPY; + + XD3_ASSERT (inst->type >= XD3_CPY && inst->type < 12); + + inst->code1 = 19 + 16 * mode; + + if (inst->size <= 18 && inst->size >= 4) + { + inst->code1 += inst->size - 3; + + if ( (prev != NULL) && + (prev->type == XD3_ADD) && + (prev->size <= 4) ) + { + if ( (inst->size <= 6) && + (mode <= 5) ) + { + prev->code2 = 163 + (mode * 12) + (3 * (prev->size - 1)) + (inst->size - 4); + + XD3_ASSERT (prev->code2 <= 234); + } + else if ( (inst->size == 4) && + (mode >= 6) ) + { + prev->code2 = 235 + ((mode - 6) * 4) + (prev->size - 1); + + XD3_ASSERT (prev->code2 <= 246); + } + } + } + + XD3_ASSERT (inst->code1 <= 162); + } + break; + } +} +#endif /* GENERIC_ENCODE_TABLES */ + +/*********************************************************************** + Instruction table encoder/decoder + ***********************************************************************/ + +#if GENERIC_ENCODE_TABLES +#if GENERIC_ENCODE_TABLES_COMPUTE == 0 + +/* In this case, we hard-code the result of + * compute_code_table_encoding for each alternate code table, + * presuming that saves time/space. This has been 131 bytes, but + * secondary compression was turned off. */ +static const uint8_t __alternate_code_table_compressed[178] = +{0xd6,0xc3,0xc4,0x00,0x00,0x01,0x8a,0x6f,0x40,0x81,0x27,0x8c,0x00,0x00,0x4a,0x4a,0x0d,0x02,0x01,0x03, +0x01,0x03,0x00,0x01,0x00,0x00,0x01,0x02,0x03,0x04,0x05,0x06,0x07,0x08,0x09,0x0a,0x0b,0x0c,0x0d,0x0e, +0x0f,0x10,0x11,0x12,0x13,0x14,0x15,0x16,0x17,0x00,0x01,0x01,0x01,0x02,0x02,0x02,0x03,0x03,0x03,0x04, +0x04,0x04,0x04,0x00,0x04,0x05,0x06,0x01,0x02,0x03,0x00,0x01,0x02,0x03,0x04,0x05,0x06,0x05,0x05,0x05, +0x06,0x06,0x06,0x00,0x01,0x02,0x03,0x04,0x05,0x06,0x00,0x02,0x00,0x18,0x13,0x63,0x00,0x1b,0x00,0x54, +0x00,0x15,0x23,0x6f,0x00,0x28,0x13,0x54,0x00,0x15,0x01,0x1a,0x31,0x23,0x6c,0x0d,0x23,0x48,0x00,0x15, +0x93,0x6f,0x00,0x28,0x04,0x23,0x51,0x04,0x32,0x00,0x2b,0x00,0x12,0x00,0x12,0x00,0x12,0x00,0x12,0x00, +0x12,0x00,0x12,0x53,0x57,0x9c,0x07,0x43,0x6f,0x00,0x34,0x00,0x0c,0x00,0x0c,0x00,0x0c,0x00,0x0c,0x00, +0x0c,0x00,0x0c,0x00,0x15,0x00,0x82,0x6f,0x00,0x15,0x12,0x0c,0x00,0x03,0x03,0x00,0x06,0x00,}; + +static int +xd3_compute_alternate_table_encoding (xd3_stream *stream, const uint8_t **data, usize_t *size) +{ + (*data) = __alternate_code_table_compressed; + (*size) = sizeof (__alternate_code_table_compressed); + return 0; +} + +#else + +/* The alternate code table will be computed and stored here. */ +static uint8_t __alternate_code_table_compressed[CODE_TABLE_VCDIFF_SIZE]; +static usize_t __alternate_code_table_compressed_size; + +/* This function generates a delta describing the code table for + * encoding within a VCDIFF file. This function is NOT thread safe + * because it is only intended that this function is used to generate + * statically-compiled strings. "comp_string" must be sized + * CODE_TABLE_VCDIFF_SIZE. */ +int xd3_compute_code_table_encoding (xd3_stream *in_stream, + const xd3_dinst *code_table, + uint8_t *comp_string, + usize_t *comp_string_size) +{ + /* Use DJW secondary compression if it is on by default. This saves + * about 20 bytes. */ + uint8_t dflt_string[CODE_TABLE_STRING_SIZE]; + uint8_t code_string[CODE_TABLE_STRING_SIZE]; + + xd3_compute_code_table_string (xd3_rfc3284_code_table (), dflt_string); + xd3_compute_code_table_string (code_table, code_string); + + return xd3_encode_memory (code_string, CODE_TABLE_STRING_SIZE, + dflt_string, CODE_TABLE_STRING_SIZE, + comp_string, comp_string_size, + CODE_TABLE_VCDIFF_SIZE, + /* flags */ 0); +} + +/* Compute a delta between alternate and rfc3284 tables. As soon as + * another alternate table is added, this code should become generic. + * For now there is only one alternate table for testing. */ +static int +xd3_compute_alternate_table_encoding (xd3_stream *stream, const uint8_t **data, usize_t *size) +{ + int ret; + + if (__alternate_code_table_compressed[0] == 0) + { + if ((ret = xd3_compute_code_table_encoding (stream, xd3_alternate_code_table (), + __alternate_code_table_compressed, + & __alternate_code_table_compressed_size))) + { + return ret; + } + + /* During development of a new code table, enable this variable to print + * the new static contents and determine its size. At run time the + * table will be filled in appropriately, but at least it should have + * the proper size beforehand. */ +#if GENERIC_ENCODE_TABLES_COMPUTE_PRINT + { + int i; + + DP(RINT, "\nstatic const usize_t __alternate_code_table_compressed_size = %u;\n", + __alternate_code_table_compressed_size); + + DP(RINT, "static const uint8_t __alternate_code_table_compressed[%u] =\n{", + __alternate_code_table_compressed_size); + + for (i = 0; i < __alternate_code_table_compressed_size; i += 1) + { + DP(RINT, "0x%02x,", __alternate_code_table_compressed[i]); + if ((i % 20) == 19) { DP(RINT, "\n"); } + } + + DP(RINT, "};\n"); + } +#endif + } + + (*data) = __alternate_code_table_compressed; + (*size) = __alternate_code_table_compressed_size; + + return 0; +} +#endif /* GENERIC_ENCODE_TABLES_COMPUTE != 0 */ +#endif /* GENERIC_ENCODE_TABLES */ + +#endif /* XD3_ENCODER */ + +/* This function generates the 1536-byte string specified in sections 5.4 and + * 7 of rfc3284, which is used to represent a code table within a VCDIFF + * file. */ +void xd3_compute_code_table_string (const xd3_dinst *code_table, uint8_t *str) +{ + int i, s; + + XD3_ASSERT (CODE_TABLE_STRING_SIZE == 6 * 256); + + for (s = 0; s < 6; s += 1) + { + for (i = 0; i < 256; i += 1) + { + switch (s) + { + case 0: *str++ = (code_table[i].type1 >= XD3_CPY ? XD3_CPY : code_table[i].type1); break; + case 1: *str++ = (code_table[i].type2 >= XD3_CPY ? XD3_CPY : code_table[i].type2); break; + case 2: *str++ = (code_table[i].size1); break; + case 3: *str++ = (code_table[i].size2); break; + case 4: *str++ = (code_table[i].type1 >= XD3_CPY ? code_table[i].type1 - XD3_CPY : 0); break; + case 5: *str++ = (code_table[i].type2 >= XD3_CPY ? code_table[i].type2 - XD3_CPY : 0); break; + } + } + } +} + +/* This function translates the code table string into the internal representation. The + * stream's near and same-modes should already be set. */ +static int +xd3_apply_table_string (xd3_stream *stream, const uint8_t *code_string) +{ + int i, s; + int modes = TOTAL_MODES (stream); + xd3_dinst *code_table; + + if ((code_table = stream->code_table_alloc = + (xd3_dinst*) xd3_alloc (stream, + (usize_t) sizeof (xd3_dinst), + 256)) == NULL) + { + return ENOMEM; + } + + for (s = 0; s < 6; s += 1) + { + for (i = 0; i < 256; i += 1) + { + switch (s) + { + case 0: + if (*code_string > XD3_CPY) + { + stream->msg = "invalid code-table opcode"; + return XD3_INTERNAL; + } + code_table[i].type1 = *code_string++; + break; + case 1: + if (*code_string > XD3_CPY) + { + stream->msg = "invalid code-table opcode"; + return XD3_INTERNAL; + } + code_table[i].type2 = *code_string++; + break; + case 2: + if (*code_string != 0 && code_table[i].type1 == XD3_NOOP) + { + stream->msg = "invalid code-table size"; + return XD3_INTERNAL; + } + code_table[i].size1 = *code_string++; + break; + case 3: + if (*code_string != 0 && code_table[i].type2 == XD3_NOOP) + { + stream->msg = "invalid code-table size"; + return XD3_INTERNAL; + } + code_table[i].size2 = *code_string++; + break; + case 4: + if (*code_string >= modes) + { + stream->msg = "invalid code-table mode"; + return XD3_INTERNAL; + } + if (*code_string != 0 && code_table[i].type1 != XD3_CPY) + { + stream->msg = "invalid code-table mode"; + return XD3_INTERNAL; + } + code_table[i].type1 += *code_string++; + break; + case 5: + if (*code_string >= modes) + { + stream->msg = "invalid code-table mode"; + return XD3_INTERNAL; + } + if (*code_string != 0 && code_table[i].type2 != XD3_CPY) + { + stream->msg = "invalid code-table mode"; + return XD3_INTERNAL; + } + code_table[i].type2 += *code_string++; + break; + } + } + } + + stream->code_table = code_table; + return 0; +} + +/* This function applies a code table delta and returns an actual code table. */ +static int +xd3_apply_table_encoding (xd3_stream *in_stream, const uint8_t *data, usize_t size) +{ + uint8_t dflt_string[CODE_TABLE_STRING_SIZE]; + uint8_t code_string[CODE_TABLE_STRING_SIZE]; + usize_t code_size; + int ret; + + xd3_compute_code_table_string (xd3_rfc3284_code_table (), dflt_string); + + if ((ret = xd3_decode_memory (data, size, + dflt_string, CODE_TABLE_STRING_SIZE, + code_string, &code_size, + CODE_TABLE_STRING_SIZE, + 0))) { return ret; } + + if (code_size != sizeof (code_string)) + { + in_stream->msg = "corrupt code-table encoding"; + return XD3_INTERNAL; + } + + return xd3_apply_table_string (in_stream, code_string); +} + +/***********************************************************************/ + +static inline void +xd3_swap_uint8p (uint8_t** p1, uint8_t** p2) +{ + uint8_t *t = (*p1); + (*p1) = (*p2); + (*p2) = t; +} + +static inline void +xd3_swap_usize_t (usize_t* p1, usize_t* p2) +{ + usize_t t = (*p1); + (*p1) = (*p2); + (*p2) = t; +} + +/* It's not constant time, but it computes the log. */ +static int +xd3_check_pow2 (xoff_t value, usize_t *logof) +{ + xoff_t x = 1; + usize_t nolog; + if (logof == NULL) { + logof = &nolog; + } + + *logof = 0; + + for (; x != 0; x <<= 1, *logof += 1) + { + if (x == value) + { + return 0; + } + } + + return XD3_INTERNAL; +} + +static size_t +xd3_pow2_roundup (size_t x) +{ + size_t i = 1; + while (x > i) { + i <<= 1U; + } + return i; +} + +static xoff_t +xd3_xoff_roundup (xoff_t x) +{ + xoff_t i = 1; + while (x > i) { + i <<= 1U; + } + return i; +} + +static usize_t +xd3_round_blksize (usize_t sz, usize_t blksz) +{ + usize_t mod = sz & (blksz-1); + + XD3_ASSERT (xd3_check_pow2 (blksz, NULL) == 0); + + return mod ? (sz + (blksz - mod)) : sz; +} + +/*********************************************************************** + Adler32 stream function: code copied from Zlib, defined in RFC1950 + ***********************************************************************/ + +#define A32_BASE 65521L /* Largest prime smaller than 2^16 */ +#define A32_NMAX 5552 /* NMAX is the largest n such that 255n(n+1)/2 + (n+1)(BASE-1) <= 2^32-1 */ + +#define A32_DO1(buf,i) {s1 += buf[i]; s2 += s1;} +#define A32_DO2(buf,i) A32_DO1(buf,i); A32_DO1(buf,i+1); +#define A32_DO4(buf,i) A32_DO2(buf,i); A32_DO2(buf,i+2); +#define A32_DO8(buf,i) A32_DO4(buf,i); A32_DO4(buf,i+4); +#define A32_DO16(buf) A32_DO8(buf,0); A32_DO8(buf,8); + +static unsigned long adler32 (unsigned long adler, const uint8_t *buf, + usize_t len) +{ + unsigned long s1 = adler & 0xffff; + unsigned long s2 = (adler >> 16) & 0xffff; + int k; + + while (len > 0) + { + k = (len < A32_NMAX) ? len : A32_NMAX; + len -= k; + + while (k >= 16) + { + A32_DO16(buf); + buf += 16; + k -= 16; + } + + if (k != 0) + { + do + { + s1 += *buf++; + s2 += s1; + } + while (--k); + } + + s1 %= A32_BASE; + s2 %= A32_BASE; + } + + return (s2 << 16) | s1; +} + +/*********************************************************************** + Run-length function + ***********************************************************************/ + +#if XD3_ENCODER +static usize_t +xd3_comprun (const uint8_t *seg, usize_t slook, uint8_t *run_cp) +{ + usize_t i; + usize_t run_l = 0; + uint8_t run_c = 0; + + for (i = 0; i < slook; i += 1) + { + NEXTRUN(seg[i]); + } + + (*run_cp) = run_c; + + return run_l; +} +#endif + +/*********************************************************************** + Basic encoder/decoder functions + ***********************************************************************/ + +static inline int +xd3_decode_byte (xd3_stream *stream, usize_t *val) +{ + if (stream->avail_in == 0) + { + stream->msg = "further input required"; + return XD3_INPUT; + } + + (*val) = stream->next_in[0]; + + DECODE_INPUT (1); + return 0; +} + +static inline int +xd3_decode_bytes (xd3_stream *stream, uint8_t *buf, usize_t *pos, usize_t size) +{ + usize_t want; + usize_t take; + + /* Note: The case where (*pos == size) happens when a zero-length + * appheader or code table is transmitted, but there is nothing in + * the standard against that. */ + while (*pos < size) + { + if (stream->avail_in == 0) + { + stream->msg = "further input required"; + return XD3_INPUT; + } + + want = size - *pos; + take = min (want, stream->avail_in); + + memcpy (buf + *pos, stream->next_in, (size_t) take); + + DECODE_INPUT (take); + (*pos) += take; + } + + return 0; +} + +#if XD3_ENCODER +static inline int +xd3_emit_byte (xd3_stream *stream, + xd3_output **outputp, + uint8_t code) +{ + xd3_output *output = (*outputp); + + if (output->next == output->avail) + { + xd3_output *aoutput; + + if ((aoutput = xd3_alloc_output (stream, output)) == NULL) + { + return ENOMEM; + } + + output = (*outputp) = aoutput; + } + + output->base[output->next++] = code; + + return 0; +} + +static inline int +xd3_emit_bytes (xd3_stream *stream, + xd3_output **outputp, + const uint8_t *base, + usize_t size) +{ + xd3_output *output = (*outputp); + + do + { + usize_t take; + + if (output->next == output->avail) + { + xd3_output *aoutput; + + if ((aoutput = xd3_alloc_output (stream, output)) == NULL) + { + return ENOMEM; + } + + output = (*outputp) = aoutput; + } + + take = min (output->avail - output->next, size); + + memcpy (output->base + output->next, base, (size_t) take); + + output->next += take; + size -= take; + base += take; + } + while (size > 0); + + return 0; +} +#endif /* XD3_ENCODER */ + +/********************************************************************* + Integer encoder/decoder functions + **********************************************************************/ + +#define DECODE_INTEGER_TYPE(PART,OFLOW) \ + while (stream->avail_in != 0) \ + { \ + usize_t next = stream->next_in[0]; \ + \ + DECODE_INPUT(1); \ + \ + if (PART & OFLOW) \ + { \ + stream->msg = "overflow in decode_integer"; \ + return XD3_INVALID_INPUT; \ + } \ + \ + PART = (PART << 7) | (next & 127); \ + \ + if ((next & 128) == 0) \ + { \ + (*val) = PART; \ + PART = 0; \ + return 0; \ + } \ + } \ + \ + stream->msg = "further input required"; \ + return XD3_INPUT + +#define READ_INTEGER_TYPE(TYPE, OFLOW) \ + TYPE val = 0; \ + const uint8_t *inp = (*inpp); \ + usize_t next; \ + \ + do \ + { \ + if (inp == max) \ + { \ + stream->msg = "end-of-input in read_integer"; \ + return XD3_INVALID_INPUT; \ + } \ + \ + if (val & OFLOW) \ + { \ + stream->msg = "overflow in read_intger"; \ + return XD3_INVALID_INPUT; \ + } \ + \ + next = (*inp++); \ + val = (val << 7) | (next & 127); \ + } \ + while (next & 128); \ + \ + (*valp) = val; \ + (*inpp) = inp; \ + \ + return 0 + +#define EMIT_INTEGER_TYPE() \ + /* max 64-bit value in base-7 encoding is 9.1 bytes */ \ + uint8_t buf[10]; \ + usize_t bufi = 10; \ + \ + /* This loop performs division and turns on all MSBs. */ \ + do \ + { \ + buf[--bufi] = (num & 127) | 128; \ + num >>= 7U; \ + } \ + while (num != 0); \ + \ + /* Turn off MSB of the last byte. */ \ + buf[9] &= 127; \ + \ + return xd3_emit_bytes (stream, output, buf + bufi, 10 - bufi) + +#define IF_SIZEOF32(x) if (num < (1U << (7 * (x)))) return (x); +#define IF_SIZEOF64(x) if (num < (1ULL << (7 * (x)))) return (x); + +#if USE_UINT32 +static inline uint32_t +xd3_sizeof_uint32_t (uint32_t num) +{ + IF_SIZEOF32(1); + IF_SIZEOF32(2); + IF_SIZEOF32(3); + IF_SIZEOF32(4); + return 5; +} + +static inline int +xd3_decode_uint32_t (xd3_stream *stream, uint32_t *val) +{ DECODE_INTEGER_TYPE (stream->dec_32part, UINT32_OFLOW_MASK); } + +static inline int +xd3_read_uint32_t (xd3_stream *stream, const uint8_t **inpp, + const uint8_t *max, uint32_t *valp) +{ READ_INTEGER_TYPE (uint32_t, UINT32_OFLOW_MASK); } + +#if XD3_ENCODER +static inline int +xd3_emit_uint32_t (xd3_stream *stream, xd3_output **output, uint32_t num) +{ EMIT_INTEGER_TYPE (); } +#endif +#endif + +#if USE_UINT64 +static inline int +xd3_decode_uint64_t (xd3_stream *stream, uint64_t *val) +{ DECODE_INTEGER_TYPE (stream->dec_64part, UINT64_OFLOW_MASK); } + +#if XD3_ENCODER +static inline int +xd3_emit_uint64_t (xd3_stream *stream, xd3_output **output, uint64_t num) +{ EMIT_INTEGER_TYPE (); } +#endif + +/* These are tested but not used */ +#if REGRESSION_TEST +static int +xd3_read_uint64_t (xd3_stream *stream, const uint8_t **inpp, + const uint8_t *max, uint64_t *valp) +{ READ_INTEGER_TYPE (uint64_t, UINT64_OFLOW_MASK); } + +static uint32_t +xd3_sizeof_uint64_t (uint64_t num) +{ + IF_SIZEOF64(1); + IF_SIZEOF64(2); + IF_SIZEOF64(3); + IF_SIZEOF64(4); + IF_SIZEOF64(5); + IF_SIZEOF64(6); + IF_SIZEOF64(7); + IF_SIZEOF64(8); + IF_SIZEOF64(9); + + return 10; +} +#endif + +#endif + +/*********************************************************************** + Address cache stuff + ***********************************************************************/ + +static int +xd3_alloc_cache (xd3_stream *stream) +{ + if (stream->acache.near_array != NULL) + { + xd3_free (stream, stream->acache.near_array); + } + + if (stream->acache.same_array != NULL) + { + xd3_free (stream, stream->acache.same_array); + } + + if (((stream->acache.s_near > 0) && + (stream->acache.near_array = (usize_t*) + xd3_alloc (stream, stream->acache.s_near, + (usize_t) sizeof (usize_t))) + == NULL) || + ((stream->acache.s_same > 0) && + (stream->acache.same_array = (usize_t*) + xd3_alloc (stream, stream->acache.s_same * 256, + (usize_t) sizeof (usize_t))) + == NULL)) + { + return ENOMEM; + } + + return 0; +} + +void +xd3_init_cache (xd3_addr_cache* acache) +{ + if (acache->s_near > 0) + { + memset (acache->near_array, 0, acache->s_near * sizeof (usize_t)); + acache->next_slot = 0; + } + + if (acache->s_same > 0) + { + memset (acache->same_array, 0, acache->s_same * 256 * sizeof (usize_t)); + } +} + +static void +xd3_update_cache (xd3_addr_cache* acache, usize_t addr) +{ + if (acache->s_near > 0) + { + acache->near_array[acache->next_slot] = addr; + acache->next_slot = (acache->next_slot + 1) % acache->s_near; + } + + if (acache->s_same > 0) + { + acache->same_array[addr % (acache->s_same*256)] = addr; + } +} + +#if XD3_ENCODER +/* OPT: this gets called a lot, can it be optimized? */ +static int +xd3_encode_address (xd3_stream *stream, + usize_t addr, + usize_t here, + uint8_t* mode) +{ + usize_t d, bestd; + usize_t i, bestm, ret; + xd3_addr_cache* acache = & stream->acache; + +#define SMALLEST_INT(x) do { if (((x) & ~127U) == 0) { goto good; } } while (0) + + /* Attempt to find the address mode that yields the smallest integer value + * for "d", the encoded address value, thereby minimizing the encoded size + * of the address. */ + bestd = addr; + bestm = VCD_SELF; + + XD3_ASSERT (addr < here); + + SMALLEST_INT (bestd); + + if ((d = here-addr) < bestd) + { + bestd = d; + bestm = VCD_HERE; + + SMALLEST_INT (bestd); + } + + for (i = 0; i < acache->s_near; i += 1) + { + /* Note: If we used signed computation here, we'd could compte d + * and then check (d >= 0 && d < bestd). */ + if (addr >= acache->near_array[i]) + { + d = addr - acache->near_array[i]; + + if (d < bestd) + { + bestd = d; + bestm = i+2; /* 2 counts the VCD_SELF, VCD_HERE modes */ + + SMALLEST_INT (bestd); + } + } + } + + if (acache->s_same > 0 && + acache->same_array[d = addr%(acache->s_same*256)] == addr) + { + bestd = d%256; + /* 2 + s_near offsets past the VCD_NEAR modes */ + bestm = acache->s_near + 2 + d/256; + + if ((ret = xd3_emit_byte (stream, & ADDR_TAIL (stream), bestd))) + { + return ret; + } + } + else + { + good: + + if ((ret = xd3_emit_size (stream, & ADDR_TAIL (stream), bestd))) + { + return ret; + } + } + + xd3_update_cache (acache, addr); + + (*mode) += bestm; + + return 0; +} +#endif + +static int +xd3_decode_address (xd3_stream *stream, usize_t here, + usize_t mode, const uint8_t **inpp, + const uint8_t *max, uint32_t *valp) +{ + int ret; + usize_t same_start = 2 + stream->acache.s_near; + + if (mode < same_start) + { + if ((ret = xd3_read_size (stream, inpp, max, valp))) { return ret; } + + switch (mode) + { + case VCD_SELF: + break; + case VCD_HERE: + (*valp) = here - (*valp); + break; + default: + (*valp) += stream->acache.near_array[mode - 2]; + break; + } + } + else + { + if (*inpp == max) + { + stream->msg = "address underflow"; + return XD3_INVALID_INPUT; + } + + mode -= same_start; + + (*valp) = stream->acache.same_array[mode*256 + (**inpp)]; + + (*inpp) += 1; + } + + xd3_update_cache (& stream->acache, *valp); + + return 0; +} + +/*********************************************************************** + Alloc/free +***********************************************************************/ + +static void* +__xd3_alloc_func (void* opaque, size_t items, usize_t size) +{ + return malloc (items * (size_t) size); +} + +static void +__xd3_free_func (void* opaque, void* address) +{ + free (address); +} + +static void* +xd3_alloc (xd3_stream *stream, + usize_t elts, + usize_t size) +{ + void *a = stream->alloc (stream->opaque, elts, size); + + if (a != NULL) + { + IF_DEBUG (stream->alloc_cnt += 1); + IF_DEBUG2 (DP(RINT "[stream %p malloc] size %u ptr %p\n", + stream, elts * size, a)); + } + else + { + stream->msg = "out of memory"; + } + + return a; +} + +static void +xd3_free (xd3_stream *stream, + void *ptr) +{ + if (ptr != NULL) + { + IF_DEBUG (stream->free_cnt += 1); + XD3_ASSERT (stream->free_cnt <= stream->alloc_cnt); + IF_DEBUG2 (DP(RINT "[stream %p free] %p\n", + stream, ptr)); + stream->free (stream->opaque, ptr); + } +} + +#if XD3_ENCODER +static void* +xd3_alloc0 (xd3_stream *stream, + usize_t elts, + usize_t size) +{ + void *a = xd3_alloc (stream, elts, size); + + if (a != NULL) + { + memset (a, 0, (size_t) (elts * size)); + } + + return a; +} + +static xd3_output* +xd3_alloc_output (xd3_stream *stream, + xd3_output *old_output) +{ + xd3_output *output; + uint8_t *base; + + if (stream->enc_free != NULL) + { + output = stream->enc_free; + stream->enc_free = output->next_page; + } + else + { + if ((output = (xd3_output*) xd3_alloc (stream, 1, + (usize_t) sizeof (xd3_output))) + == NULL) + { + return NULL; + } + + if ((base = (uint8_t*) xd3_alloc (stream, XD3_ALLOCSIZE, + sizeof (uint8_t))) == NULL) + { + xd3_free (stream, output); + return NULL; + } + + output->base = base; + output->avail = XD3_ALLOCSIZE; + } + + output->next = 0; + + if (old_output) + { + old_output->next_page = output; + } + + output->next_page = NULL; + + return output; +} + +static usize_t +xd3_sizeof_output (xd3_output *output) +{ + usize_t s = 0; + + for (; output; output = output->next_page) + { + s += output->next; + } + + return s; +} + +static void +xd3_freelist_output (xd3_stream *stream, + xd3_output *output) +{ + xd3_output *tmp; + + while (output) + { + tmp = output; + output = output->next_page; + + tmp->next = 0; + tmp->next_page = stream->enc_free; + stream->enc_free = tmp; + } +} + +static void +xd3_free_output (xd3_stream *stream, + xd3_output *output) +{ + xd3_output *next; + + again: + if (output == NULL) + { + return; + } + + next = output->next_page; + + xd3_free (stream, output->base); + xd3_free (stream, output); + + output = next; + goto again; +} +#endif /* XD3_ENCODER */ + +void +xd3_free_stream (xd3_stream *stream) +{ + xd3_iopt_buflist *blist = stream->iopt_alloc; + + while (blist != NULL) + { + xd3_iopt_buflist *tmp = blist; + blist = blist->next; + xd3_free (stream, tmp->buffer); + xd3_free (stream, tmp); + } + + xd3_free (stream, stream->large_table); + xd3_free (stream, stream->small_table); + xd3_free (stream, stream->small_prev); + +#if XD3_ENCODER + { + int i; + for (i = 0; i < ENC_SECTS; i += 1) + { + xd3_free_output (stream, stream->enc_heads[i]); + } + xd3_free_output (stream, stream->enc_free); + } +#endif + + xd3_free (stream, stream->acache.near_array); + xd3_free (stream, stream->acache.same_array); + + xd3_free (stream, stream->inst_sect.copied1); + xd3_free (stream, stream->addr_sect.copied1); + xd3_free (stream, stream->data_sect.copied1); + + xd3_free (stream, stream->dec_buffer); + xd3_free (stream, (uint8_t*) stream->dec_lastwin); + + xd3_free (stream, stream->buf_in); + xd3_free (stream, stream->dec_appheader); + xd3_free (stream, stream->dec_codetbl); + xd3_free (stream, stream->code_table_alloc); + +#if SECONDARY_ANY + xd3_free (stream, stream->inst_sect.copied2); + xd3_free (stream, stream->addr_sect.copied2); + xd3_free (stream, stream->data_sect.copied2); + + if (stream->sec_type != NULL) + { + stream->sec_type->destroy (stream, stream->sec_stream_d); + stream->sec_type->destroy (stream, stream->sec_stream_i); + stream->sec_type->destroy (stream, stream->sec_stream_a); + } +#endif + + xd3_free (stream, stream->whole_target.adds); + xd3_free (stream, stream->whole_target.inst); + xd3_free (stream, stream->whole_target.wininfo); + + XD3_ASSERT (stream->alloc_cnt == stream->free_cnt); + + memset (stream, 0, sizeof (xd3_stream)); +} + +#if (XD3_DEBUG > 1 || VCDIFF_TOOLS) +static const char* +xd3_rtype_to_string (xd3_rtype type, int print_mode) +{ + switch (type) + { + case XD3_NOOP: + return "NOOP "; + case XD3_RUN: + return "RUN "; + case XD3_ADD: + return "ADD "; + default: break; + } + if (! print_mode) + { + return "CPY "; + } + switch (type) + { + case XD3_CPY + 0: return "CPY_0"; + case XD3_CPY + 1: return "CPY_1"; + case XD3_CPY + 2: return "CPY_2"; + case XD3_CPY + 3: return "CPY_3"; + case XD3_CPY + 4: return "CPY_4"; + case XD3_CPY + 5: return "CPY_5"; + case XD3_CPY + 6: return "CPY_6"; + case XD3_CPY + 7: return "CPY_7"; + case XD3_CPY + 8: return "CPY_8"; + case XD3_CPY + 9: return "CPY_9"; + default: return "CPY>9"; + } +} +#endif + +/**************************************************************** + Stream configuration + ******************************************************************/ + +int +xd3_config_stream(xd3_stream *stream, + xd3_config *config) +{ + int ret; + xd3_config defcfg; + xd3_smatcher *smatcher = &stream->smatcher; + + if (config == NULL) + { + config = & defcfg; + memset (config, 0, sizeof (*config)); + } + + /* Initial setup: no error checks yet */ + memset (stream, 0, sizeof (*stream)); + + stream->winsize = config->winsize ? config->winsize : XD3_DEFAULT_WINSIZE; + stream->sprevsz = config->sprevsz ? config->sprevsz : XD3_DEFAULT_SPREVSZ; + + if (config->iopt_size == 0) + { + stream->iopt_size = XD3_ALLOCSIZE / sizeof(xd3_rinst); + stream->iopt_unlimited = 1; + } + else + { + stream->iopt_size = config->iopt_size; + } + + stream->getblk = config->getblk; + stream->alloc = config->alloc ? config->alloc : __xd3_alloc_func; + stream->free = config->freef ? config->freef : __xd3_free_func; + stream->opaque = config->opaque; + stream->flags = config->flags; + + /* Secondary setup. */ + stream->sec_data = config->sec_data; + stream->sec_inst = config->sec_inst; + stream->sec_addr = config->sec_addr; + + stream->sec_data.data_type = DATA_SECTION; + stream->sec_inst.data_type = INST_SECTION; + stream->sec_addr.data_type = ADDR_SECTION; + + /* Check static sizes. */ + if (sizeof (usize_t) != SIZEOF_USIZE_T || + sizeof (xoff_t) != SIZEOF_XOFF_T || + (ret = xd3_check_pow2(XD3_ALLOCSIZE, NULL))) + { + stream->msg = "incorrect compilation: wrong integer sizes"; + return XD3_INTERNAL; + } + + /* Check/set secondary compressor. */ + switch (stream->flags & XD3_SEC_TYPE) + { + case 0: + if (stream->flags & XD3_SEC_NOALL) + { + stream->msg = "XD3_SEC flags require a secondary compressor type"; + return XD3_INTERNAL; + } + break; + case XD3_SEC_FGK: + FGK_CASE (stream); + case XD3_SEC_DJW: + DJW_CASE (stream); + case XD3_SEC_LZMA: + LZMA_CASE (stream); + default: + stream->msg = "too many secondary compressor types set"; + return XD3_INTERNAL; + } + + /* Check/set encoder code table. */ + switch (stream->flags & XD3_ALT_CODE_TABLE) { + case 0: + stream->code_table_desc = & __rfc3284_code_table_desc; + stream->code_table_func = xd3_rfc3284_code_table; + break; +#if GENERIC_ENCODE_TABLES + case XD3_ALT_CODE_TABLE: + stream->code_table_desc = & __alternate_code_table_desc; + stream->code_table_func = xd3_alternate_code_table; + stream->comp_table_func = xd3_compute_alternate_table_encoding; + break; +#endif + default: + stream->msg = "alternate code table support was not compiled"; + return XD3_INTERNAL; + } + + /* Check sprevsz */ + if (smatcher->small_chain == 1 && + smatcher->small_lchain == 1) + { + stream->sprevsz = 0; + } + else + { + if ((ret = xd3_check_pow2 (stream->sprevsz, NULL))) + { + stream->msg = "sprevsz is required to be a power of two"; + return XD3_INTERNAL; + } + + stream->sprevmask = stream->sprevsz - 1; + } + + /* Default scanner settings. */ +#if XD3_ENCODER + switch (config->smatch_cfg) + { + IF_BUILD_SOFT(case XD3_SMATCH_SOFT: + { + *smatcher = config->smatcher_soft; + smatcher->string_match = __smatcher_soft.string_match; + smatcher->name = __smatcher_soft.name; + if (smatcher->large_look < MIN_MATCH || + smatcher->large_step < 1 || + smatcher->small_look < MIN_MATCH) + { + stream->msg = "invalid soft string-match config"; + return XD3_INVALID; + } + break; + }) + + IF_BUILD_DEFAULT(case XD3_SMATCH_DEFAULT: + *smatcher = __smatcher_default; + break;) + IF_BUILD_SLOW(case XD3_SMATCH_SLOW: + *smatcher = __smatcher_slow; + break;) + IF_BUILD_FASTEST(case XD3_SMATCH_FASTEST: + *smatcher = __smatcher_fastest; + break;) + IF_BUILD_FASTER(case XD3_SMATCH_FASTER: + *smatcher = __smatcher_faster; + break;) + IF_BUILD_FAST(case XD3_SMATCH_FAST: + *smatcher = __smatcher_fast; + break;) + default: + stream->msg = "invalid string match config type"; + return XD3_INTERNAL; + } + + if (config->smatch_cfg == XD3_SMATCH_DEFAULT && + (stream->flags & XD3_COMPLEVEL_MASK) != 0) + { + int level = (stream->flags & XD3_COMPLEVEL_MASK) >> XD3_COMPLEVEL_SHIFT; + + switch (level) + { + case 1: + IF_BUILD_FASTEST(*smatcher = __smatcher_fastest; + break;) + case 2: + IF_BUILD_FASTER(*smatcher = __smatcher_faster; + break;) + case 3: case 4: case 5: + IF_BUILD_FAST(*smatcher = __smatcher_fast; + break;) + case 6: + IF_BUILD_DEFAULT(*smatcher = __smatcher_default; + break;) + default: + IF_BUILD_SLOW(*smatcher = __smatcher_slow; + break;) + IF_BUILD_DEFAULT(*smatcher = __smatcher_default; + break;) + IF_BUILD_FAST(*smatcher = __smatcher_fast; + break;) + IF_BUILD_FASTER(*smatcher = __smatcher_faster; + break;) + IF_BUILD_FASTEST(*smatcher = __smatcher_fastest; + break;) + } + } +#endif + + return 0; +} + +/*********************************************************** + Getblk interface + ***********************************************************/ + +inline +xoff_t xd3_source_eof(const xd3_source *src) +{ + xoff_t r = (src->blksize * src->max_blkno) + (xoff_t)src->onlastblk; + return r; +} + +inline +usize_t xd3_bytes_on_srcblk (xd3_source *src, xoff_t blkno) +{ + usize_t r = (blkno == src->max_blkno ? + src->onlastblk : + src->blksize); + return r; +} + +/* This function interfaces with the client getblk function, checks + * its results, updates frontier_blkno, max_blkno, onlastblk, eof_known. */ +static int +xd3_getblk (xd3_stream *stream, xoff_t blkno) +{ + int ret; + xd3_source *source = stream->src; + + if (source->curblk == NULL || blkno != source->curblkno) + { + source->getblkno = blkno; + + if (stream->getblk == NULL) + { + stream->msg = "getblk source input"; + return XD3_GETSRCBLK; + } + + ret = stream->getblk (stream, source, blkno); + if (ret != 0) + { + IF_DEBUG1 (DP(RINT "[getblk] app error blkno %"Q": %s\n", + blkno, xd3_strerror (ret))); + return ret; + } + } + + if (blkno >= source->frontier_blkno) + { + if (blkno > source->max_blkno) + { + source->max_blkno = blkno; + source->onlastblk = source->onblk; + } + + if (source->onblk == source->blksize) + { + source->frontier_blkno = blkno + 1; + + IF_DEBUG2 (DP(RINT "[getblk] full source blkno %"Q": " + "source length unknown %"Q"\n", + blkno, + xd3_source_eof (source))); + } + else + { + if (!source->eof_known) + { + IF_DEBUG2 (DP(RINT "[getblk] eof block has %d bytes; " + "source length known %"Q"\n", + xd3_bytes_on_srcblk (source, blkno), + xd3_source_eof (source))); + source->eof_known = 1; + } + + source->frontier_blkno = blkno; + } + } + + XD3_ASSERT (source->curblk != NULL); + IF_DEBUG2 (DP(RINT "[getblk] read source block %"Q" onblk %u blksize %u\n", + blkno, source->onblk, source->blksize)); + + if (blkno == source->max_blkno) + { + /* In case the application sets the source as 1 block w/ a + preset buffer. */ + source->onlastblk = source->onblk; + + if (source->onblk == source->blksize) + { + source->frontier_blkno = blkno + 1; + } + } + return 0; +} + +/*********************************************************** + Stream open/close + ***************************************************************/ + +int +xd3_set_source (xd3_stream *stream, + xd3_source *src) +{ + usize_t shiftby; + + stream->src = src; + src->srclen = 0; + src->srcbase = 0; + + /* Enforce power-of-two blocksize so that source-block number + * calculations are cheap. */ + if (xd3_check_pow2 (src->blksize, &shiftby) != 0) + { + src->blksize = xd3_pow2_roundup(src->blksize); + xd3_check_pow2 (src->blksize, &shiftby); + IF_DEBUG1 (DP(RINT "raising src_blksz to %u\n", src->blksize)); + } + + src->shiftby = shiftby; + src->maskby = (1 << shiftby) - 1; + + if (xd3_check_pow2 (src->max_winsize, NULL) != 0) + { + src->max_winsize = xd3_xoff_roundup(src->max_winsize); + IF_DEBUG1 (DP(RINT "raising src_maxsize to %u\n", src->blksize)); + } + src->max_winsize = max(src->max_winsize, XD3_ALLOCSIZE); + + return 0; +} + +int +xd3_set_source_and_size (xd3_stream *stream, + xd3_source *user_source, + xoff_t source_size) { + int ret = xd3_set_source (stream, user_source); + if (ret == 0) + { + stream->src->eof_known = 1; + IF_DEBUG2 (DP(RINT "[set source] size known %"Q"\n", + source_size)); + + xd3_blksize_div(source_size, + stream->src, + &stream->src->max_blkno, + &stream->src->onlastblk); + } + return ret; +} + +void +xd3_abort_stream (xd3_stream *stream) +{ + stream->dec_state = DEC_ABORTED; + stream->enc_state = ENC_ABORTED; +} + +int +xd3_close_stream (xd3_stream *stream) +{ + if (stream->enc_state != 0 && stream->enc_state != ENC_ABORTED) + { + if (stream->buf_leftover != NULL) + { + stream->msg = "encoding is incomplete"; + return XD3_INTERNAL; + } + + if (stream->enc_state == ENC_POSTWIN) + { +#if XD3_ENCODER + xd3_encode_reset (stream); +#endif + stream->current_window += 1; + stream->enc_state = ENC_INPUT; + } + + /* If encoding, should be ready for more input but not actually + have any. */ + if (stream->enc_state != ENC_INPUT || stream->avail_in != 0) + { + stream->msg = "encoding is incomplete"; + return XD3_INTERNAL; + } + } + else + { + switch (stream->dec_state) + { + case DEC_VCHEAD: + case DEC_WININD: + /* TODO: Address the zero-byte ambiguity. Does the encoder + * emit a window or not? If so, then catch an error here. + * If not, need another routine to say + * decode_at_least_one_if_empty. */ + case DEC_ABORTED: + break; + default: + /* If decoding, should be ready for the next window. */ + stream->msg = "EOF in decode"; + return XD3_INTERNAL; + } + } + + return 0; +} + +/************************************************************** + Application header + ****************************************************************/ + +int +xd3_get_appheader (xd3_stream *stream, + uint8_t **data, + usize_t *size) +{ + if (stream->dec_state < DEC_WININD) + { + stream->msg = "application header not available"; + return XD3_INTERNAL; + } + + (*data) = stream->dec_appheader; + (*size) = stream->dec_appheadsz; + return 0; +} + +/********************************************************** + Decoder stuff + *************************************************/ + +#include "xdelta3-decode.h" + +/**************************************************************** + Encoder stuff + *****************************************************************/ + +#if XD3_ENCODER +void +xd3_set_appheader (xd3_stream *stream, + const uint8_t *data, + usize_t size) +{ + stream->enc_appheader = data; + stream->enc_appheadsz = size; +} + +#if XD3_DEBUG +static int +xd3_iopt_check (xd3_stream *stream) +{ + usize_t ul = xd3_rlist_length (& stream->iopt_used); + usize_t fl = xd3_rlist_length (& stream->iopt_free); + + return (ul + fl + (stream->iout ? 