Message-Id: <201305121933.r4CJXUCv008550@delorie.com> Date: Sun, 12 May 2013 21:32:16 +0200 From: Juan Manuel Guerrero To: djgpp-announce AT delorie DOT com Subject: ANNOUNCE: DJGPP port of GNU Patch 2.7.1 uploaded. Content-Type: text/plain; charset=ISO-8859-15; format=flowed Reply-To: djgpp AT delorie DOT com This is a port of GNU patch 2.7.1 to MSDOS/DJGPP. Patch is a program to apply patches to files. It accepts a "patch script", usually produced by file-comparison utility such as Diff, and produces modified files by applying this script to old versions of files. More often than not, Patch is used with text, usually source, files. But it can also patch binary files, although the patch scripts for binary files tend to be large. DJGPP specific changes. ======================= The port offers the same djgpp specific functionality than the previous port. These are: - support for numbered backups even on 8+3 file systems; - Patch can now ignore the CR characters at the end of each line of the patch script, so patching Unix-style text files does not convert them to DOS-style files anymore (if you use the --binary switch). To build this port you will need: - to have an OS with LFN support to configure and compile the source package. I have no plans to support SFN systems to build the patch binaries. - to run the test suite you will need to install the latest mktmpNNb.zip As usual, all djgpp specific files (config.bat, diffs, README files, etc.) are located in the /djgpp directory. Please note that there is a directory called /pc/djgpp that contains files that are part of the official FSF patch distribution. I have inspected those files but I have never used them nor I have plans to change this in the future. Due to the massive use of gnulib code it is no longer possible to compile the official FSF distribution out-of-the-box. So I will neither try to use those files nor to keep them up to date. I will only use my own ones. If someone else would like to keep those files up to date it will be OK with me. Please also note that some of the tests of the test suite will fail. This is not a port issue but a test suite issue. The test suite is very posix centric so that some test will fail in an environment like DOS/Windows. The most notorious test to fail is crlf-handling. Here the behavior of patch in a posix environment is tested and how it works when files to patch or the diffs file has DOS style EOL. This is not the same behavior than a DOS/DJGPP port of patch to work in a DOS/WINDOWS environment. Not all those test will fail but some of them. It makes no sense to try to adjust them because they test for a behaviour that the port will not offer. Other tests that will fully or partially fail are preserve-mode-and-timestamp, file-modes and fifo. For the DJGPP 2.03 port also the symlink test will completely fail. This is no surprise because DJGPP 2.03 has no real symlink support. To implement symlink support for the DJGPP 2.04 port, among other things, the use of __solve_symlinks and __solve_dir_symlinks is required. Because these functions had some bugs that have been fixed in the repository, I have decided to compile the 2.04 port using a freshly build library from the current cvs repository. If you want to recompile this port version you will have to checkout the sources from the repository and compile the library yourself. One of the most notorious change between this patch version and the previous one is that this one uses a hash table to administrate the names of the files it wants to modify or create. The hash keys are created using the st_dev and st_ino values. Especially DJGPP creates the st_ino value using the starting cluster of the file. This has the consequence that every time the file is changed by the patch program by appending some new contents it gets a new st_ino. This has the consequence that stat()ing again produces a different st_ino for the same file and indeed a new hash key that makes it impossible to find the already existing entry for this file in the hash table. This has the nasty consequence, that when patch appends something new to this file, it does not see that the file already exists and instead of appending to it, it creates a new one at the same location with the same name overwriting the old one with the new contains. To solve this issue it was necessary to create unique and persistent inode numbers for every file produced by patch. Wrapper functions to stat, fstat, open, fopen, close and fclose have been written that will keep in sync file name, file descriptor and invented inode number for every file produced by patch. See /pc/pc_inode.c for details concerning the implementation. Another change is that the patch program relaies on some explicit values of the file type bits stored in the st_mode member of stat. Especially it is assumed that every S_IF* bit is encoded by 1 and not by 0. But unfortunately this is the case in DJGPP. DJGPP encodes S_IFREG with zero. This makes the code fail. To avoid to have to change alot of code lines, S_IFREG is mapped to another value different from zero before used. This port supports git-style of patch files. Unfortunately this type of patches encode the file type (like regular file or symlink) of the files to be patched as number in the header of the patch file. This means that the patch program reads a git-style patch file and parses the header and tries to determinate if the file to be patched will be a regular