1 : 0)) == stream->iopt_size; +} +#endif + +static xd3_rinst* +xd3_iopt_free (xd3_stream *stream, xd3_rinst *i) +{ + xd3_rinst *n = xd3_rlist_remove (i); + xd3_rlist_push_back (& stream->iopt_free, i); + return n; +} + +static void +xd3_iopt_free_nonadd (xd3_stream *stream, xd3_rinst *i) +{ + if (i->type != XD3_ADD) + { + xd3_rlist_push_back (& stream->iopt_free, i); + } +} + +/* When an instruction is ready to flush from the iopt buffer, this + * function is called to produce an encoding. It writes the + * instruction plus size, address, and data to the various encoding + * sections. */ +static int +xd3_iopt_finish_encoding (xd3_stream *stream, xd3_rinst *inst) +{ + int ret; + + /* Check for input overflow. */ + XD3_ASSERT (inst->pos + inst->size <= stream->avail_in); + + switch (inst->type) + { + case XD3_CPY: + { + /* the address may have an offset if there is a source window. */ + usize_t addr; + xd3_source *src = stream->src; + + if (src != NULL) + { + /* If there is a source copy, the source must have its + * source window decided before we can encode. This can + * be bad -- we have to make this decision even if no + * source matches have been found. */ + if (stream->srcwin_decided == 0) + { + if ((ret = xd3_srcwin_setup (stream))) { return ret; } + } + else + { + stream->srcwin_decided_early = (!stream->src->eof_known || + (stream->srcwin_cksum_pos < + xd3_source_eof (stream->src))); + } + + /* xtra field indicates the copy is from the source */ + if (inst->xtra) + { + XD3_ASSERT (inst->addr >= src->srcbase); + XD3_ASSERT (inst->addr + inst->size <= + src->srcbase + src->srclen); + addr = (usize_t)(inst->addr - src->srcbase); + stream->n_scpy += 1; + stream->l_scpy += (xoff_t) inst->size; + } + else + { + /* with source window: target copy address is offset + * by taroff. */ + addr = stream->taroff + (usize_t) inst->addr; + stream->n_tcpy += 1; + stream->l_tcpy += (xoff_t) inst->size; + } + } + else + { + addr = (usize_t) inst->addr; + stream->n_tcpy += 1; + stream->l_tcpy += inst->size; + } + + /* Note: used to assert inst->size >= MIN_MATCH, but not true + * for merge operations & identical match heuristics. */ + /* the "here" position is always offset by taroff */ + if ((ret = xd3_encode_address (stream, addr, inst->pos + stream->taroff, + & inst->type))) + { + return ret; + } + + IF_DEBUG2 ({ + static int cnt; + DP(RINT "[iopt copy:%d] pos %"Q"-%"Q" addr %"Q"-%"Q" size %u\n", + cnt++, + stream->total_in + inst->pos, + stream->total_in + inst->pos + inst->size, + inst->addr, inst->addr + inst->size, inst->size); + }); + break; + } + case XD3_RUN: + { + XD3_ASSERT (inst->size >= MIN_MATCH); + + if ((ret = xd3_emit_byte (stream, & DATA_TAIL (stream), inst->xtra))) { return ret; } + + stream->n_run += 1; + stream->l_run += inst->size; + + IF_DEBUG2 ({ + static int cnt; + DP(RINT "[iopt run:%d] pos %"Q" size %u\n", cnt++, stream->total_in + inst->pos, inst->size); + }); + break; + } + case XD3_ADD: + { + if ((ret = xd3_emit_bytes (stream, & DATA_TAIL (stream), + stream->next_in + inst->pos, inst->size))) { return ret; } + + stream->n_add += 1; + stream->l_add += inst->size; + + IF_DEBUG2 ({ + static int cnt; + DP(RINT "[iopt add:%d] pos %"Q" size %u\n", cnt++, stream->total_in + inst->pos, inst->size); + }); + + break; + } + } + + /* This is the only place stream->unencoded_offset is incremented. */ + XD3_ASSERT (stream->unencoded_offset == inst->pos); + stream->unencoded_offset += inst->size; + + inst->code2 = 0; + + XD3_CHOOSE_INSTRUCTION (stream, stream->iout, inst); + + if (stream->iout != NULL) + { + if (stream->iout->code2 != 0) + { + if ((ret = xd3_emit_double (stream, stream->iout, inst, stream->iout->code2))) { return ret; } + + xd3_iopt_free_nonadd (stream, stream->iout); + xd3_iopt_free_nonadd (stream, inst); + stream->iout = NULL; + return 0; + } + else + { + if ((ret = xd3_emit_single (stream, stream->iout, stream->iout->code1))) { return ret; } + + xd3_iopt_free_nonadd (stream, stream->iout); + } + } + + stream->iout = inst; + + return 0; +} + +/* This possibly encodes an add instruction, iadd, which must remain + * on the stack until the following call to + * xd3_iopt_finish_encoding. */ +static int +xd3_iopt_add (xd3_stream *stream, usize_t pos, xd3_rinst *iadd) +{ + int ret; + usize_t off = stream->unencoded_offset; + + if (pos > off) + { + iadd->type = XD3_ADD; + iadd->pos = off; + iadd->size = pos - off; + + if ((ret = xd3_iopt_finish_encoding (stream, iadd))) { return ret; } + } + + return 0; +} + +/* This function calls xd3_iopt_finish_encoding to finish encoding an + * instruction, and it may also produce an add instruction for an + * unmatched region. */ +static int +xd3_iopt_add_encoding (xd3_stream *stream, xd3_rinst *inst) +{ + int ret; + xd3_rinst iadd; + + if ((ret = xd3_iopt_add (stream, inst->pos, & iadd))) { return ret; } + + if ((ret = xd3_iopt_finish_encoding (stream, inst))) { return ret; } + + return 0; +} + +/* Generates a final add instruction to encode the remaining input. */ +static int +xd3_iopt_add_finalize (xd3_stream *stream) +{ + int ret; + xd3_rinst iadd; + + if ((ret = xd3_iopt_add (stream, stream->avail_in, & iadd))) { return ret; } + + if (stream->iout) + { + if ((ret = xd3_emit_single (stream, stream->iout, stream->iout->code1))) { return ret; } + + xd3_iopt_free_nonadd (stream, stream->iout); + stream->iout = NULL; + } + + return 0; +} + +/* Compact the instruction buffer by choosing the best non-overlapping + * instructions when lazy string-matching. There are no ADDs in the + * iopt buffer because those are synthesized in xd3_iopt_add_encoding + * and during xd3_iopt_add_finalize. */ +static int +xd3_iopt_flush_instructions (xd3_stream *stream, int force) +{ + xd3_rinst *r1 = xd3_rlist_front (& stream->iopt_used); + xd3_rinst *r2; + xd3_rinst *r3; + usize_t r1end; + usize_t r2end; + usize_t r2off; + usize_t r2moff; + usize_t gap; + usize_t flushed; + int ret; + + XD3_ASSERT (xd3_iopt_check (stream)); + + /* Note: once tried to skip this step if it's possible to assert + * there are no overlapping instructions. Doesn't work because + * xd3_opt_erase leaves overlapping instructions. */ + while (! xd3_rlist_end (& stream->iopt_used, r1) && + ! xd3_rlist_end (& stream->iopt_used, r2 = xd3_rlist_next (r1))) + { + r1end = r1->pos + r1->size; + + /* If the instructions do not overlap, continue. */ + if (r1end <= r2->pos) + { + r1 = r2; + continue; + } + + r2end = r2->pos + r2->size; + + /* The min_match adjustments prevent this. */ + XD3_ASSERT (r2end > (r1end + LEAST_MATCH_INCR)); + + /* If r3 is available... */ + if (! xd3_rlist_end (& stream->iopt_used, r3 = xd3_rlist_next (r2))) + { + /* If r3 starts before r1 finishes or just about, r2 is irrelevant */ + if (r3->pos <= r1end + 1) + { + xd3_iopt_free (stream, r2); + continue; + } + } + else if (! force) + { + /* Unless force, end the loop when r3 is not available. */ + break; + } + + r2off = r2->pos - r1->pos; + r2moff = r2end - r1end; + gap = r2end - r1->pos; + + /* If the two matches overlap almost entirely, choose the better match + * and discard the other. The else branch can still create inefficient + * copies, e.g., a 4-byte copy that takes 4 bytes to encode, which + * xd3_smatch() wouldn't allow by its crude efficiency check. However, + * in this case there are adjacent copies which mean the add would cost + * one extra byte. Allow the inefficiency here. */ + if (gap < 2*MIN_MATCH || r2moff <= 2 || r2off <= 2) + { + /* Only one match should be used, choose the longer one. */ + if (r1->size < r2->size) + { + xd3_iopt_free (stream, r1); + r1 = r2; + } + else + { + /* We are guaranteed that r1 does not overlap now, so advance past r2 */ + r1 = xd3_iopt_free (stream, r2); + } + continue; + } + else + { + /* Shorten one of the instructions -- could be optimized + * based on the address cache. */ + usize_t average; + usize_t newsize; + usize_t adjust1; + + XD3_ASSERT (r1end > r2->pos && r2end > r1->pos); + + /* Try to balance the length of both instructions, but avoid + * making both longer than MAX_MATCH_SPLIT . */ + average = gap / 2; + newsize = min (MAX_MATCH_SPLIT, gap - average); + + /* Should be possible to simplify this code. */ + if (newsize > r1->size) + { + /* shorten r2 */ + adjust1 = r1end - r2->pos; + } + else if (newsize > r2->size) + { + /* shorten r1 */ + adjust1 = r1end - r2->pos; + + XD3_ASSERT (r1->size > adjust1); + + r1->size -= adjust1; + + /* don't shorten r2 */ + adjust1 = 0; + } + else + { + /* shorten r1 */ + adjust1 = r1->size - newsize; + + if (r2->pos > r1end - adjust1) + { + adjust1 -= r2->pos - (r1end - adjust1); + } + + XD3_ASSERT (r1->size > adjust1); + + r1->size -= adjust1; + + /* shorten r2 */ + XD3_ASSERT (r1->pos + r1->size >= r2->pos); + + adjust1 = r1->pos + r1->size - r2->pos; + } + + /* Fallthrough above if-else, shorten r2 */ + XD3_ASSERT (r2->size > adjust1); + + r2->size -= adjust1; + r2->pos += adjust1; + r2->addr += adjust1; + + XD3_ASSERT (r1->size >= MIN_MATCH); + XD3_ASSERT (r2->size >= MIN_MATCH); + + r1 = r2; + } + } + + XD3_ASSERT (xd3_iopt_check (stream)); + + /* If forcing, pick instructions until the list is empty, otherwise + * this empties 50% of the queue. */ + for (flushed = 0; ! xd3_rlist_empty (& stream->iopt_used); ) + { + xd3_rinst *renc = xd3_rlist_pop_front (& stream->iopt_used); + if ((ret = xd3_iopt_add_encoding (stream, renc))) + { + return ret; + } + + if (! force) + { + if (++flushed > stream->iopt_size / 2) + { + break; + } + + /* If there are only two instructions remaining, break, + * because they were not optimized. This means there were + * more than 50% eliminated by the loop above. */ + r1 = xd3_rlist_front (& stream->iopt_used); + if (xd3_rlist_end(& stream->iopt_used, r1) || + xd3_rlist_end(& stream->iopt_used, r2 = xd3_rlist_next (r1)) || + xd3_rlist_end(& stream->iopt_used, r3 = xd3_rlist_next (r2))) + { + break; + } + } + } + + XD3_ASSERT (xd3_iopt_check (stream)); + + XD3_ASSERT (!force || xd3_rlist_length (& stream->iopt_used) == 0); + + return 0; +} + +static int +xd3_iopt_get_slot (xd3_stream *stream, xd3_rinst** iptr) +{ + xd3_rinst *i; + int ret; + + if (xd3_rlist_empty (& stream->iopt_free)) + { + if (stream->iopt_unlimited) + { + usize_t elts = XD3_ALLOCSIZE / sizeof(xd3_rinst); + + if ((ret = xd3_alloc_iopt (stream, elts))) + { + return ret; + } + + stream->iopt_size += elts; + } + else + { + if ((ret = xd3_iopt_flush_instructions (stream, 0))) { return ret; } + + XD3_ASSERT (! xd3_rlist_empty (& stream->iopt_free)); + } + } + + i = xd3_rlist_pop_back (& stream->iopt_free); + + xd3_rlist_push_back (& stream->iopt_used, i); + + (*iptr) = i; + + ++stream->i_slots_used; + + return 0; +} + +/* A copy is about to be emitted that extends backwards to POS, + * therefore it may completely cover some existing instructions in the + * buffer. If an instruction is completely covered by this new match, + * erase it. If the new instruction is covered by the previous one, + * return 1 to skip it. */ +static void +xd3_iopt_erase (xd3_stream *stream, usize_t pos, usize_t size) +{ + while (! xd3_rlist_empty (& stream->iopt_used)) + { + xd3_rinst *r = xd3_rlist_back (& stream->iopt_used); + + /* Verify that greedy is working. The previous instruction + * should end before the new one begins. */ + XD3_ASSERT ((stream->flags & XD3_BEGREEDY) == 0 || (r->pos + r->size <= pos)); + /* Verify that min_match is working. The previous instruction + * should end before the new one ends. */ + XD3_ASSERT ((stream->flags & XD3_BEGREEDY) != 0 || (r->pos + r->size < pos + size)); + + /* See if the last instruction starts before the new + * instruction. If so, there is nothing to erase. */ + if (r->pos < pos) + { + return; + } + + /* Otherwise, the new instruction covers the old one, delete it + and repeat. */ + xd3_rlist_remove (r); + xd3_rlist_push_back (& stream->iopt_free, r); + --stream->i_slots_used; + } +} + +/* This function tells the last matched input position. */ +static usize_t +xd3_iopt_last_matched (xd3_stream *stream) +{ + xd3_rinst *r; + + if (xd3_rlist_empty (& stream->iopt_used)) + { + return 0; + } + + r = xd3_rlist_back (& stream->iopt_used); + + return r->pos + r->size; +} + +/********************************************************* + Emit routines + ***********************************************************/ + +static int +xd3_emit_single (xd3_stream *stream, xd3_rinst *single, usize_t code) +{ + int has_size = stream->code_table[code].size1 == 0; + int ret; + + IF_DEBUG2 (DP(RINT "[emit1] %u %s (%u) code %u\n", + single->pos, + xd3_rtype_to_string ((xd3_rtype) single->type, 0), + single->size, + code)); + + if ((ret = xd3_emit_byte (stream, & INST_TAIL (stream), code))) + { + return ret; + } + + if (has_size) + { + if ((ret = xd3_emit_size (stream, & INST_TAIL (stream), single->size))) + { + return ret; + } + } + + return 0; +} + +static int +xd3_emit_double (xd3_stream *stream, xd3_rinst *first, + xd3_rinst *second, usize_t code) +{ + int ret; + + /* All double instructions use fixed sizes, so all we need to do is + * output the instruction code, no sizes. */ + XD3_ASSERT (stream->code_table[code].size1 != 0 && + stream->code_table[code].size2 != 0); + + if ((ret = xd3_emit_byte (stream, & INST_TAIL (stream), code))) + { + return ret; + } + + IF_DEBUG2 (DP(RINT "[emit2]: %u %s (%u) %s (%u) code %u\n", + first->pos, + xd3_rtype_to_string ((xd3_rtype) first->type, 0), + first->size, + xd3_rtype_to_string ((xd3_rtype) second->type, 0), + second->size, + code)); + + return 0; +} + +/* This enters a potential run instruction into the iopt buffer. The + * position argument is relative to the target window. */ +static int +xd3_emit_run (xd3_stream *stream, usize_t pos, usize_t size, uint8_t *run_c) +{ + xd3_rinst* ri; + int ret; + + if ((ret = xd3_iopt_get_slot (stream, & ri))) { return ret; } + + ri->type = XD3_RUN; + ri->xtra = *run_c; + ri->pos = pos; + ri->size = size; + + return 0; +} + +/* This enters a potential copy instruction into the iopt buffer. The + * position argument is relative to the target window.. */ +int +xd3_found_match (xd3_stream *stream, usize_t pos, + usize_t size, xoff_t addr, int is_source) +{ + xd3_rinst* ri; + int ret; + + if ((ret = xd3_iopt_get_slot (stream, & ri))) { return ret; } + + ri->type = XD3_CPY; + ri->xtra = is_source; + ri->pos = pos; + ri->size = size; + ri->addr = addr; + + return 0; +} + +static int +xd3_emit_hdr (xd3_stream *stream) +{ + int ret; + int use_secondary = stream->sec_type != NULL; + int use_adler32 = stream->flags & (XD3_ADLER32 | XD3_ADLER32_RECODE); + int vcd_source = xd3_encoder_used_source (stream); + usize_t win_ind = 0; + usize_t del_ind = 0; + usize_t enc_len; + usize_t tgt_len; + usize_t data_len; + usize_t inst_len; + usize_t addr_len; + + if (stream->current_window == 0) + { + usize_t hdr_ind = 0; + int use_appheader = stream->enc_appheader != NULL; + int use_gencodetbl = GENERIC_ENCODE_TABLES && + (stream->code_table_desc != & __rfc3284_code_table_desc); + + if (use_secondary) { hdr_ind |= VCD_SECONDARY; } + if (use_gencodetbl) { hdr_ind |= VCD_CODETABLE; } + if (use_appheader) { hdr_ind |= VCD_APPHEADER; } + + if ((ret = xd3_emit_byte (stream, & HDR_TAIL (stream), + VCDIFF_MAGIC1)) != 0 || + (ret = xd3_emit_byte (stream, & HDR_TAIL (stream), + VCDIFF_MAGIC2)) != 0 || + (ret = xd3_emit_byte (stream, & HDR_TAIL (stream), + VCDIFF_MAGIC3)) != 0 || + (ret = xd3_emit_byte (stream, & HDR_TAIL (stream), + VCDIFF_VERSION)) != 0 || + (ret = xd3_emit_byte (stream, & HDR_TAIL (stream), hdr_ind)) != 0) + { + return ret; + } + + /* Secondary compressor ID */ +#if SECONDARY_ANY + if (use_secondary && + (ret = xd3_emit_byte (stream, & HDR_TAIL (stream), + stream->sec_type->id))) + { + return ret; + } +#endif + + /* Compressed code table */ + if (use_gencodetbl) + { + usize_t code_table_size; + const uint8_t *code_table_data; + + if ((ret = stream->comp_table_func (stream, & code_table_data, + & code_table_size))) + { + return ret; + } + + if ((ret = xd3_emit_size (stream, & HDR_TAIL (stream), + code_table_size + 2)) || + (ret = xd3_emit_byte (stream, & HDR_TAIL (stream), + stream->code_table_desc->near_modes)) || + (ret = xd3_emit_byte (stream, & HDR_TAIL (stream), + stream->code_table_desc->same_modes)) || + (ret = xd3_emit_bytes (stream, & HDR_TAIL (stream), + code_table_data, code_table_size))) + { + return ret; + } + } + + /* Application header */ + if (use_appheader) + { + if ((ret = xd3_emit_size (stream, & HDR_TAIL (stream), + stream->enc_appheadsz)) || + (ret = xd3_emit_bytes (stream, & HDR_TAIL (stream), + stream->enc_appheader, + stream->enc_appheadsz))) + { + return ret; + } + } + } + + /* try to compress this window */ +#if SECONDARY_ANY + if (use_secondary) + { + int data_sec = 0; + int inst_sec = 0; + int addr_sec = 0; + +# define ENCODE_SECONDARY_SECTION(UPPER,LOWER) \ + ((stream->flags & XD3_SEC_NO ## UPPER) == 0 && \ + (ret = xd3_encode_secondary (stream, \ + & UPPER ## _HEAD (stream), \ + & UPPER ## _TAIL (stream), \ + & xd3_sec_ ## LOWER (stream), \ + & stream->sec_ ## LOWER, \ + & LOWER ## _sec))) + + if (ENCODE_SECONDARY_SECTION (DATA, data) || + ENCODE_SECONDARY_SECTION (INST, inst) || + ENCODE_SECONDARY_SECTION (ADDR, addr)) + { + return ret; + } + + del_ind |= (data_sec ? VCD_DATACOMP : 0); + del_ind |= (inst_sec ? VCD_INSTCOMP : 0); + del_ind |= (addr_sec ? VCD_ADDRCOMP : 0); + } +#endif + + /* if (vcd_target) { win_ind |= VCD_TARGET; } */ + if (vcd_source) { win_ind |= VCD_SOURCE; } + if (use_adler32) { win_ind |= VCD_ADLER32; } + + /* window indicator */ + if ((ret = xd3_emit_byte (stream, & HDR_TAIL (stream), win_ind))) + { + return ret; + } + + /* source window */ + if (vcd_source) + { + /* or (vcd_target) { ... } */ + if ((ret = xd3_emit_size (stream, & HDR_TAIL (stream), + stream->src->srclen)) || + (ret = xd3_emit_offset (stream, & HDR_TAIL (stream), + stream->src->srcbase))) { return ret; } + } + + tgt_len = stream->avail_in; + data_len = xd3_sizeof_output (DATA_HEAD (stream)); + inst_len = xd3_sizeof_output (INST_HEAD (stream)); + addr_len = xd3_sizeof_output (ADDR_HEAD (stream)); + + /* The enc_len field is a redundency for future extensions.*/ + enc_len = (1 + (xd3_sizeof_size (tgt_len) + + xd3_sizeof_size (data_len) + + xd3_sizeof_size (inst_len) + + xd3_sizeof_size (addr_len)) + + data_len + + inst_len + + addr_len + + (use_adler32 ? 4 : 0)); + + if ((ret = xd3_emit_size (stream, & HDR_TAIL (stream), enc_len)) || + (ret = xd3_emit_size (stream, & HDR_TAIL (stream), tgt_len)) || + (ret = xd3_emit_byte (stream, & HDR_TAIL (stream), del_ind)) || + (ret = xd3_emit_size (stream, & HDR_TAIL (stream), data_len)) || + (ret = xd3_emit_size (stream, & HDR_TAIL (stream), inst_len)) || + (ret = xd3_emit_size (stream, & HDR_TAIL (stream), addr_len))) + { + return ret; + } + + if (use_adler32) + { + uint8_t send[4]; + uint32_t a32; + + if (stream->flags & XD3_ADLER32) + { + a32 = adler32 (1L, stream->next_in, stream->avail_in); + } + else + { + a32 = stream->recode_adler32; + } + + /* Four bytes. */ + send[0] = (uint8_t) (a32 >> 24); + send[1] = (uint8_t) (a32 >> 16); + send[2] = (uint8_t) (a32 >> 8); + send[3] = (uint8_t) (a32 & 0x000000FFU); + + if ((ret = xd3_emit_bytes (stream, & HDR_TAIL (stream), send, 4))) + { + return ret; + } + } + + return 0; +} + +/**************************************************************** + Encode routines + ****************************************************************/ + +static int +xd3_encode_buffer_leftover (xd3_stream *stream) +{ + usize_t take; + usize_t room; + + /* Allocate the buffer. */ + if (stream->buf_in == NULL && + (stream->buf_in = (uint8_t*) xd3_alloc (stream, stream->winsize, 1)) == NULL) + { + return ENOMEM; + } + + IF_DEBUG2 (DP(RINT "[leftover] flush?=%s\n", (stream->flags & XD3_FLUSH) ? "yes" : "no")); + + /* Take leftover input first. */ + if (stream->buf_leftover != NULL) + { + XD3_ASSERT (stream->buf_avail == 0); + XD3_ASSERT (stream->buf_leftavail < stream->winsize); + + IF_DEBUG2 (DP(RINT "[leftover] previous %u avail %u\n", stream->buf_leftavail, stream->avail_in)); + + memcpy (stream->buf_in, stream->buf_leftover, stream->buf_leftavail); + + stream->buf_leftover = NULL; + stream->buf_avail = stream->buf_leftavail; + } + + /* Copy into the buffer. */ + room = stream->winsize - stream->buf_avail; + take = min (room, stream->avail_in); + + memcpy (stream->buf_in + stream->buf_avail, stream->next_in, take); + + stream->buf_avail += take; + + if (take < stream->avail_in) + { + /* Buffer is full */ + stream->buf_leftover = stream->next_in + take; + stream->buf_leftavail = stream->avail_in - take; + } + else if ((stream->buf_avail < stream->winsize) && !(stream->flags & XD3_FLUSH)) + { + /* Buffer has space */ + IF_DEBUG2 (DP(RINT "[leftover] emptied %u\n", take)); + return XD3_INPUT; + } + + /* Use the buffer: */ + IF_DEBUG2 (DP(RINT "[leftover] take %u remaining %u\n", take, stream->buf_leftavail)); + stream->next_in = stream->buf_in; + stream->avail_in = stream->buf_avail; + stream->buf_avail = 0; + + return 0; +} + +/* Allocates one block of xd3_rlist elements */ +static int +xd3_alloc_iopt (xd3_stream *stream, usize_t elts) +{ + usize_t i; + xd3_iopt_buflist* last = + (xd3_iopt_buflist*) xd3_alloc (stream, sizeof (xd3_iopt_buflist), 1); + + if (last == NULL || + (last->buffer = (xd3_rinst*) xd3_alloc (stream, sizeof (xd3_rinst), elts)) == NULL) + { + return ENOMEM; + } + + last->next = stream->iopt_alloc; + stream->iopt_alloc = last; + + for (i = 0; i < elts; i += 1) + { + xd3_rlist_push_back (& stream->iopt_free, & last->buffer[i]); + } + + return 0; +} + +/* This function allocates all memory initially used by the encoder. */ +static int +xd3_encode_init (xd3_stream *stream, int full_init) +{ + int i; + + if (full_init) + { + int large_comp = (stream->src != NULL); + int small_comp = ! (stream->flags & XD3_NOCOMPRESS); + + /* Memory allocations for checksum tables are delayed until + * xd3_string_match_init in the first call to string_match--that way + * identical or short inputs require no table allocation. */ + if (large_comp) + { + usize_t hash_values = (stream->src->max_winsize / + stream->smatcher.large_step); + + xd3_size_hashtable (stream, + hash_values, + & stream->large_hash); + } + + if (small_comp) + { + /* TODO: This is under devel: used to have min(sprevsz) here, which sort + * of makes sense, but observed fast performance w/ larger tables, which + * also sort of makes sense. @@@ */ + usize_t hash_values = stream->winsize; + + xd3_size_hashtable (stream, + hash_values, + & stream->small_hash); + } + } + + /* data buffers */ + for (i = 0; i < ENC_SECTS; i += 1) + { + if ((stream->enc_heads[i] = + stream->enc_tails[i] = + xd3_alloc_output (stream, NULL)) == NULL) + { + return ENOMEM; + } + } + + /* iopt buffer */ + xd3_rlist_init (& stream->iopt_used); + xd3_rlist_init (& stream->iopt_free); + + if (xd3_alloc_iopt (stream, stream->iopt_size) != 0) { goto fail; } + + XD3_ASSERT (xd3_rlist_length (& stream->iopt_free) == stream->iopt_size); + XD3_ASSERT (xd3_rlist_length (& stream->iopt_used) == 0); + + /* address cache, code table */ + stream->acache.s_near = stream->code_table_desc->near_modes; + stream->acache.s_same = stream->code_table_desc->same_modes; + stream->code_table = stream->code_table_func (); + + return xd3_alloc_cache (stream); + + fail: + + return ENOMEM; +} + +int +xd3_encode_init_full (xd3_stream *stream) +{ + return xd3_encode_init (stream, 1); +} + +int +xd3_encode_init_partial (xd3_stream *stream) +{ + return xd3_encode_init (stream, 0); +} + +/* Called after the ENC_POSTOUT state, this puts the output buffers + * back into separate lists and re-initializes some variables. (The + * output lists were spliced together during the ENC_FLUSH state.) */ +static void +xd3_encode_reset (xd3_stream *stream) +{ + int i; + xd3_output *olist; + + stream->avail_in = 0; + stream->small_reset = 1; + stream->i_slots_used = 0; + + if (stream->src != NULL) + { + stream->src->srcbase = 0; + stream->src->srclen = 0; + stream->srcwin_decided = 0; + stream->srcwin_decided_early = 0; + stream->match_minaddr = 0; + stream->match_maxaddr = 0; + stream->taroff = 0; + } + + /* Reset output chains. */ + olist = stream->enc_heads[0]; + + for (i = 0; i < ENC_SECTS; i += 1) + { + XD3_ASSERT (olist != NULL); + + stream->enc_heads[i] = olist; + stream->enc_tails[i] = olist; + olist = olist->next_page; + + stream->enc_heads[i]->next = 0; + stream->enc_heads[i]->next_page = NULL; + + stream->enc_tails[i]->next_page = NULL; + stream->enc_tails[i] = stream->enc_heads[i]; + } + + xd3_freelist_output (stream, olist); +} + +/* The main encoding routine. */ +int +xd3_encode_input (xd3_stream *stream) +{ + int ret, i; + + if (stream->dec_state != 0) + { + stream->msg = "encoder/decoder transition"; + return XD3_INTERNAL; + } + + switch (stream->enc_state) + { + case ENC_INIT: + /* Only reached on first time through: memory setup. */ + if ((ret = xd3_encode_init_full (stream))) { return ret; } + + stream->enc_state = ENC_INPUT; + + case ENC_INPUT: + + /* If there is no input yet, just return. This checks for + * next_in == NULL, not avail_in == 0 since zero bytes is a + * valid input. There is an assertion in xd3_avail_input() that + * next_in != NULL for this reason. By returning right away we + * avoid creating an input buffer before the caller has supplied + * its first data. It is possible for xd3_avail_input to be + * called both before and after the first call to + * xd3_encode_input(). */ + if (stream->next_in == NULL) + { + return XD3_INPUT; + } + + enc_flush: + /* See if we should buffer the input: either if there is already + * a leftover buffer, or if the input is short of winsize + * without flush. The label at this point is reached by a goto + * below, when there is leftover input after postout. */ + if ((stream->buf_leftover != NULL) || + (stream->buf_avail != 0) || + (stream->avail_in < stream->winsize && ! (stream->flags & XD3_FLUSH))) + { + if ((ret = xd3_encode_buffer_leftover (stream))) { return ret; } + } + + /* Initalize the address cache before each window. */ + xd3_init_cache (& stream->acache); + + stream->input_position = 0; + stream->min_match = MIN_MATCH; + stream->unencoded_offset = 0; + + stream->enc_state = ENC_SEARCH; + + IF_DEBUG2 (DP(RINT "[WINSTART:%"Q"] input bytes %u offset %"Q"\n", + stream->current_window, stream->avail_in, + stream->total_in)); + return XD3_WINSTART; + + case ENC_SEARCH: + IF_DEBUG2 (DP(RINT "[SEARCH] match_state %d avail_in %u %s\n", + stream->match_state, stream->avail_in, + stream->src ? "source" : "no source")); + + /* Reentrant matching. */ + if (stream->src != NULL) + { + switch (stream->match_state) + { + case MATCH_TARGET: + /* Try matching forward at the start of the target. + * This is entered the first time through, to check for + * a perfect match, and whenever there is a source match + * that extends to the end of the previous window. The + * match_srcpos field is initially zero and later set + * during xd3_source_extend_match. */ + + if (stream->avail_in > 0) + { + /* This call can't fail because the source window is + * unrestricted. */ + ret = xd3_source_match_setup (stream, stream->match_srcpos); + XD3_ASSERT (ret == 0); + stream->match_state = MATCH_FORWARD; + } + else + { + stream->match_state = MATCH_SEARCHING; + stream->match_fwd = 0; + } + XD3_ASSERT (stream->match_fwd == 0); + + case MATCH_FORWARD: + case MATCH_BACKWARD: + if (stream->avail_in != 0) + { + if ((ret = xd3_source_extend_match (stream)) != 0) + { + return ret; + } + + /* The search has to make forward progress here + * or else it can get stuck in a match-backward + * (getsrcblk) then match-forward (getsrcblk), + * find insufficient match length, then repeat + * exactly the same search. + */ + stream->input_position += stream->match_fwd; + } + + case MATCH_SEARCHING: + /* Continue string matching. (It's possible that the + * initial match continued through the entire input, in + * which case we're still in MATCH_FORWARD and should + * remain so for the next input window.) */ + break; + } + } + + /* String matching... */ + if (stream->avail_in != 0 && + (ret = stream->smatcher.string_match (stream))) + { + return ret; + } + + stream->enc_state = ENC_INSTR; + + case ENC_INSTR: + /* Note: Jump here to encode VCDIFF deltas w/o using this + * string-matching code. Merging code code enters here. */ + + /* Flush the instrution buffer, then possibly add one more + * instruction, then emit the header. */ + if ((ret = xd3_iopt_flush_instructions (stream, 1)) || + (ret = xd3_iopt_add_finalize (stream))) + { + return ret; + } + + stream->enc_state = ENC_FLUSH; + + case ENC_FLUSH: + /* Note: main_recode_func() bypasses string-matching by setting + * ENC_FLUSH. */ + if ((ret = xd3_emit_hdr (stream))) + { + return ret; + } + + /* Begin output. */ + stream->enc_current = HDR_HEAD (stream); + + /* Chain all the outputs together. After doing this, it looks + * as if there is only one section. The other enc_heads are set + * to NULL to avoid freeing them more than once. */ + for (i = 1; i < ENC_SECTS; i += 1) + { + stream->enc_tails[i-1]->next_page = stream->enc_heads[i]; + stream->enc_heads[i] = NULL; + } + + enc_output: + + stream->enc_state = ENC_POSTOUT; + stream->next_out = stream->enc_current->base; + stream->avail_out = stream->enc_current->next; + stream->total_out += (xoff_t) stream->avail_out; + + /* If there is any output in this buffer, return it, otherwise + * fall through to handle the next buffer or finish the window + * after all buffers have been output. */ + if (stream->avail_out > 0) + { + /* This is the only place xd3_encode returns XD3_OUTPUT */ + return XD3_OUTPUT; + } + + case ENC_POSTOUT: + + if (stream->avail_out != 0) + { + stream->msg = "missed call to consume output"; + return XD3_INTERNAL; + } + + /* Continue outputting one buffer at a time, until the next is NULL. */ + if ((stream->enc_current = stream->enc_current->next_page) != NULL) + { + goto enc_output; + } + + stream->total_in += (xoff_t) stream->avail_in; + stream->enc_state = ENC_POSTWIN; + + IF_DEBUG2 (DP(RINT "[WINFINISH:%"Q"] in=%"Q"\n", + stream->current_window, + stream->total_in)); + return XD3_WINFINISH; + + case ENC_POSTWIN: + + xd3_encode_reset (stream); + + stream->current_window += 1; + stream->enc_state = ENC_INPUT; + + /* If there is leftover input to flush, repeat. */ + if (stream->buf_leftover != NULL) + { + goto enc_flush; + } + + /* Ready for more input. */ + return XD3_INPUT; + + default: + stream->msg = "invalid state"; + return XD3_INTERNAL; + } +} +#endif /* XD3_ENCODER */ + +/***************************************************************** + Client convenience functions + ******************************************************************/ + +static int +xd3_process_stream (int is_encode, + xd3_stream *stream, + int (*func) (xd3_stream *), + int close_stream, + const uint8_t *input, + usize_t input_size, + uint8_t *output, + usize_t *output_size, + usize_t output_size_max) +{ + usize_t ipos = 0; + usize_t n = min(stream->winsize, input_size); + + (*output_size) = 0; + + stream->flags |= XD3_FLUSH; + + xd3_avail_input (stream, input + ipos, n); + ipos += n; + + for (;;) + { + int ret; + switch((ret = func (stream))) + { + case XD3_OUTPUT: { /* memcpy below */ break; } + case XD3_INPUT: { + n = min(stream->winsize, input_size - ipos); + if (n == 0) { + goto done; + } + xd3_avail_input (stream, input + ipos, n); + ipos += n; + continue; + } + case XD3_GOTHEADER: { /* ignore */ continue; } + case XD3_WINSTART: { /* ignore */ continue; } + case XD3_WINFINISH: { /* ignore */ continue; } + case XD3_GETSRCBLK: + { + stream->msg = "stream requires source input"; + return XD3_INTERNAL; + } + case 0: + { + /* xd3_encode_input/xd3_decode_input never return 0 */ + stream->msg = "invalid return: 0"; + return XD3_INTERNAL; + } + default: + return ret; + } + + if (*output_size + stream->avail_out > output_size_max) + { + stream->msg = "insufficient output space"; + return ENOSPC; + } + + memcpy (output + *output_size, stream->next_out, stream->avail_out); + + *output_size += stream->avail_out; + + xd3_consume_output (stream); + } + done: + return (close_stream == 0) ? 0 : xd3_close_stream (stream); +} + +static int +xd3_process_memory (int is_encode, + int (*func) (xd3_stream *), + int close_stream, + const uint8_t *input, + usize_t input_size, + const uint8_t *source, + usize_t source_size, + uint8_t *output, + usize_t *output_size, + usize_t output_size_max, + int flags) { + xd3_stream stream; + xd3_config config; + xd3_source src; + int ret; + + memset (& stream, 0, sizeof (stream)); + memset (& config, 0, sizeof (config)); + + if (input == NULL || output == NULL) { + stream.msg = "invalid input/output buffer"; + ret = XD3_INTERNAL; + goto exit; + } + + config.flags = flags; + + if (is_encode) + { + config.winsize = min(input_size, (usize_t) XD3_DEFAULT_WINSIZE); + config.iopt_size = min(input_size / 32, XD3_DEFAULT_IOPT_SIZE); + config.iopt_size = max(config.iopt_size, 128U); + config.sprevsz = xd3_pow2_roundup (config.winsize); + } + + if ((ret = xd3_config_stream (&stream, &config)) != 0) + { + goto exit; + } + + if (source != NULL) + { + memset (& src, 0, sizeof (src)); + + src.blksize = source_size; + src.onblk = source_size; + src.curblk = source; + src.curblkno = 0; + src.max_winsize = source_size; + + if ((ret = xd3_set_source_and_size (&stream, &src, source_size)) != 0) + { + goto exit; + } + } + + if ((ret = xd3_process_stream (is_encode, + & stream, + func, 1, + input, input_size, + output, + output_size, + output_size_max)) != 0) + { + goto exit; + } + + exit: + if (ret != 0) + { + IF_DEBUG2 (DP(RINT "process_memory: %d: %s\n", ret, stream.msg)); + } + xd3_free_stream(&stream); + return ret; +} + +int +xd3_decode_stream (xd3_stream *stream, + const uint8_t *input, + usize_t input_size, + uint8_t *output, + usize_t *output_size, + usize_t output_size_max) +{ + return xd3_process_stream (0, stream, & xd3_decode_input, 1, + input, input_size, + output, output_size, output_size_max); +} + +int +xd3_decode_memory (const uint8_t *input, + usize_t input_size, + const uint8_t *source, + usize_t source_size, + uint8_t *output, + usize_t *output_size, + usize_t output_size_max, + int flags) { + return xd3_process_memory (0, & xd3_decode_input, 1, + input, input_size, + source, source_size, + output, output_size, output_size_max, + flags); +} + + +#if XD3_ENCODER +int +xd3_encode_stream (xd3_stream *stream, + const uint8_t *input, + usize_t input_size, + uint8_t *output, + usize_t *output_size, + usize_t output_size_max) +{ + return xd3_process_stream (1, stream, & xd3_encode_input, 1, + input, input_size, + output, output_size, output_size_max); +} + +int +xd3_encode_memory (const uint8_t *input, + usize_t input_size, + const uint8_t *source, + usize_t source_size, + uint8_t *output, + usize_t *output_size, + usize_t output_size_max, + int flags) { + return xd3_process_memory (1, & xd3_encode_input, 1, + input, input_size, + source, source_size, + output, output_size, output_size_max, + flags); +} +#endif + + +/************************************************************* + String matching helpers + *************************************************************/ + +#if XD3_ENCODER +/* Do the initial xd3_string_match() checksum table setup. + * Allocations are delayed until first use to avoid allocation + * sometimes (e.g., perfect matches, zero-length inputs). */ +static int +xd3_string_match_init (xd3_stream *stream) +{ + const int DO_SMALL = ! (stream->flags & XD3_NOCOMPRESS); + const int DO_LARGE = (stream->src != NULL); + + if (DO_LARGE && stream->large_table == NULL) + { + if ((stream->large_table = + (usize_t*) xd3_alloc0 (stream, stream->large_hash.size, sizeof (usize_t))) == NULL) + { + return ENOMEM; + } + } + + if (DO_SMALL) + { + /* Subsequent calls can return immediately after checking reset. */ + if (stream->small_table != NULL) + { + /* The target hash table is reinitialized once per window. */ + /* TODO: This would not have to be reinitialized if absolute + * offsets were being stored. */ + if (stream->small_reset) + { + stream->small_reset = 0; + memset (stream->small_table, 0, + sizeof (usize_t) * stream->small_hash.size); + } + + return 0; + } + + if ((stream->small_table = + (usize_t*) xd3_alloc0 (stream, + stream->small_hash.size, + sizeof (usize_t))) == NULL) + { + return ENOMEM; + } + + /* If there is a previous table needed. */ + if (stream->smatcher.small_lchain > 1 || + stream->smatcher.small_chain > 1) + { + if ((stream->small_prev = + (xd3_slist*) xd3_alloc (stream, + stream->sprevsz, + sizeof (xd3_slist))) == NULL) + { + return ENOMEM; + } + } + } + + return 0; +} + +#if XD3_USE_LARGEFILE64 +/* This function handles the 32/64bit ambiguity -- file positions are 64bit + * but the hash table for source-offsets is 32bit. */ +static xoff_t +xd3_source_cksum_offset(xd3_stream *stream, usize_t low) +{ + xoff_t scp = stream->srcwin_cksum_pos; + xoff_t s0 = scp >> 32; + + usize_t sr = (usize_t) scp; + + if (s0 == 0) { + return low; + } + + /* This should not be >= because srcwin_cksum_pos is the next + * position to index. */ + if (low > sr) { + return (--s0 << 32) | low; + } + + return (s0 << 32) | low; +} +#else +static xoff_t +xd3_source_cksum_offset(xd3_stream *stream, usize_t low) +{ + return (xoff_t) low; +} +#endif + +/* This function sets up the stream->src fields srcbase, srclen. The + * call is delayed until these values are needed to encode a copy + * address. At this point the decision has to be made. */ +static int +xd3_srcwin_setup (xd3_stream *stream) +{ + xd3_source *src = stream->src; + xoff_t length, x; + + /* Check the undecided state. */ + XD3_ASSERT (src->srclen == 0 && src->srcbase == 0); + + /* Avoid repeating this call. */ + stream->srcwin_decided = 1; + + /* If the stream is flushing, then the iopt buffer was able to + * contain the complete encoding. If no copies were issued no + * source window is actually needed. This prevents the VCDIFF + * header from including source