file (S_IFREG) or a symlink (S_IFLNK) or something else. This means that the values of S_IFREG or S_IFLNK are not stored as some kind of symbols but as those real numbers used to encode them on the OS where the patch file has been produced. This has the consequence that if on a different OS/compiler this encoding does not match, the patch program will not work properly. An inspection of sys/stat.h of linux, bsd and cygwin shows that they all use the same encodings for all values stored in the st_mode member of the stat structure. Unfortunately DJGPP encodes these values in a incompatible way. When this port parses that file type number, it _ALWAYS_ interprets it as linux/bsd/cygwin encoding of the S_IF* bits and maps it to the DJGPP specifc values of those bits. Please note, that the port has no way to determinate if the number really represents linux/bsd/cygwin encoding or not. If the number is not a linux/bsd/cygwin encoding, the mapping will fail an the patch cannot be applied. The port assumes that the following encoding for the S_IF* bits is valid: /* Encoding of the file mode. */ #define S_IFMT 0170000 /* These bits determine file type. */ /* File types. */ #define S_IFDIR 0040000 /* Directory. */ #define S_IFCHR 0020000 /* Character device. */ #define S_IFBLK 0060000 /* Block device. */ #define S_IFREG 0100000 /* Regular file. */ #define S_IFIFO 0010000 /* FIFO. */ #define S_IFLNK 0120000 /* Symbolic link. */ #define S_IFSOCK 0140000 /* Socket. */ If this assumption is not true, the mapping to DJGPP values of S_IF* bits will fail. Please not that the 2.03 port of patch will always fail applying git-style patches because it does not support symlinks at all. Neitherless I do not think that this is a great inconvenience because I think that no one will ever port and use git on DOS at all. Please note that the FSF distribution of patch no longer can be compiled with DJGPP out-of-the-box. If something seems not to work with this port, please report it first at c.o.m.d. and not at bug-patch AT gnu DOT org. The maintainer will have no idea about what I have done. The source package is now distributed configured for both DJGPP 2.03 and DJGPP 2.04. In the top srcdir there is a "_build.203" directory and a "_build.204" directory. For further information about GNU patch please read the info docs and NEWS file. Here is an extract of the NEWS file showing the user visible changes from the last port (patch 2.6.1) to this one: Changes in version 2.7.1: * Two critical bug fixes in the "diff --git" format support. * Clarify the message printed when a patch is expected to empty out and delete a file, but the file does not become empty. * Various improvements to messages when applying a patch to a file of different type (regular file vs. symlink), when there are line ending differences (LF vs. CRLF), and when in --dry-run mode. * When in the root directory, allow file names that are absolute or that contain a component of "..". * New --follow-symlinks option to allow to treat symlinks as files: this was patch's behavior before version 2.7. * Ignore when extended attributes cannot be preserved because they are unsupported or because permission to set them is denied. * License clarifications in NEWS and README. * Portability bug fixes. Changes in version 2.7: * Patch no longer gets a failed assertion for certain mangled patches. * Ignore destination file names that are absolute or that contain a component of "..". This addresses CVE-2010-4651. * Support for most features of the "diff --git" format, including renames and copies, permission changes, and symlink diffs. Binary diffs are not supported yet; patch will complain and skip them. * Support for double-quoted filenames: when a filename starts with a double quote, it is interpreted as a C string literal. The escape sequences \\, \", \a, \b, \f, \n, \r, \t, \v, and \ooo (a three-digit octal number between 0 and 255) are recognized. * Refuse to apply a normal patch to a symlink. (Previous versions of patch were replacing the symlink with a regular file.) * When trying to modify a read-only file, warn about the potential problem by default. The --read-only command line option allows to change this behavior. * Files to be deleted are deleted once the entire input has been processed, not immediately. This fixes a bug with numbered backup files. * When a timestamp specifies a time zone, honor that instead of assuming the local time zone (--set-date) or Universal Coordinated Time (--set-utc). * Support for nanosecond precision timestamps. * Many portability and bug fixes. ---------------------------------------------------------------------------- The port has been compiled using stock djdev203 (patchlevel 2) and consists of the two packages that can be downloaded from ftp.delorie.com and mirrors as (time stamp 2013-05-10): Patch 2.7.1 binary and man format documentation: ftp://ftp.delorie.com/pub/djgpp/current/v2gnu/pat271b.zip Patch 2.7.1 source: ftp://ftp.delorie.com/pub/djgpp/current/v2gnu/pat271s.zip The binaries have been produced a second time using a freshly compiled DJGPP library from the cvs repository. This package is available at ftp.delorie.com and mirrors as (time stamp 2013-05-10): Patch 2.7.1 binary and man format documentation: ftp://ftp.delorie.com/pub/djgpp/beta/v2gnu/pat271b.zip Send Patch specific bug reports to . Send suggestions and bug reports concerning the DJGPP port to comp.os.msdos.djgpp or . Enjoy. Guerrero, Juan Manuel