base/len. xd3_emit_hdr checks for + * srclen == 0. */ + if (stream->enc_state == ENC_INSTR && stream->match_maxaddr == 0) + { + goto done; + } + + /* Check for overflow, srclen is usize_t - this can't happen unless + * XD3_DEFAULT_SRCBACK and related parameters are extreme - should + * use smaller windows. */ + length = stream->match_maxaddr - stream->match_minaddr; + + x = (xoff_t) USIZE_T_MAX; + if (length > x) + { + stream->msg = "source window length overflow (not 64bit)"; + return XD3_INTERNAL; + } + + /* If ENC_INSTR, then we know the exact source window to use because + * no more copies can be issued. */ + if (stream->enc_state == ENC_INSTR) + { + src->srcbase = stream->match_minaddr; + src->srclen = (usize_t) length; + XD3_ASSERT (src->srclen); + goto done; + } + + /* Otherwise, we have to make a guess. More copies may still be + * issued, but we have to decide the source window base and length + * now. */ + src->srcbase = stream->match_minaddr; + src->srclen = max ((usize_t) length, + stream->avail_in + (stream->avail_in >> 2)); + + /* OPT: If we know the source size, it might be possible to reduce + * srclen. */ + XD3_ASSERT (src->srclen); + done: + /* Set the taroff. This convenience variable is used even when + stream->src == NULL. */ + stream->taroff = src->srclen; + return 0; +} + +/* Sets the bounding region for a newly discovered source match, prior + * to calling xd3_source_extend_match(). This sets the match_maxfwd, + * match_maxback variables. Note: srcpos is an absolute position + * (xoff_t) but the match_maxfwd, match_maxback variables are usize_t. + * Returns 0 if the setup succeeds, or 1 if the source position lies + * outside an already-decided srcbase/srclen window. */ +static int +xd3_source_match_setup (xd3_stream *stream, xoff_t srcpos) +{ + xd3_source *src = stream->src; + usize_t greedy_or_not; + xoff_t frontier_pos; + + stream->match_maxback = 0; + stream->match_maxfwd = 0; + stream->match_back = 0; + stream->match_fwd = 0; + + /* This avoids a non-blocking endless loop caused by scanning + * backwards across a block boundary, only to find not enough + * matching bytes to beat the current min_match due to a better lazy + * target match: the re-entry to xd3_string_match() repeats the same + * long match because the input position hasn't changed. TODO: if + * ever duplicates are added to the source hash table, this logic + * won't suffice to avoid loops. See testing/regtest.cc's + * TestNonBlockingProgress test! */ + if (srcpos != 0 && srcpos == stream->match_last_srcpos) + { + IF_DEBUG2(DP(RINT "[match_setup] looping failure\n")); + goto bad; + } + + /* Implement src->max_winsize, which prevents the encoder from seeking + * back further than the LRU cache maintaining FIFO discipline, (to + * avoid seeking). */ + frontier_pos = stream->src->frontier_blkno * stream->src->blksize; + IF_DEBUG2(DP(RINT "[match_setup] frontier_pos %"Q", srcpos %"Q", " + "src->max_winsize %"Q"\n", + frontier_pos, srcpos, stream->src->max_winsize)); + if (srcpos < frontier_pos && + frontier_pos - srcpos > stream->src->max_winsize) { + IF_DEBUG1(DP(RINT "[match_setup] rejected due to src->max_winsize " + "distance eof=%"Q" srcpos=%"Q" maxsz=%"Q"\n", + xd3_source_eof (stream->src), + srcpos, stream->src->max_winsize)); + goto bad; + } + + /* Going backwards, the 1.5-pass algorithm allows some + * already-matched input may be covered by a longer source match. + * The greedy algorithm does not allow this. */ + if (stream->flags & XD3_BEGREEDY) + { + /* The greedy algorithm allows backward matching to the last + matched position. */ + greedy_or_not = xd3_iopt_last_matched (stream); + } + else + { + /* The 1.5-pass algorithm allows backward matching to go back as + * far as the unencoded offset, which is updated as instructions + * pass out of the iopt buffer. If this (default) is chosen, it + * means xd3_iopt_erase may be called to eliminate instructions + * when a covering source match is found. */ + greedy_or_not = stream->unencoded_offset; + } + + /* Backward target match limit. */ + XD3_ASSERT (stream->input_position >= greedy_or_not); + stream->match_maxback = stream->input_position - greedy_or_not; + + /* Forward target match limit. */ + XD3_ASSERT (stream->avail_in > stream->input_position); + stream->match_maxfwd = stream->avail_in - stream->input_position; + + /* Now we take the source position into account. It depends whether + * the srclen/srcbase have been decided yet. */ + if (stream->srcwin_decided == 0) + { + /* Unrestricted case: the match can cover the entire source, + * 0--src->size. We compare the usize_t + * match_maxfwd/match_maxback against the xoff_t + * src->size/srcpos values and take the min. */ + if (srcpos < (xoff_t) stream->match_maxback) + { + stream->match_maxback = (usize_t) srcpos; + } + + if (stream->src->eof_known) + { + xoff_t srcavail = xd3_source_eof (stream->src) - srcpos; + + if (srcavail < (xoff_t) stream->match_maxfwd) + { + stream->match_maxfwd = (usize_t) srcavail; + } + } + + IF_DEBUG2(DP(RINT + "[match_setup] srcpos %"Q" (tgtpos %"Q") " + "unrestricted maxback %u maxfwd %u\n", + srcpos, + stream->total_in + stream->input_position, + stream->match_maxback, + stream->match_maxfwd)); + goto good; + } + + /* Decided some source window. */ + XD3_ASSERT (src->srclen > 0); + + /* Restricted case: fail if the srcpos lies outside the source window */ + if ((srcpos < src->srcbase) || + (srcpos > (src->srcbase + (xoff_t) src->srclen))) + { + IF_DEBUG1(DP(RINT "[match_setup] restricted source window failure\n")); + goto bad; + } + else + { + usize_t srcavail; + + srcavail = (usize_t) (srcpos - src->srcbase); + if (srcavail < stream->match_maxback) + { + stream->match_maxback = srcavail; + } + + srcavail = (usize_t) (src->srcbase + (xoff_t) src->srclen - srcpos); + if (srcavail < stream->match_maxfwd) + { + stream->match_maxfwd = srcavail; + } + + IF_DEBUG1(DP(RINT + "[match_setup] srcpos %"Q" (tgtpos %"Q") " + "restricted maxback %u maxfwd %u\n", + srcpos, + stream->total_in + stream->input_position, + stream->match_maxback, + stream->match_maxfwd)); + goto good; + } + + good: + stream->match_state = MATCH_BACKWARD; + stream->match_srcpos = srcpos; + stream->match_last_srcpos = srcpos; + return 0; + + bad: + stream->match_state = MATCH_SEARCHING; + return 1; +} + +static inline int +xd3_forward_match(const uint8_t *s1c, const uint8_t *s2c, int n) +{ + int i = 0; +#if UNALIGNED_OK + int nint = n / sizeof(int); + + if (nint >> 3) + { + int j = 0; + const int *s1 = (const int*)s1c; + const int *s2 = (const int*)s2c; + int nint_8 = nint - 8; + + while (i <= nint_8 && + s1[i++] == s2[j++] && + s1[i++] == s2[j++] && + s1[i++] == s2[j++] && + s1[i++] == s2[j++] && + s1[i++] == s2[j++] && + s1[i++] == s2[j++] && + s1[i++] == s2[j++] && + s1[i++] == s2[j++]) { } + + i = (i - 1) * sizeof(int); + } +#endif + + while (i < n && s1c[i] == s2c[i]) + { + i++; + } + return i; +} + +/* This function expands the source match backward and forward. It is + * reentrant, since xd3_getblk may return XD3_GETSRCBLK, so most + * variables are kept in xd3_stream. There are two callers of this + * function, the string_matching routine when a checksum match is + * discovered, and xd3_encode_input whenever a continuing (or initial) + * match is suspected. The two callers do different things with the + * input_position, thus this function leaves that variable untouched. + * If a match is taken the resulting stream->match_fwd is left + * non-zero. */ +static int +xd3_source_extend_match (xd3_stream *stream) +{ + int ret; + xd3_source *src = stream->src; + xoff_t matchoff; /* matchoff is the current right/left-boundary of + the source match being tested. */ + usize_t streamoff; /* streamoff is the current right/left-boundary + of the input match being tested. */ + xoff_t tryblk; /* tryblk, tryoff are the block, offset position + of matchoff */ + usize_t tryoff; + usize_t tryrem; /* tryrem is the number of matchable bytes */ + usize_t matched; + + IF_DEBUG2(DP(RINT "[extend match] srcpos %"Q"\n", + stream->match_srcpos)); + + XD3_ASSERT (src != NULL); + + /* Does it make sense to compute backward match AFTER forward match? */ + if (stream->match_state == MATCH_BACKWARD) + { + /* Note: this code is practically duplicated below, substituting + * match_fwd/match_back and direction. TODO: Consolidate? */ + matchoff = stream->match_srcpos - stream->match_back; + streamoff = stream->input_position - stream->match_back; + xd3_blksize_div (matchoff, src, &tryblk, &tryoff); + + /* this loops backward over source blocks */ + while (stream->match_back < stream->match_maxback) + { + /* see if we're backing across a source block boundary */ + if (tryoff == 0) + { + tryoff = src->blksize; + tryblk -= 1; + } + + if ((ret = xd3_getblk (stream, tryblk))) + { + /* if search went too far back, continue forward. */ + if (ret == XD3_TOOFARBACK) + { + break; + } + + /* could be a XD3_GETSRCBLK failure. */ + return ret; + } + + tryrem = min (tryoff, stream->match_maxback - stream->match_back); + + IF_DEBUG2(DP(RINT "[maxback] maxback %u trysrc %"Q"/%u tgt %u tryrem %u\n", + stream->match_maxback, tryblk, tryoff, streamoff, tryrem)); + + /* TODO: This code can be optimized similar to xd3_match_forward() */ + for (; tryrem != 0; tryrem -= 1, stream->match_back += 1) + { + if (src->curblk[tryoff-1] != stream->next_in[streamoff-1]) + { + goto doneback; + } + + tryoff -= 1; + streamoff -= 1; + } + } + + doneback: + stream->match_state = MATCH_FORWARD; + } + + XD3_ASSERT (stream->match_state == MATCH_FORWARD); + + matchoff = stream->match_srcpos + stream->match_fwd; + streamoff = stream->input_position + stream->match_fwd; + xd3_blksize_div (matchoff, src, & tryblk, & tryoff); + + /* Note: practically the same code as backwards case above: same comments */ + while (stream->match_fwd < stream->match_maxfwd) + { + if (tryoff == src->blksize) + { + tryoff = 0; + tryblk += 1; + } + + if ((ret = xd3_getblk (stream, tryblk))) + { + /* if search went too far back, continue forward. */ + if (ret == XD3_TOOFARBACK) + { + break; + } + + /* could be a XD3_GETSRCBLK failure. */ + return ret; + } + + tryrem = min(stream->match_maxfwd - stream->match_fwd, + src->onblk - tryoff); + + if (tryrem == 0) + { + /* Generally, this means we have a power-of-two size source + * and we just found the end-of-file, in this case it's an + * empty block. */ + XD3_ASSERT (src->onblk < src->blksize); + break; + } + + matched = xd3_forward_match(src->curblk + tryoff, + stream->next_in + streamoff, + tryrem); + tryoff += matched; + streamoff += matched; + stream->match_fwd += matched; + + if (tryrem != matched) + { + break; + } + } + + stream->match_state = MATCH_SEARCHING; + + /* If the match ends short of the last instruction end, we probably + * don't want it. There is the possibility that a copy ends short + * of the last copy but also goes further back, in which case we + * might want it. This code does not implement such: if so we would + * need more complicated xd3_iopt_erase logic. */ + if (stream->match_fwd < stream->min_match) + { + stream->match_fwd = 0; + } + else + { + usize_t total = stream->match_fwd + stream->match_back; + + /* Correct the variables to remove match_back from the equation. */ + usize_t target_position = stream->input_position - stream->match_back; + usize_t match_length = stream->match_back + stream->match_fwd; + xoff_t match_position = stream->match_srcpos - stream->match_back; + xoff_t match_end = stream->match_srcpos + stream->match_fwd; + + /* At this point we may have to erase any iopt-buffer + * instructions that are fully covered by a backward-extending + * copy. */ + if (stream->match_back > 0) + { + xd3_iopt_erase (stream, target_position, total); + } + + stream->match_back = 0; + + /* Update ranges. The first source match occurs with both + values set to 0. */ + if (stream->match_maxaddr == 0 || + match_position < stream->match_minaddr) + { + stream->match_minaddr = match_position; + } + + if (match_end > stream->match_maxaddr) + { + /* Note: per-window */ + stream->match_maxaddr = match_end; + } + + if (match_end > stream->maxsrcaddr) + { + /* Note: across windows */ + stream->maxsrcaddr = match_end; + } + + IF_DEBUG2 ({ + static int x = 0; + DP(RINT "[source match:%d] (%s) [ %u bytes ]\n", + x++, + stream->total_in + target_position, + stream->total_in + target_position + match_length, + match_position, + match_position + match_length, + (stream->total_in + target_position == match_position) ? "same" : "diff", + match_length); + }); + + if ((ret = xd3_found_match (stream, + /* decoder position */ target_position, + /* length */ match_length, + /* address */ match_position, + /* is_source */ 1))) + { + return ret; + } + + /* If the match ends with the available input: */ + if (target_position + match_length == stream->avail_in) + { + /* Setup continuing match for the next window. */ + stream->match_state = MATCH_TARGET; + stream->match_srcpos = match_end; + } + } + + return 0; +} + +/* Update the small hash. Values in the small_table are offset by + * HASH_CKOFFSET (1) to distinguish empty buckets from real offsets. */ +static void +xd3_scksum_insert (xd3_stream *stream, + usize_t inx, + usize_t scksum, + usize_t pos) +{ + /* If we are maintaining previous duplicates. */ + if (stream->small_prev) + { + usize_t last_pos = stream->small_table[inx]; + xd3_slist *pos_list = & stream->small_prev[pos & stream->sprevmask]; + + /* Note last_pos is offset by HASH_CKOFFSET. */ + pos_list->last_pos = last_pos; + } + + /* Enter the new position into the hash bucket. */ + stream->small_table[inx] = pos + HASH_CKOFFSET; +} + +#if XD3_DEBUG +static int +xd3_check_smatch (const uint8_t *ref0, const uint8_t *inp0, + const uint8_t *inp_max, usize_t cmp_len) +{ + usize_t i; + + for (i = 0; i < cmp_len; i += 1) + { + XD3_ASSERT (ref0[i] == inp0[i]); + } + + if (inp0 + cmp_len < inp_max) + { + XD3_ASSERT (inp0[i] != ref0[i]); + } + + return 1; +} +#endif /* XD3_DEBUG */ + +/* When the hash table indicates a possible small string match, it + * calls this routine to find the best match. The first matching + * position is taken from the small_table, HASH_CKOFFSET is subtracted + * to get the actual position. After checking that match, if previous + * linked lists are in use (because stream->smatcher.small_chain > 1), + * previous matches are tested searching for the longest match. If + * (stream->min_match > MIN_MATCH) then a lazy match is in effect. + */ +static usize_t +xd3_smatch (xd3_stream *stream, + usize_t base, + usize_t scksum, + usize_t *match_offset) +{ + usize_t cmp_len; + usize_t match_length = 0; + usize_t chain = (stream->min_match == MIN_MATCH ? + stream->smatcher.small_chain : + stream->smatcher.small_lchain); + const uint8_t *inp_max = stream->next_in + stream->avail_in; + const uint8_t *inp; + const uint8_t *ref; + + SMALL_HASH_DEBUG1 (stream, stream->next_in + stream->input_position); + + XD3_ASSERT (stream->min_match + stream->input_position <= stream->avail_in); + + base -= HASH_CKOFFSET; + + again: + + IF_DEBUG2 (DP(RINT "smatch at base=%u inp=%u cksum=%u\n", base, + stream->input_position, scksum)); + + /* For small matches, we can always go to the end-of-input because + * the matching position must be less than the input position. */ + XD3_ASSERT (base < stream->input_position); + + ref = stream->next_in + base; + inp = stream->next_in + stream->input_position; + + SMALL_HASH_DEBUG2 (stream, ref); + + /* Expand potential match forward. */ + while (inp < inp_max && *inp == *ref) + { + ++inp; + ++ref; + } + + cmp_len = (usize_t)(inp - (stream->next_in + stream->input_position)); + + /* Verify correctness */ + XD3_ASSERT (xd3_check_smatch (stream->next_in + base, + stream->next_in + stream->input_position, + inp_max, cmp_len)); + + /* Update longest match */ + if (cmp_len > match_length) + { + ( match_length) = cmp_len; + (*match_offset) = base; + + /* Stop if we match the entire input or have a long_enough match. */ + if (inp == inp_max || cmp_len >= stream->smatcher.long_enough) + { + goto done; + } + } + + /* If we have not reached the chain limit, see if there is another + previous position. */ + while (--chain != 0) + { + /* Calculate the previous offset. */ + usize_t prev_pos = stream->small_prev[base & stream->sprevmask].last_pos; + usize_t diff_pos; + + if (prev_pos == 0) + { + break; + } + + prev_pos -= HASH_CKOFFSET; + + if (prev_pos > base) + { + break; + } + + base = prev_pos; + + XD3_ASSERT (stream->input_position > base); + diff_pos = stream->input_position - base; + + /* Stop searching if we go beyond sprevsz, since those entries + * are for unrelated checksum entries. */ + if (diff_pos & ~stream->sprevmask) + { + break; + } + + goto again; + } + + done: + /* Crude efficiency test: if the match is very short and very far back, it's + * unlikely to help, but the exact calculation requires knowing the state of + * the address cache and adjacent instructions, which we can't do here. + * Rather than encode a probably inefficient copy here and check it later + * (which complicates the code a lot), do this: + */ + if (match_length == 4 && stream->input_position - (*match_offset) >= 1<<14) + { + /* It probably takes >2 bytes to encode an address >= 2^14 from here */ + return 0; + } + if (match_length == 5 && stream->input_position - (*match_offset) >= 1<<21) + { + /* It probably takes >3 bytes to encode an address >= 2^21 from here */ + return 0; + } + + /* It's unlikely that a window is large enough for the (match_length == 6 && + * address >= 2^28) check */ + return match_length; +} + +#if XD3_DEBUG +static void +xd3_verify_small_state (xd3_stream *stream, + const uint8_t *inp, + uint32_t x_cksum) +{ + uint32_t state; + uint32_t cksum = xd3_scksum (&state, inp, stream->smatcher.small_look); + + XD3_ASSERT (cksum == x_cksum); +} + +static void +xd3_verify_large_state (xd3_stream *stream, + const uint8_t *inp, + uint32_t x_cksum) +{ + uint32_t cksum = xd3_lcksum (inp, stream->smatcher.large_look); + XD3_ASSERT (cksum == x_cksum); +} +static void +xd3_verify_run_state (xd3_stream *stream, + const uint8_t *inp, + usize_t x_run_l, + uint8_t *x_run_c) +{ + usize_t slook = stream->smatcher.small_look; + uint8_t run_c; + usize_t run_l = xd3_comprun (inp, slook, &run_c); + + XD3_ASSERT (run_l == 0 || run_c == *x_run_c); + XD3_ASSERT (x_run_l > slook || run_l == x_run_l); +} +#endif /* XD3_DEBUG */ + +/* This function computes more source checksums to advance the window. + * Called at every entrance to the string-match loop and each time + * stream->input_position reaches the value returned as + * *next_move_point. NB: this is one of the most expensive functions + * in this code and also the most critical for good compression. + * TODO: optimize the inner loop + */ +static int +xd3_srcwin_move_point (xd3_stream *stream, usize_t *next_move_point) +{ + xoff_t logical_input_cksum_pos; + xoff_t source_size; + + if (stream->src->eof_known) + { + source_size = xd3_source_eof (stream->src); + XD3_ASSERT(stream->srcwin_cksum_pos <= source_size); + + if (stream->srcwin_cksum_pos == source_size) + { + *next_move_point = USIZE_T_MAX; + return 0; + } + } + + /* Begin by advancing at twice the input rate, up to half the + * maximum window size. */ + logical_input_cksum_pos = min((stream->total_in + stream->input_position) * 2, + (stream->total_in + stream->input_position) + + (stream->src->max_winsize / 2)); + + /* If srcwin_cksum_pos is already greater, wait until the difference + * is met. */ + if (stream->srcwin_cksum_pos > logical_input_cksum_pos) + { + *next_move_point = stream->input_position + + (usize_t)(stream->srcwin_cksum_pos - logical_input_cksum_pos); + return 0; + } + + /* A long match may have extended past srcwin_cksum_pos. Don't + * start checksumming already-matched source data. */ + if (stream->maxsrcaddr > stream->srcwin_cksum_pos) + { + stream->srcwin_cksum_pos = stream->maxsrcaddr; + } + + if (logical_input_cksum_pos < stream->srcwin_cksum_pos) + { + logical_input_cksum_pos = stream->srcwin_cksum_pos; + } + + /* Advance at least one source block. With the command-line + * defaults this means: + * + * if (src->size <= src->max_winsize), index the entire source at once + * using the position of the first non-match. This is good for + * small inputs, especially when the content may have moved anywhere + * in the file (e.g., tar files). + * + * if (src->size > src->max_winsize), index at least one block ahead + * of the logical position. This is good for different reasons: + * when a long match spanning several source blocks is encountered, + * this avoids computing checksums for those blocks. + */ + logical_input_cksum_pos += stream->src->blksize; + + while (stream->srcwin_cksum_pos < logical_input_cksum_pos && + (!stream->src->eof_known || + stream->srcwin_cksum_pos < xd3_source_eof (stream->src))) + { + xoff_t blkno; + xoff_t blkbaseoffset; + usize_t blkrem; + ssize_t oldpos; /* Using ssize_t because of a */ + ssize_t blkpos; /* do { blkpos-- } + while (blkpos >= oldpos); */ + int ret; + xd3_blksize_div (stream->srcwin_cksum_pos, + stream->src, &blkno, &blkrem); + oldpos = blkrem; + + if ((ret = xd3_getblk (stream, blkno))) + { + /* TOOFARBACK should never occur here, since we read forward. */ + if (ret == XD3_TOOFARBACK) + { + ret = XD3_INTERNAL; + } + IF_DEBUG1 (DP(RINT + "[srcwin_move_point] async getblk return for %"Q"\n", + blkno)); + return ret; + } + + IF_DEBUG1 (DP(RINT + "[srcwin_move_point] T=%"Q"{%"Q"} S=%"Q" EOF=%"Q" %s\n", + stream->total_in + stream->input_position, + logical_input_cksum_pos, + stream->srcwin_cksum_pos, + xd3_source_eof (stream->src), + stream->src->eof_known ? "known" : "unknown")); + + blkpos = xd3_bytes_on_srcblk (stream->src, blkno); + + if (blkpos < (ssize_t) stream->smatcher.large_look) + { + stream->srcwin_cksum_pos = (blkno + 1) * stream->src->blksize; + IF_DEBUG1 (DP(RINT "[srcwin_move_point] continue (end-of-block)\n")); + continue; + } + + /* This inserts checksums for the entire block, in reverse, + * starting from the end of the block. This logic does not test + * stream->srcwin_cksum_pos because it always advances it to the + * start of the next block. + * + * oldpos is the srcwin_cksum_pos within this block. blkpos is + * the number of bytes available. Each iteration inspects + * large_look bytes then steps back large_step bytes. The + * if-stmt above ensures at least one large_look of data. */ + blkpos -= stream->smatcher.large_look; + blkbaseoffset = stream->src->blksize * blkno; + + do + { + uint32_t cksum = xd3_lcksum (stream->src->curblk + blkpos, + stream->smatcher.large_look); + usize_t hval = xd3_checksum_hash (& stream->large_hash, cksum); + + stream->large_table[hval] = + (usize_t) (blkbaseoffset + + (xoff_t)(blkpos + HASH_CKOFFSET)); + + IF_DEBUG (stream->large_ckcnt += 1); + + blkpos -= stream->smatcher.large_step; + } + while (blkpos >= oldpos); + + stream->srcwin_cksum_pos = (blkno + 1) * stream->src->blksize; + } + + IF_DEBUG1 (DP(RINT + "[srcwin_move_point] exited loop T=%"Q"{%"Q"} " + "S=%"Q" EOF=%"Q" %s\n", + stream->total_in + stream->input_position, + logical_input_cksum_pos, + stream->srcwin_cksum_pos, + xd3_source_eof (stream->src), + stream->src->eof_known ? "known" : "unknown")); + + if (stream->src->eof_known) + { + source_size = xd3_source_eof (stream->src); + + if (stream->srcwin_cksum_pos >= source_size) + { + /* This invariant is needed for xd3_source_cksum_offset() */ + stream->srcwin_cksum_pos = source_size; + *next_move_point = USIZE_T_MAX; + IF_DEBUG1 (DP(RINT + "[srcwin_move_point] finished with source input\n")); + return 0; + } + } + + /* How long until this function should be called again. */ + XD3_ASSERT(stream->srcwin_cksum_pos >= logical_input_cksum_pos); + *next_move_point = stream->input_position + 1 + + (usize_t)(stream->srcwin_cksum_pos - logical_input_cksum_pos); + return 0; +} + +#endif /* XD3_ENCODER */ + +/******************************************************************** + TEMPLATE pass + *********************************************************************/ + +#endif /* __XDELTA3_C_INLINE_PASS__ */ +#ifdef __XDELTA3_C_TEMPLATE_PASS__ + +#if XD3_ENCODER + +/******************************************************************** + Templates + *******************************************************************/ + +/* Template macros */ +#define XD3_TEMPLATE(x) XD3_TEMPLATE2(x,TEMPLATE) +#define XD3_TEMPLATE2(x,n) XD3_TEMPLATE3(x,n) +#define XD3_TEMPLATE3(x,n) x ## n +#define XD3_STRINGIFY(x) XD3_STRINGIFY2(x) +#define XD3_STRINGIFY2(x) #x + +static int XD3_TEMPLATE(xd3_string_match_) (xd3_stream *stream); + +static const xd3_smatcher XD3_TEMPLATE(__smatcher_) = +{ + XD3_STRINGIFY(TEMPLATE), + XD3_TEMPLATE(xd3_string_match_), +#if SOFTCFG == 1 + 0, 0, 0, 0, 0, 0, 0 +#else + LLOOK, LSTEP, SLOOK, SCHAIN, SLCHAIN, MAXLAZY, LONGENOUGH +#endif +}; + +static int +XD3_TEMPLATE(xd3_string_match_) (xd3_stream *stream) +{ + const int DO_SMALL = ! (stream->flags & XD3_NOCOMPRESS); + const int DO_LARGE = (stream->src != NULL); + const int DO_RUN = (1); + + const uint8_t *inp; + uint32_t scksum = 0; + uint32_t scksum_state = 0; + uint32_t lcksum = 0; + usize_t sinx; + usize_t linx; + uint8_t run_c; + usize_t run_l; + int ret; + usize_t match_length; + usize_t match_offset = 0; + usize_t next_move_point; + + /* If there will be no compression due to settings or short input, + * skip it entirely. */ + if (! (DO_SMALL || DO_LARGE || DO_RUN) || + stream->input_position + SLOOK > stream->avail_in) { goto loopnomore; } + + if ((ret = xd3_string_match_init (stream))) { return ret; } + + /* The restartloop label is reached when the incremental loop state + * needs to be reset. */ + restartloop: + + /* If there is not enough input remaining for any kind of match, + skip it. */ + if (stream->input_position + SLOOK > stream->avail_in) { goto loopnomore; } + + /* Now reset the incremental loop state: */ + + /* The min_match variable is updated to avoid matching the same lazy + * match over and over again. For example, if you find a (small) + * match of length 9 at one position, you will likely find a match + * of length 8 at the next position. */ + if (xd3_iopt_last_matched (stream) > stream->input_position) + { + stream->min_match = max(MIN_MATCH, + 1 + xd3_iopt_last_matched(stream) - + stream->input_position); + } + else + { + stream->min_match = MIN_MATCH; + } + + /* The current input byte. */ + inp = stream->next_in + stream->input_position; + + /* Small match state. */ + if (DO_SMALL) + { + scksum = xd3_scksum (&scksum_state, inp, SLOOK); + } + + /* Run state. */ + if (DO_RUN) + { + run_l = xd3_comprun (inp, SLOOK, & run_c); + } + + /* Large match state. We continue the loop even after not enough + * bytes for LLOOK remain, so always check stream->input_position in + * DO_LARGE code. */ + if (DO_LARGE && (stream->input_position + LLOOK <= stream->avail_in)) + { + /* Source window: next_move_point is the point that + * stream->input_position must reach before computing more + * source checksum. */ + if ((ret = xd3_srcwin_move_point (stream, & next_move_point))) + { + return ret; + } + + lcksum = xd3_lcksum (inp, LLOOK); + } + + /* TRYLAZYLEN: True if a certain length match should be followed by + * lazy search. This checks that LEN is shorter than MAXLAZY and + * that there is enough leftover data to consider lazy matching. + * "Enough" is set to 2 since the next match will start at the next + * offset, it must match two extra characters. */ +#define TRYLAZYLEN(LEN,POS,MAX) ((MAXLAZY) > 0 && (LEN) < (MAXLAZY) \ + && (POS) + (LEN) <= (MAX) - 2) + + /* HANDLELAZY: This statement is called each time an instruciton is + * emitted (three cases). If the instruction is large enough, the + * loop is restarted, otherwise lazy matching may ensue. */ +#define HANDLELAZY(mlen) \ + if (TRYLAZYLEN ((mlen), (stream->input_position), (stream->avail_in))) \ + { stream->min_match = (mlen) + LEAST_MATCH_INCR; goto updateone; } \ + else \ + { stream->input_position += (mlen); goto restartloop; } + + /* Now loop over one input byte at a time until a match is found... */ + for (;; inp += 1, stream->input_position += 1) + { + /* Now we try three kinds of string match in order of expense: + * run, large match, small match. */ + + /* Expand the start of a RUN. The test for (run_l == SLOOK) + * avoids repeating this check when we pass through a run area + * performing lazy matching. The run is only expanded once when + * the min_match is first reached. If lazy matching is + * performed, the run_l variable will remain inconsistent until + * the first non-running input character is reached, at which + * time the run_l may then again grow to SLOOK. */ + if (DO_RUN && run_l == SLOOK) + { + usize_t max_len = stream->avail_in - stream->input_position; + + IF_DEBUG (xd3_verify_run_state (stream, inp, run_l, &run_c)); + + while (run_l < max_len && inp[run_l] == run_c) { run_l += 1; } + + /* Output a RUN instruction. */ + if (run_l >= stream->min_match && run_l >= MIN_RUN) + { + if ((ret = xd3_emit_run (stream, stream->input_position, + run_l, &run_c))) { return ret; } + + HANDLELAZY (run_l); + } + } + + /* If there is enough input remaining. */ + if (DO_LARGE && (stream->input_position + LLOOK <= stream->avail_in)) + { + if ((stream->input_position >= next_move_point) && + (ret = xd3_srcwin_move_point (stream, & next_move_point))) + { + return ret; + } + + linx = xd3_checksum_hash (& stream->large_hash, lcksum); + + IF_DEBUG (xd3_verify_large_state (stream, inp, lcksum)); + + if (stream->large_table[linx] != 0) + { + /* the match_setup will fail if the source window has + * been decided and the match lies outside it. + * OPT: Consider forcing a window at this point to + * permit a new source window. */ + xoff_t adj_offset = + xd3_source_cksum_offset(stream, + stream->large_table[linx] - + HASH_CKOFFSET); + if (xd3_source_match_setup (stream, adj_offset) == 0) + { + if ((ret = xd3_source_extend_match (stream))) + { + return ret; + } + + /* Update stream position. match_fwd is zero if no + * match. */ + if (stream->match_fwd > 0) + { + HANDLELAZY (stream->match_fwd); + } + } + } + } + + /* Small matches. */ + if (DO_SMALL) + { + sinx = xd3_checksum_hash (& stream->small_hash, scksum); + + /* Verify incremental state in debugging mode. */ + IF_DEBUG (xd3_verify_small_state (stream, inp, scksum)); + + /* Search for the longest match */ + if (stream->small_table[sinx] != 0) + { + match_length = xd3_smatch (stream, + stream->small_table[sinx], + scksum, + & match_offset); + } + else + { + match_length = 0; + } + + /* Insert a hash for this string. */ + xd3_scksum_insert (stream, sinx, scksum, stream->input_position); + + /* Maybe output a COPY instruction */ + if (match_length >= stream->min_match) + { + IF_DEBUG2 ({ + static int x = 0; + DP(RINT "[target match:%d] " + "(-%d) [ %u bytes ]\n", + x++, + stream->input_position, + stream->input_position + match_length, + match_offset, + match_offset + match_length, + stream->input_position - match_offset, + match_length); + }); + + if ((ret = xd3_found_match (stream, + /* decoder position */ + stream->input_position, + /* length */ match_length, + /* address */ (xoff_t) match_offset, + /* is_source */ 0))) + { + return ret; + } + + /* Copy instruction. */ + HANDLELAZY (match_length); + } + } + + /* The logic above prevents excess work during lazy matching by + * increasing min_match to avoid smaller matches. Each time we + * advance stream->input_position by one, the minimum match + * shortens as well. */ + if (stream->min_match > MIN_MATCH) + { + stream->min_match -= 1; + } + + updateone: + + /* See if there are no more incremental cksums to compute. */ + if (stream->input_position + SLOOK == stream->avail_in) + { + goto loopnomore; + } + + /* Compute next RUN, CKSUM */ + if (DO_RUN) + { + NEXTRUN (inp[SLOOK]); + } + + if (DO_SMALL) + { + scksum = xd3_small_cksum_update (&scksum_state, inp, SLOOK); + } + + if (DO_LARGE && (stream->input_position + LLOOK < stream->avail_in)) + { + lcksum = xd3_large_cksum_update (lcksum, inp, LLOOK); + } + } + + loopnomore: + return 0; +} + +#endif /* XD3_ENCODER */ +#endif /* __XDELTA3_C_TEMPLATE_PASS__ */ diff --git a/3rdparty/xdelta3-3.0.8/xdelta3.h b/3rdparty/xdelta3-3.0.8/xdelta3.h new file mode 100644 index 0000000..d6471fa --- /dev/null +++ b/3rdparty/xdelta3-3.0.8/xdelta3.h @@ -0,0 +1,1373 @@ +/* xdelta 3 - delta compression tools and library + * Copyright (C) 2001, 2003, 2004, 2005, 2006, 2007, + * 2008, 2009, 2010, 2011, 2012, 2013. Joshua P. MacDonald + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +/* To know more about Xdelta, start by reading xdelta3.c. If you are + * ready to use the API, continue reading here. There are two + * interfaces -- xd3_encode_input and xd3_decode_input -- plus a dozen + * or so related calls. This interface is styled after Zlib. */ + +#ifndef _XDELTA3_H_ +#define _XDELTA3_H_ + +#define _POSIX_SOURCE +#define _ISOC99_SOURCE +#define _C99_SOURCE + +#if HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include +#include +#include +#include +#include +#include + +/****************************************************************/ + +/* Default configured value of stream->winsize. If the program + * supplies xd3_encode_input() with data smaller than winsize the + * stream will automatically buffer the input, otherwise the input + * buffer is used directly. + */ +#ifndef XD3_DEFAULT_WINSIZE +#define XD3_DEFAULT_WINSIZE (1U << 23) +#endif + +/* Default total size of the source window used in xdelta3-main.h */ +#ifndef XD3_DEFAULT_SRCWINSZ +#define XD3_DEFAULT_SRCWINSZ (1U << 26) +#endif + +/* When Xdelta requests a memory allocation for certain buffers, it + * rounds up to units of at least this size. The code assumes (and + * asserts) that this is a power-of-two. */ +#ifndef XD3_ALLOCSIZE +#define XD3_ALLOCSIZE (1U<<14) +#endif + +/* The XD3_HARDMAXWINSIZE parameter is a safety mechanism to protect + * decoders against malicious files. The decoder will never decode a + * window larger than this. If the file specifies VCD_TARGET the + * decoder may require two buffers of this size. + * + * 8-16MB is reasonable, probably don't need to go larger. */ +#ifndef XD3_HARDMAXWINSIZE +#define XD3_HARDMAXWINSIZE (1U<<24) +#endif +/* The IOPT_SIZE value sets the size of a buffer used to batch + * overlapping copy instructions before they are optimized by picking + * the best non-overlapping ranges. The larger this buffer, the + * longer a forced xd3_srcwin_setup() decision is held off. Setting + * this value to 0 causes an unlimited buffer to be used. */ +#ifndef XD3_DEFAULT_IOPT_SIZE +#define XD3_DEFAULT_IOPT_SIZE (1U<<15) +#endif + +/* The maximum distance backward to search for small matches */ +#ifndef XD3_DEFAULT_SPREVSZ +#define XD3_DEFAULT_SPREVSZ (1U<<18) +#endif + +/* The default compression level + */ +#ifndef XD3_DEFAULT_LEVEL +#define XD3_DEFAULT_LEVEL 3 +#endif + +#ifndef XD3_DEFAULT_SECONDARY_LEVEL +#define XD3_DEFAULT_SECONDARY_LEVEL 6 +#endif + +#ifndef XD3_USE_LARGEFILE64 +#define XD3_USE_LARGEFILE64 1 +#endif + +/* Sizes and addresses within VCDIFF windows are represented as usize_t + * + * For source-file offsets and total file sizes, total input and + * output counts, the xoff_t type is used. The decoder and encoder + * generally check for overflow of the xoff_t size (this is tested at + * the 32bit boundary [xdelta3-test.h]). + */ +#ifndef _WIN32 +#include +#else +#define WIN32_LEAN_AND_MEAN +#if XD3_USE_LARGEFILE64 +/* 64 bit file offsets: uses GetFileSizeEx and SetFilePointerEx. + * requires Win2000 or newer version of WinNT */ +#define WINVER 0x0500 +#define _WIN32_WINNT 0x0500 +#else +/* 32 bit (DWORD) file offsets: uses GetFileSize and + * SetFilePointer. compatible with win9x-me and WinNT4 */ +#define WINVER 0x0400 +#define _WIN32_WINNT 0x0400 +#endif +#include +#ifdef _MSC_VER +#define inline +typedef signed int ssize_t; +#if _MSC_VER < 1600 +typedef unsigned char uint8_t; +typedef unsigned short uint16_t; +typedef unsigned long uint32_t; +typedef ULONGLONG uint64_t; +#else +/* For MSVC10 and above */ +#include +#endif +#else +/* mingw32, lcc and watcom provide a proper header */ +#include +#endif +#endif + +typedef uint32_t usize_t; + +#if XD3_USE_LARGEFILE64 +#define __USE_FILE_OFFSET64 1 /* GLIBC: for 64bit fileops, ... ? */ +#ifndef _LARGEFILE_SOURCE +#define _LARGEFILE_SOURCE +#endif +#ifndef _FILE_OFFSET_BITS +#define _FILE_OFFSET_BITS 64 +#endif + +typedef uint64_t xoff_t; +#define SIZEOF_XOFF_T 8 +#define SIZEOF_USIZE_T 4 +#ifndef WIN32 +#if SIZEOF_SIZE_T == 8 +#define Q "z" +#else +#define Q "ll" +#endif +#else +#define Q "I64" +#endif +#else +typedef uint32_t xoff_t; +#define SIZEOF_XOFF_T 4 +#define SIZEOF_USIZE_T 4 +#define Q +#endif + +#define USE_UINT32 (SIZEOF_USIZE_T == 4 || \ + SIZEOF_XOFF_T == 4 || REGRESSION_TEST) +#define USE_UINT64 (SIZEOF_USIZE_T == 8 || \ + SIZEOF_XOFF_T == 8 || REGRESSION_TEST) + +#ifndef UNALIGNED_OK +#ifdef HAVE_ALIGNED_ACCESS_REQUIRED +#define UNALIGNED_OK 0 +#else +/* This generally includes all Windows builds. */ +#define UNALIGNED_OK 1 +#endif +#endif + +/**********************************************************************/ + +/* Whether to build the encoder, otherwise only build the decoder. */ +#ifndef XD3_ENCODER +#define XD3_ENCODER 1 +#endif + +/* The code returned when main() fails, also defined in system + includes. */ +#ifndef EXIT_FAILURE +#define EXIT_FAILURE 1 +#endif + +/* REGRESSION TEST enables the "xdelta3 test" command, which runs a + series of self-tests. */ +#ifndef REGRESSION_TEST +#define REGRESSION_TEST 0 +#endif + +/* XD3_DEBUG=1 enables assertions and various statistics. Levels > 1 + * enable some additional output only useful during development and + * debugging. */ +#ifndef XD3_DEBUG +#define XD3_DEBUG 0 +#endif + +#ifndef PYTHON_MODULE +#define PYTHON_MODULE 0 +#endif + +#ifndef SWIG_MODULE +#define SWIG_MODULE 0 +#endif + +/* There are three string matching functions supplied: one fast, one + * slow (default), and one soft-configurable. To disable any of + * these, use the following definitions. */ +#ifndef XD3_BUILD_SLOW +#define XD3_BUILD_SLOW 1 +#endif +#ifndef XD3_BUILD_FAST +#define XD3_BUILD_FAST 1 +#endif +#ifndef XD3_BUILD_FASTER +#define XD3_BUILD_FASTER 1 +#endif +#ifndef XD3_BUILD_FASTEST +#define XD3_BUILD_FASTEST 1 +#endif +#ifndef XD3_BUILD_SOFT +#define XD3_BUILD_SOFT 1 +#endif +#ifndef XD3_BUILD_DEFAULT +#define XD3_BUILD_DEFAULT 1 +#endif + +#if XD3_DEBUG +#include +#endif + +/* XPRINT. Debug output and VCDIFF_TOOLS functions report to stderr. + * I have used an irregular style to abbreviate [fprintf(stderr, "] as + * [DP(RINT "]. */ +#define DP fprintf +#define RINT stderr, + +typedef struct _xd3_stream xd3_stream; +typedef struct _xd3_source xd3_source; +typedef struct _xd3_hash_cfg xd3_hash_cfg; +typedef struct _xd3_smatcher xd3_smatcher; +typedef struct _xd3_rinst xd3_rinst; +typedef struct _xd3_dinst xd3_dinst; +typedef struct _xd3_hinst xd3_hinst; +typedef struct _xd3_winst xd3_winst; +typedef struct _xd3_rpage xd3_rpage; +typedef struct _xd3_addr_cache xd3_addr_cache; +typedef struct _xd3_output xd3_output; +typedef struct _xd3_desect xd3_desect; +typedef struct _xd3_iopt_buflist xd3_iopt_buflist; +typedef struct _xd3_rlist xd3_rlist; +typedef struct _xd3_sec_type xd3_sec_type; +typedef struct _xd3_sec_cfg xd3_sec_cfg; +typedef struct _xd3_sec_stream xd3_sec_stream; +typedef struct _xd3_config xd3_config; +typedef struct _xd3_code_table_desc xd3_code_table_desc; +typedef struct _xd3_code_table_sizes xd3_code_table_sizes; +typedef struct _xd3_slist xd3_slist; +typedef struct _xd3_whole_state xd3_whole_state; +typedef struct _xd3_wininfo xd3_wininfo; + +/* The stream configuration has three callbacks functions, all of + * which may be supplied with NULL values. If config->getblk is + * provided as NULL, the stream returns XD3_GETSRCBLK. */ + +typedef void* (xd3_alloc_func) (void *opaque, + size_t items, + usize_t size); +typedef void (xd3_free_func) (void *opaque, + void *address); + +typedef int (xd3_getblk_func) (xd3_stream *stream, + xd3_source *source, + xoff_t blkno); + +/* These are internal functions to delay construction of encoding + * tables and support alternate code tables. See the comments & code + * enabled by GENERIC_ENCODE_TABLES. */ + +typedef const xd3_dinst* (xd3_code_table_func) (void); +typedef int (xd3_comp_table_func) (xd3_stream *stream, + const uint8_t **data, + usize_t *size); + + + +#if XD3_DEBUG +#define XD3_ASSERT(x) \ + do { if (! (x)) { DP(RINT "%s:%d: XD3 assertion failed: %s\n", __FILE__, __LINE__, #x); \ + abort (); } } while (0) +#else +#define XD3_ASSERT(x) (void)0 +#endif /* XD3_DEBUG */ + +#ifndef max +#define max(x,y) ((x) < (y) ? (y) : (x)) +#endif +#ifndef min +#define min(x,y) ((x) < (y) ? (x) : (y)) +#endif + +/**************************************************************** + PUBLIC ENUMS + ******************************************************************/ + +/* These are the five ordinary status codes returned by the + * xd3_encode_input() and xd3_decode_input() state machines. */ +typedef enum { + + /* An application must be prepared to handle these five return + * values from either xd3_encode_input or xd3_decode_input, except + * in the case of no-source compression, in which case XD3_GETSRCBLK + * is never returned. More detailed comments for these are given in + * xd3_encode_input and xd3_decode_input comments, below. */ + XD3_INPUT = -17703, /* need input */ + XD3_OUTPUT = -17704, /* have output */ + XD3_GETSRCBLK = -17705, /* need a block of source input (with no + * xd3_getblk function), a chance to do + * non-blocking read. */ + XD3_GOTHEADER = -17706, /* (decode-only) after the initial VCDIFF & + first window header */ + XD3_WINSTART = -17707, /* notification: returned before a window is + * processed, giving a chance to + * XD3_SKIP_WINDOW or not XD3_SKIP_EMIT that + * window. */ + XD3_WINFINISH = -17708, /* notification: returned after + encode/decode & output for a window */ + XD3_TOOFARBACK = -17709, /* (encoder only) may be returned by + getblk() if the block is too old */ + XD3_INTERNAL = -17710, /* internal error */ + XD3_INVALID = -17711, /* invalid config */ + XD3_INVALID_INPUT = -17712, /* invalid input/decoder error */ + XD3_NOSECOND = -17713, /* when secondary compression finds no + improvement. */ + XD3_UNIMPLEMENTED = -17714 /* currently VCD_TARGET */ +} xd3_rvalues; + +/* special values in config->flags */ +typedef enum +{ + XD3_JUST_HDR = (1 << 1), /* used by VCDIFF tools, see + xdelta3-main.h. */ + XD3_SKIP_WINDOW = (1 << 2), /* used by VCDIFF tools, see + xdelta3-main.h. */ + XD3_SKIP_EMIT = (1 << 3), /* used by VCDIFF tools, see + xdelta3-main.h. */ + XD3_FLUSH = (1 << 4), /* flush the stream buffer to + prepare for + xd3_stream_close(). */ + + XD3_SEC_DJW = (1 << 5), /* use DJW static huffman */ + XD3_SEC_FGK = (1 << 6), /* use FGK adaptive huffman */ + XD3_SEC_LZMA = (1 << 24), /* use LZMA secondary */ + + XD3_SEC_TYPE = (XD3_SEC_DJW | XD3_SEC_FGK | XD3_SEC_LZMA), + + XD3_SEC_NODATA = (1 << 7), /* disable secondary compression of + the data section. */ + XD3_SEC_NOINST = (1 << 8), /* disable secondary compression of + the inst section. */ + XD3_SEC_NOADDR = (1 << 9), /* disable secondary compression of + the addr section. */ + + XD3_SEC_NOALL = (XD3_SEC_NODATA | XD3_SEC_NOINST | XD3_SEC_NOADDR), + + XD3_ADLER32 = (1 << 10), /* enable checksum computation in + the encoder. */ + XD3_ADLER32_NOVER = (1 << 11), /* disable checksum verification in + the decoder. */ + + XD3_ALT_CODE_TABLE = (1 << 12), /* for testing th + e alternate code table encoding. */ + + XD3_NOCOMPRESS = (1 << 13), /* disable ordinary data + * compression feature, only search + * the source, not the target. */ + XD3_BEGREEDY = (1 << 14), /* disable the "1.5-pass + * algorithm", instead use greedy + * matching. Greedy is off by + * default. */ + XD3_ADLER32_RECODE = (1 << 15), /* used by "recode". */ + + /* 4 bits to set the compression level the same as the command-line + * setting -1 through -9 (-0 corresponds to the XD3_NOCOMPRESS flag, + * and is independent of compression level). This is for + * convenience, especially with xd3_encode_memory(). */ + + XD3_COMPLEVEL_SHIFT = 20, /* 20 - 23 */ + XD3_COMPLEVEL_MASK = (0xF << XD3_COMPLEVEL_SHIFT), + XD3_COMPLEVEL_1 = (1 << XD3_COMPLEVEL_SHIFT), + XD3_COMPLEVEL_2 = (2 << XD3_COMPLEVEL_SHIFT), + XD3_COMPLEVEL_3 = (3 << XD3_COMPLEVEL_SHIFT), + XD3_COMPLEVEL_6 = (6 << XD3_COMPLEVEL_SHIFT), + XD3_COMPLEVEL_9 = (9 << XD3_COMPLEVEL_SHIFT) + +} xd3_flags; + +/* The values of this enumeration are set in xd3_config using the + * smatch_cfg variable. It can be set to default, slow, fast, etc., + * and soft. */ +typedef enum +{ + XD3_SMATCH_DEFAULT = 0, /* Flags may contain XD3_COMPLEVEL bits, + else default. */ + XD3_SMATCH_SLOW = 1, + XD3_SMATCH_FAST = 2, + XD3_SMATCH_FASTER = 3, + XD3_SMATCH_FASTEST = 4, + XD3_SMATCH_SOFT = 5 +} xd3_smatch_cfg; + +/********************************************************************* + PRIVATE ENUMS +**********************************************************************/ + +/* stream->match_state is part of the xd3_encode_input state machine + * for source matching: + * + * 1. the XD3_GETSRCBLK block-read mechanism means reentrant matching + * 2. this state spans encoder windows: a match and end-of-window + * will continue in the next 3. the initial target byte and source + * byte are a presumed match, to avoid some computation in case the + * inputs are identical. + */ +typedef enum { + + MATCH_TARGET = 0, /* in this state, attempt to match the start of + * the target with the previously set source + * address (initially 0). */ + MATCH_BACKWARD = 1, /* currently expanding a match backward in the + source/target. */ + MATCH_FORWARD = 2, /* currently expanding a match forward in the + source/target. */ + MATCH_SEARCHING = 3 /* currently searching for a match. */ + +} xd3_match_state; + +/* The xd3_encode_input state machine steps through these states in + * the following order. The matcher is reentrant and returns + * XD3_INPUT whenever it requires more data. After receiving + * XD3_INPUT, if the application reads EOF it should call + * xd3_stream_close(). + */ +typedef enum { + + ENC_INIT = 0, /* xd3_encode_input has never been called. */ + ENC_INPUT = 1, /* waiting for xd3_avail_input () to be called. */ + ENC_SEARCH = 2, /* currently searching for matches. */ + ENC_INSTR = 3, /* currently formatting output. */ + ENC_FLUSH = 4, /* currently emitting output. */ + ENC_POSTOUT = 5, /* after an output section. */ + ENC_POSTWIN = 6, /* after all output sections. */ + ENC_ABORTED = 7 /* abort. */ +} xd3_encode_state; + +/* The xd3_decode_input state machine steps through these states in + * the following order. The matcher is reentrant and returns + * XD3_INPUT whenever it requires more data. After receiving + * XD3_INPUT, if the application reads EOF it should call + * xd3_stream_close(). + * + * 0-8: the VCDIFF header + * 9-18: the VCDIFF window header + * 19-21: the three primary sections: data, inst, addr + * 22: producing output: returns XD3_OUTPUT, possibly XD3_GETSRCBLK, + * 23: return XD3_WINFINISH, set state=9 to decode more input + */ +typedef enum { + + DEC_VCHEAD = 0, /* VCDIFF header */ + DEC_HDRIND = 1, /* header indicator */ + + DEC_SECONDID = 2, /* secondary compressor ID */ + + DEC_TABLEN = 3, /* code table length */ + DEC_NEAR = 4, /* code table near */ + DEC_SAME = 5, /* code table same */ + DEC_TABDAT = 6, /* code table data */ + + DEC_APPLEN = 7, /* application data length */ + DEC_APPDAT = 8, /* application data */ + + DEC_WININD = 9, /* window indicator */ + + DEC_CPYLEN = 10, /* copy window length */ + DEC_CPYOFF = 11, /* copy window offset */ + + DEC_ENCLEN = 12, /* length of delta encoding */ + DEC_TGTLEN = 13, /* length of target window */ + DEC_DELIND = 14, /* delta indicator */ + + DEC_DATALEN = 15, /* length of ADD+RUN data */ + DEC_INSTLEN = 16, /* length of instruction data */ + DEC_ADDRLEN = 17, /* length of address data */ + + DEC_CKSUM = 18, /* window checksum */ + + DEC_DATA = 19, /* data section */ + DEC_INST = 20, /* instruction section */ + DEC_ADDR = 21, /* address section */ + + DEC_EMIT = 22, /* producing data */ + + DEC_FINISH = 23, /* window finished */ + + DEC_ABORTED = 24 /* xd3_abort_stream */ +} xd3_decode_state; + +/************************************************************ + internal types + ************************************************************/ + +/* instruction lists used in the IOPT buffer */ +struct _xd3_rlist +{ + xd3_rlist *next; + xd3_rlist *prev; +}; + +/* the raw encoding of an instruction used in the IOPT buffer */ +struct _xd3_rinst +{ + uint8_t type; + uint8_t xtra; + uint8_t code1; + uint8_t code2; + usize_t pos; + usize_t size; + xoff_t addr; + xd3_rlist link; +}; + +/* the code-table form of an single- or double-instruction */ +struct _xd3_dinst +{ + uint8_t type1; + uint8_t size1; + uint8_t type2; + uint8_t size2; +}; + +/* the decoded form of a single (half) instruction. */ +struct _xd3_hinst +{ + uint8_t type; + uint32_t size; /* TODO: why decode breaks if this is usize_t? */ + uint32_t addr; /* TODO: why decode breaks if this is usize_t? */ +}; + +/* the form of a whole-file instruction */ +struct _xd3_winst +{ + uint8_t type; /* RUN, ADD, COPY */ + uint8_t mode; /* 0, VCD_SOURCE, VCD_TARGET */ + usize_t size; + xoff_t addr; + xoff_t position; /* absolute position of this inst */ +}; + +/* used by the encoder to buffer output in sections. list of blocks. */ +struct _xd3_output +{ + uint8_t *base; + usize_t next; + usize_t avail; + xd3_output *next_page; +}; + +/* used by the decoder to buffer input in sections. */ +struct _xd3_desect +{ + const uint8_t *buf; + const uint8_t *buf_max; + uint32_t size; /* TODO: why decode breaks if this is usize_t? */ + usize_t pos; + + /* used in xdelta3-decode.h */ + uint8_t *copied1; + usize_t alloc1; + + /* used in xdelta3-second.h */ + uint8_t *copied2; + usize_t alloc2; +}; + +/* the VCDIFF address cache, see the RFC */ +struct _xd3_addr_cache +{ + usize_t s_near; + usize_t s_same; + usize_t next_slot; /* the circular index for near */ + usize_t *near_array; /* array of size s_near */ + usize_t *same_array; /* array of size s_same*256 */ +}; + +/* the IOPT buffer list is just a list of buffers, which may be allocated + * during encode when using an unlimited buffer. */ +struct _xd3_iopt_buflist +{ + xd3_rinst *buffer; + xd3_iopt_buflist *next; +}; + +/* This is the record of a pre-compiled configuration, a subset of + xd3_config. */ +struct _xd3_smatcher +{ + const char *name; + int (*string_match) (xd3_stream *stream); + usize_t large_look; + usize_t large_step; + usize_t small_look; + usize_t small_chain; + usize_t small_lchain; + usize_t max_lazy; + usize_t long_enough; +}; + +/* hash table size & power-of-two hash function. */ +struct _xd3_hash_cfg +{ + usize_t size; + usize_t shift; + usize_t mask; +}; + +/* the sprev list */ +struct _xd3_slist +{ + usize_t last_pos; +}; + +/* window info (for whole state) */ +struct _xd3_wininfo { + xoff_t offset; + usize_t length; + uint32_t adler32; +}; + +/* whole state for, e.g., merge */ +struct _xd3_whole_state { + usize_t addslen; + uint8_t *adds; + usize_t adds_alloc; + + usize_t instlen; + xd3_winst *inst; + usize_t inst_alloc; + + usize_t wininfolen; + xd3_wininfo *wininfo; + usize_t wininfo_alloc; + + xoff_t length; +}; + +/******************************************************************** + public types + *******************************************************************/ + +/* Settings for the secondary compressor. */ +struct _xd3_sec_cfg +{ + int data_type; /* Which section. (set automatically) */ + usize_t ngroups; /* Number of DJW Huffman groups. */ + usize_t sector_size; /* Sector size. */ + int inefficient; /* If true, ignore efficiency check [avoid XD3_NOSECOND]. */ +}; + +/* This is the user-visible stream configuration. */ +struct _xd3_config +{ + usize_t winsize; /* The encoder window size. */ + usize_t sprevsz; /* How far back small string + matching goes */ + usize_t iopt_size; /* entries in the + instruction-optimizing + buffer */ + + xd3_getblk_func *getblk; /* The three callbacks. */ + xd3_alloc_func *alloc; + xd3_free_func *freef; + void *opaque; /* Not used. */ + int flags; /* stream->flags are initialized + * from xd3_config & never + * modified by the library. Use + * xd3_set_flags to modify flags + * settings mid-stream. */ + + xd3_sec_cfg sec_data; /* Secondary compressor config: data */ + xd3_sec_cfg sec_inst; /* Secondary compressor config: inst */ + xd3_sec_cfg sec_addr; /* Secondary compressor config: addr */ + + xd3_smatch_cfg smatch_cfg; /* See enum: use fields below for + soft config */ + xd3_smatcher smatcher_soft; +}; + +/* The primary source file object. You create one of these objects and + * initialize the first four fields. This library maintains the next + * 5 fields. The configured getblk implementation is responsible for + * setting the final 3 fields when called (and/or when XD3_GETSRCBLK + * is returned). + */ +struct _xd3_source +{ + /* you set */ + usize_t blksize; /* block size */ + const char *name; /* its name, for debug/print + purposes */ + void *ioh; /* opaque handle */ + xoff_t max_winsize; /* maximum visible buffer */ + + /* getblk sets */ + xoff_t curblkno; /* current block number: client + sets after getblk request */ + usize_t onblk; /* number of bytes on current + block: client sets, must be >= 0 + and <= blksize */ + const uint8_t *curblk; /* current block array: client + sets after getblk request */ + + /* xd3 sets */ + usize_t srclen; /* length of this source window */ + xoff_t srcbase; /* offset of this source window + in the source itself */ + int shiftby; /* for power-of-two blocksizes */ + int maskby; /* for power-of-two blocksizes */ + xoff_t cpyoff_blocks; /* offset of dec_cpyoff in blocks */ + usize_t cpyoff_blkoff; /* offset of copy window in + blocks, remainder */ + xoff_t getblkno; /* request block number: xd3 sets + current getblk request */ + + /* See xd3_getblk() */ + xoff_t max_blkno; /* Maximum block, if eof is known, + * otherwise, equals frontier_blkno + * (initially 0). */ + xoff_t frontier_blkno; /* If eof is unknown, the next + * source position to be read. + * Otherwise, equal to + * max_blkno. */ + usize_t onlastblk; /* Number of bytes on max_blkno */ + int eof_known; /* Set to true when the first + * partial block is read. */ +}; + +/* The primary xd3_stream object, used for encoding and decoding. You + * may access only two fields: avail_out, next_out. Use the methods + * above to operate on xd3_stream. */ +struct _xd3_stream +{ + /* input state */ + const uint8_t *next_in; /* next input byte */ + usize_t avail_in; /* number of bytes available at + next_in */ + xoff_t total_in; /* how many bytes in */ + + /* output state */ + uint8_t *next_out; /* next output byte */ + usize_t avail_out; /* number of bytes available at + next_out */ + usize_t space_out; /* total out space */ + xoff_t current_window; /* number of windows encoded/decoded */ + xoff_t total_out; /* how many bytes out */ + + /* to indicate an error, xd3 sets */ + const char *msg; /* last error message, NULL if + no error */ + + /* source configuration */ + xd3_source *src; /* source array */ + + /* encoder memory configuration */ + usize_t winsize; /* suggested window size */ + usize_t sprevsz; /* small string, previous window + size (power of 2) */ + usize_t sprevmask; /* small string, previous window + size mask */ + usize_t iopt_size; + usize_t iopt_unlimited; + + /* general configuration */ + xd3_getblk_func *getblk; /* set nxtblk, nxtblkno to scanblkno */ + xd3_alloc_func *alloc; /* malloc function */ + xd3_free_func *free; /* free function */ + void* opaque; /* private data object passed to + alloc, free, and getblk */ + int flags; /* various options */ + + /* secondary compressor configuration */ + xd3_sec_cfg sec_data; /* Secondary compressor config: data */ + xd3_sec_cfg sec_inst; /* Secondary compressor config: inst */ + xd3_sec_cfg sec_addr; /* Secondary compressor config: addr */ + + xd3_smatcher smatcher; + + usize_t *large_table; /* table of large checksums */ + xd3_hash_cfg large_hash; /* large hash config */ + + usize_t *small_table; /* table of small checksums */ + xd3_slist *small_prev; /* table of previous offsets, + circular linked list */ + int small_reset; /* true if small table should + be reset */ + + xd3_hash_cfg small_hash; /* small hash config */ + xd3_addr_cache acache; /* the vcdiff address cache */ + xd3_encode_state enc_state; /* state of the encoder */ + + usize_t taroff; /* base offset of the target input */ + usize_t input_position; /* current input position */ + usize_t min_match; /* current minimum match + length, avoids redundent + matches */ + usize_t unencoded_offset; /* current input, first + * unencoded offset. this value + * is <= the first instruction's + * position in the iopt buffer, + * if there is at least one + * match in the buffer. */ + + /* SRCWIN */ + int srcwin_decided; /* boolean: true if srclen and + srcbase have been + decided. */ + int srcwin_decided_early; /* boolean: true if srclen + and srcbase were + decided early. */ + xoff_t srcwin_cksum_pos; /* Source checksum position */ + + /* MATCH */ + xd3_match_state match_state; /* encoder match state */ + xoff_t match_srcpos; /* current match source + position relative to + srcbase */ + xoff_t match_last_srcpos; /* previously attempted + * srcpos, to avoid loops. */ + xoff_t match_minaddr; /* smallest matching address to + * set window params (reset each + * window xd3_encode_reset) */ + xoff_t match_maxaddr; /* largest matching address to + * set window params (reset each + * window xd3_encode_reset) */ + usize_t match_back; /* match extends back so far */ + usize_t match_maxback; /* match extends back maximum */ + usize_t match_fwd; /* match extends forward so far */ + usize_t match_maxfwd; /* match extends forward maximum */ + + xoff_t maxsrcaddr; /* address of the last source + match (across windows) */ + + uint8_t *buf_in; /* for saving buffered input */ + usize_t buf_avail; /* amount of saved input */ + const uint8_t *buf_leftover; /* leftover content of next_in + (i.e., user's buffer) */ + usize_t buf_leftavail; /* amount of leftover content */ + + xd3_output *enc_current; /* current output buffer */ + xd3_output *enc_free; /* free output buffers */ + xd3_output *enc_heads[4]; /* array of encoded outputs: + head of chain */ + xd3_output *enc_tails[4]; /* array of encoded outputs: + tail of chain */ + uint32_t recode_adler32; /* set the adler32 checksum + * during "recode". */ + + xd3_rlist iopt_used; /* instruction optimizing buffer */ + xd3_rlist iopt_free; + xd3_rinst *iout; /* next single instruction */ + xd3_iopt_buflist *iopt_alloc; + + const uint8_t *enc_appheader; /* application header to encode */ + usize_t enc_appheadsz; /* application header size */ + + /* decoder stuff */ + xd3_decode_state dec_state; /* current DEC_XXX value */ + usize_t dec_hdr_ind; /* VCDIFF header indicator */ + usize_t dec_win_ind; /* VCDIFF window indicator */ + usize_t dec_del_ind; /* VCDIFF delta indicator */ + + uint8_t dec_magic[4]; /* First four bytes */ + usize_t dec_magicbytes; /* Magic position. */ + + usize_t dec_secondid; /* Optional secondary compressor ID. */ + + /* TODO: why decode breaks if this is usize_t? */ + uint32_t dec_codetblsz; /* Optional code table: length. */ + uint8_t *dec_codetbl; /* Optional code table: storage. */ + usize_t dec_codetblbytes; /* Optional code table: position. */ + + /* TODO: why decode breaks if this is usize_t? */ + uint32_t dec_appheadsz; /* Optional application header: + size. */ + uint8_t *dec_appheader; /* Optional application header: + storage */ + usize_t dec_appheadbytes; /* Optional application header: + position. */ + + usize_t dec_cksumbytes; /* Optional checksum: position. */ + uint8_t dec_cksum[4]; /* Optional checksum: storage. */ + uint32_t dec_adler32; /* Optional checksum: value. */ + + /* TODO: why decode breaks if this is usize_t? */ + uint32_t dec_cpylen; /* length of copy window + (VCD_SOURCE or VCD_TARGET) */ + xoff_t dec_cpyoff; /* offset of copy window + (VCD_SOURCE or VCD_TARGET) */ + /* TODO: why decode breaks if this is usize_t? */ + uint32_t dec_enclen; /* length of delta encoding */ + /* TODO: why decode breaks if this is usize_t? */ + uint32_t dec_tgtlen; /* length of target window */ + +#if USE_UINT64 + uint64_t dec_64part; /* part of a decoded uint64_t */ +#endif +#if USE_UINT32 + uint32_t dec_32part; /* part of a decoded uint32_t */ +#endif + + xoff_t dec_winstart; /* offset of the start of + current target window */ + xoff_t dec_window_count; /* == current_window + 1 in + DEC_FINISH */ + usize_t dec_winbytes; /* bytes of the three sections + so far consumed */ + usize_t dec_hdrsize; /* VCDIFF + app header size */ + + const uint8_t *dec_tgtaddrbase; /* Base of decoded target + addresses (addr >= + dec_cpylen). */ + const uint8_t *dec_cpyaddrbase; /* Base of decoded copy + addresses (addr < + dec_cpylen). */ + + usize_t dec_position; /* current decoder position + counting the cpylen + offset */ + usize_t dec_maxpos; /* maximum decoder position + counting the cpylen + offset */ + xd3_hinst dec_current1; /* current instruction */ + xd3_hinst dec_current2; /* current instruction */ + + uint8_t *dec_buffer; /* Decode buffer */ + uint8_t *dec_lastwin; /* In case of VCD_TARGET, the + last target window. */ + usize_t dec_lastlen; /* length of the last target + window */ + xoff_t dec_laststart; /* offset of the start of last + target window */ + usize_t dec_lastspace; /* allocated space of last + target window, for reuse */ + + xd3_desect inst_sect; /* staging area for decoding + window sections */ + xd3_desect addr_sect; + xd3_desect data_sect; + + xd3_code_table_func *code_table_func; + xd3_comp_table_func *comp_table_func; + const xd3_dinst *code_table; + const xd3_code_table_desc *code_table_desc; + xd3_dinst *code_table_alloc; + + /* secondary compression */ + const xd3_sec_type *sec_type; + xd3_sec_stream *sec_stream_d; + xd3_sec_stream *sec_stream_i; + xd3_sec_stream *sec_stream_a; + + /* state for reconstructing whole files (e.g., for merge), this only + * supports loading USIZE_T_MAX instructions, adds, etc. */ + xd3_whole_state whole_target; + + /* statistics */ + xoff_t n_scpy; + xoff_t n_tcpy; + xoff_t n_add; + xoff_t n_run; + + xoff_t l_scpy; + xoff_t l_tcpy; + xoff_t l_add; + xoff_t l_run; + + usize_t i_slots_used; + +#if XD3_DEBUG + usize_t large_ckcnt; + + /* memory usage */ + usize_t alloc_cnt; + usize_t free_cnt; +#endif +}; + +/************************************************************************** + PUBLIC FUNCTIONS + **************************************************************************/ + +/* This function configures an xd3_stream using the provided in-memory + * input buffer, source buffer, output buffer, and flags. The output + * array must be large enough or else ENOSPC will be returned. This + * is the simplest in-memory encoding interface. */ +int xd3_encode_memory (const uint8_t *input, + usize_t input_size, + const uint8_t *source, + usize_t source_size, + uint8_t *output_buffer, + usize_t *output_size, + usize_t avail_output, + int flags); + +/* The reverse of xd3_encode_memory. */ +int xd3_decode_memory (const uint8_t *input, + usize_t input_size, + const uint8_t *source, + usize_t source_size, + uint8_t *output_buf, + usize_t *output_size, + usize_t avail_output, + int flags); + +/* This function encodes an in-memory input using a pre-configured + * xd3_stream. This allows the caller to set a variety of options + * which are not available in the xd3_encode/decode_memory() + * functions. + * + * The output array must be large enough to hold the output or else + * ENOSPC is returned. The source (if any) should be set using + * xd3_set_source_and_size() with a single-block xd3_source. This + * calls the underlying non-blocking interfaces, + * xd3_encode/decode_input(), handling the necessary input/output + * states. This method may be considered a reference for any + * application using xd3_encode_input() directly. + * + * xd3_stream stream; + * xd3_config config; + * xd3_source src; + * + * memset (& src, 0, sizeof (src)); + * memset (& stream, 0, sizeof (stream)); + * memset (& config, 0, sizeof (config)); + * + * if (source != NULL) + * { + * src.size = source_size; + * src.blksize = source_size; + * src.curblkno = 0; + * src.onblk = source_size; + * src.curblk = source; + * src.max_winsize = source_size; + * xd3_set_source(&stream, &src); + * } + * + * config.flags = flags; + * config.winsize = input_size; + * + * ... set smatcher, appheader, encoding-table, compression-level, etc. + * + * xd3_config_stream(&stream, &config); + * xd3_encode_stream(&stream, ...); + * xd3_free_stream(&stream); + */ +int xd3_encode_stream (xd3_stream *stream, + const uint8_t *input, + usize_t input_size, + uint8_t *output, + usize_t *output_size, + usize_t avail_output); + +/* The reverse of xd3_encode_stream. */ +int xd3_decode_stream (xd3_stream *stream, + const uint8_t *input, + usize_t input_size, + uint8_t *output, + usize_t *output_size, + usize_t avail_size); + +/* This is the non-blocking interface. + * + * Handling input and output states is the same for encoding or + * decoding using the xd3_avail_input() and xd3_consume_output() + * routines, inlined below. + * + * Return values: + * + * XD3_INPUT: the process requires more input: call + * xd3_avail_input() then repeat + * + * XD3_OUTPUT: the process has more output: read stream->next_out, + * stream->avail_out, then call xd3_consume_output(), + * then repeat + * + * XD3_GOTHEADER: (decoder-only) notification returned following the + * VCDIFF header and first window header. the decoder + * may use the header to configure itself. + * + * XD3_WINSTART: a general notification returned once for each + * window except the 0-th window, which is implied by + * XD3_GOTHEADER. It is recommended to use a + * switch-stmt such as: + * + * ... + * again: + * switch ((ret = xd3_decode_input (stream))) { + * case XD3_GOTHEADER: { + * assert(stream->current_window == 0); + * stuff; + * } + * // fallthrough + * case XD3_WINSTART: { + * something(stream->current_window); + * goto again; + * } + * ... + * + * XD3_WINFINISH: a general notification, following the complete + * input & output of a window. at this point, + * stream->total_in and stream->total_out are consistent + * for either encoding or decoding. + * + * XD3_GETSRCBLK: If the xd3_getblk() callback is NULL, this value + * is returned to initiate a non-blocking source read. + */ +int xd3_decode_input (xd3_stream *stream); +int xd3_encode_input (xd3_stream *stream); + +/* The xd3_config structure is used to initialize a stream - all data + * is copied into stream so config may be a temporary variable. See + * the [documentation] or comments on the xd3_config structure. */ +int xd3_config_stream (xd3_stream *stream, + xd3_config *config); + +/* Since Xdelta3 doesn't open any files, xd3_close_stream is just an + * error check that the stream is in a proper state to be closed: this + * means the encoder is flushed and the decoder is at a window + * boundary. The application is responsible for freeing any of the + * resources it supplied. */ +int xd3_close_stream (xd3_stream *stream); + +/* This arranges for closes the stream to succeed. Does not free the + * stream.*/ +void xd3_abort_stream (xd3_stream *stream); + +/* xd3_free_stream frees all memory allocated for the stream. The + * application is responsible for freeing any of the resources it + * supplied. */ +void xd3_free_stream (xd3_stream *stream); + +/* This function informs the encoder or decoder that source matching + * (i.e., delta-compression) is possible. For encoding, this should + * be called before the first xd3_encode_input. A NULL source is + * ignored. For decoding, this should be called before the first + * window is decoded, but the appheader may be read first + * (XD3_GOTHEADER). After decoding the header, call xd3_set_source() + * if you have a source file. Note: if (stream->dec_win_ind & VCD_SOURCE) + * is true, it means the first window expects there to be a source file. + */ +int xd3_set_source (xd3_stream *stream, + xd3_source *source); + +/* If the source size is known, call this instead of xd3_set_source(). + * to avoid having stream->getblk called (and/or to avoid XD3_GETSRCBLK). + * + * Follow these steps: + xd3_source source; + memset(&source, 0, sizeof(source)); + source.blksize = size; + source.onblk = size; + source.curblk = buf; + source.curblkno = 0; + int ret = xd3_set_source_and_size(&stream, &source, size); + ... + */ +int xd3_set_source_and_size (xd3_stream *stream, + xd3_source *source, + xoff_t source_size); + +/* This should be called before the first call to xd3_encode_input() + * to include application-specific data in the VCDIFF header. */ +void xd3_set_appheader (xd3_stream *stream, + const uint8_t *data, + usize_t size); + +/* xd3_get_appheader may be called in the decoder after XD3_GOTHEADER. + * For convenience, the decoder always adds a single byte padding to + * the end of the application header, which is set to zero in case the + * application header is a string. */ +int xd3_get_appheader (xd3_stream *stream, + uint8_t **data, + usize_t *size); + +/* To generate a VCDIFF encoded delta with xd3_encode_init() from + * another format, use: + * + * xd3_encode_init_partial() -- initialze encoder state (w/o hash tables) + * xd3_init_cache() -- reset VCDIFF address cache + * xd3_found_match() -- to report a copy instruction + * + * set stream->enc_state to ENC_INSTR and call xd3_encode_input as usual. + */ +int xd3_encode_init_partial (xd3_stream *stream); +void xd3_init_cache (xd3_addr_cache* acache); +int xd3_found_match (xd3_stream *stream, + usize_t pos, usize_t size, + xoff_t addr, int is_source); + +/* Gives an error string for xdelta3-speficic errors, returns NULL for + system errors */ +const char* xd3_strerror (int ret); + +/* For convenience, zero & initialize the xd3_config structure with + specified flags. */ +static inline +void xd3_init_config (xd3_config *config, + int flags) +{ + memset (config, 0, sizeof (*config)); + config->flags = flags; +} + +/* This supplies some input to the stream. + * + * For encoding, if the input is larger than the configured window + * size (xd3_config.winsize), the entire input will be consumed and + * encoded anyway. If you wish to strictly limit the window size, + * limit the buffer passed to xd3_avail_input to the window size. + * + * For encoding, if the input is smaller than the configured window + * size (xd3_config.winsize), the library will create a window-sized + * buffer and accumulate input until a full-sized window can be + * encoded. XD3_INPUT will be returned. The input must remain valid + * until the next time xd3_encode_input() returns XD3_INPUT. + * + * For decoding, the input will be consumed entirely before XD3_INPUT + * is returned again. + */ +static inline +void xd3_avail_input (xd3_stream *stream, + const uint8_t *idata, + usize_t isize) +{ + /* Even if isize is zero, the code expects a non-NULL idata. Why? + * It uses this value to determine whether xd3_avail_input has ever + * been called. If xd3_encode_input is called before + * xd3_avail_input it will return XD3_INPUT right away without + * allocating a stream->winsize buffer. This is to avoid an + * unwanted allocation. */ + XD3_ASSERT (idata != NULL || isize == 0); + + stream->next_in = idata; + stream->avail_in = isize; +} + +/* This acknowledges receipt of output data, must be called after any + * XD3_OUTPUT return. */ +static inline +void xd3_consume_output (xd3_stream *stream) +{ + stream->avail_out = 0; +} + +/* These are set for each XD3_WINFINISH return. */ +static inline +int xd3_encoder_used_source (xd3_stream *stream) { + return stream->src != NULL && stream->src->srclen > 0; +} +static inline +xoff_t xd3_encoder_srcbase (xd3_stream *stream) { + return stream->src->srcbase; +} +static inline +usize_t xd3_encoder_srclen (xd3_stream *stream) { + return stream->src->srclen; +} + +/* Checks for legal flag changes. */ +static inline +void xd3_set_flags (xd3_stream *stream, int flags) +{ + /* The bitwise difference should contain only XD3_FLUSH or + XD3_SKIP_WINDOW */ + XD3_ASSERT(((flags ^ stream->flags) & ~(XD3_FLUSH | XD3_SKIP_WINDOW)) == 0); + stream->flags = flags; +} + +/* Gives some extra information about the latest library error, if any + * is known. */ +static inline +const char* xd3_errstring (xd3_stream *stream) +{ + return stream->msg ? stream->msg : ""; +} + + +/* 64-bit divisions are expensive, which is why we require a + * power-of-two source->blksize. To relax this restriction is + * relatively easy, see the history for xd3_blksize_div(). */ +static inline +void xd3_blksize_div (const xoff_t offset, + const xd3_source *source, + xoff_t *blkno, + usize_t *blkoff) { + *blkno = (xoff_t) (offset >> source->shiftby); + *blkoff = (usize_t) (offset & source->maskby); + XD3_ASSERT (*blkoff < source->blksize); +} + +static inline +void xd3_blksize_add (xoff_t *blkno, + usize_t *blkoff, + const xd3_source *source, + const usize_t add) +{ + usize_t blkdiff; + + /* Does not check for overflow, checked in xdelta3-decode.h. */ + *blkoff += add; + blkdiff = *blkoff >> source->shiftby; + + if (blkdiff) + { + *blkno += blkdiff; + *blkoff &= source->maskby; + } + + XD3_ASSERT (*blkoff < source->blksize); +} + +#define XD3_NOOP 0U +#define XD3_ADD 1U +#define XD3_RUN 2U +#define XD3_CPY 3U /* XD3_CPY rtypes are represented as (XD3_CPY + + * copy-mode value) */ + +#if XD3_DEBUG +#define IF_DEBUG(x) x +#else +#define IF_DEBUG(x) +#endif +#if XD3_DEBUG > 1 +#define IF_DEBUG1(x) x +#else +#define IF_DEBUG1(x) +#endif +#if XD3_DEBUG > 2 +#define IF_DEBUG2(x) x +#else +#define IF_DEBUG2(x) +#endif + +#define SIZEOF_ARRAY(x) (sizeof(x) / sizeof(x[0])) + +#endif /* _XDELTA3_H_ */