XZ Utils Installation ===================== 0. Preface 1. Supported platforms 1.1. Compilers 1.2. Platform-specific notes 1.2.1. AIX 1.2.2. IRIX 1.2.3. MINIX 3 1.2.4. OpenVMS 1.2.5. Solaris, OpenSolaris, and derivatives 1.2.6. Tru64 1.2.7. Windows 1.2.8. DOS 1.2.9. z/OS 1.3. Adding support for new platforms 2. configure options 2.1. Static vs. dynamic linking of liblzma 2.2. Optimizing xzdec and lzmadec 3. xzgrep and other scripts 3.1. Dependencies 3.2. PATH 4. Tests 4.1 Testing in parallel 4.2 Cross compiling 5. Troubleshooting 5.1. "No C99 compiler was found." 5.2. "No POSIX conforming shell (sh) was found." 5.3. configure works but build fails at crc32_x86.S 5.4. Lots of warnings about symbol visibility 5.5. "make check" fails 5.6. liblzma.so (or similar) not found when running xz 0. Preface ---------- If you aren't familiar with building packages that use GNU Autotools, see the file INSTALL.generic for generic instructions before reading further. If you are going to build a package for distribution, see also the file PACKAGERS. It contains information that should help making the binary packages as good as possible, but the information isn't very interesting to those making local builds for private use or for use in special situations like embedded systems. 1. Supported platforms ---------------------- XZ Utils are developed on GNU/Linux, but they should work on many POSIX-like operating systems like *BSDs and Solaris, and even on a few non-POSIX operating systems. 1.1. Compilers A C99 compiler is required to compile XZ Utils. If you use GCC, you need at least version 3.x.x. GCC version 2.xx.x doesn't support some C99 features used in XZ Utils source code, thus GCC 2 won't compile XZ Utils. XZ Utils takes advantage of some GNU C extensions when building with GCC. Because these extensions are used only when building with GCC, it should be possible to use any C99 compiler. 1.2. Platform-specific notes 1.2.1. AIX If you use IBM XL C compiler, pass CC=xlc_r to configure. If you use CC=xlc instead, you must disable threading support with --disable-threads (usually not recommended). 1.2.2. IRIX MIPSpro 7.4.4m has been reported to produce broken code if using the -O2 optimization flag ("make check" fails). Using -O1 should work. A problem has been reported when using shared liblzma. Passing --disable-shared to configure works around this. Alternatively, putting "-64" to CFLAGS to build a 64-bit version might help too. 1.2.3. MINIX 3 The default install of MINIX 3 includes Amsterdam Compiler Kit (ACK), which doesn't support C99. Install GCC to compile XZ Utils. MINIX 3.1.8 and older have bugs in /usr/include/stdint.h, which has to be patched before XZ Utils can be compiled correctly. See . MINIX 3.2.0 and later use a different libc and aren't affected by the above bug. XZ Utils doesn't have code to detect the amount of physical RAM and number of CPU cores on MINIX 3. See section 5.4 in this file about symbol visibility warnings (you may want to pass gl_cv_cc_visibility=no to configure). 1.2.4. OpenVMS XZ Utils can be built for OpenVMS, but the build system files are not included in the XZ Utils source package. The required OpenVMS-specific files are maintained by Jouk Jansen and can be downloaded here: http://nchrem.tnw.tudelft.nl/openvms/software2.html#xzutils 1.2.5. Solaris, OpenSolaris, and derivatives The following linker error has been reported on some x86 systems: ld: fatal: relocation error: R_386_GOTOFF: ... This can be worked around by passing gl_cv_cc_visibility=no as an argument to the configure script. test_scripts.sh in "make check" may fail if good enough tools are missing from PATH (/usr/xpg4/bin or /usr/xpg6/bin). Nowadays /usr/xpg4/bin is added to the script PATH by default on Solaris (see --enable-path-for-scripts=PREFIX in section 2), but old xz releases needed extra steps. See sections 5.5 and 3.2 for more information. 1.2.6. Tru64 If you try to use the native C compiler on Tru64 (passing CC=cc to configure), you may need the workaround mention in section 5.1 in this file (pass also ac_cv_prog_cc_c99= to configure). 1.2.7. Windows If it is enough to build liblzma (no command line tools): - There is CMake support. It should be good enough to build static liblzma or liblzma.dll with Visual Studio. The CMake support may work with MinGW or MinGW-w64. Read the comment in the beginning of CMakeLists.txt before running CMake! - There are Visual Studio project files under the "windows" directory. See windows/INSTALL-MSVC.txt. In the future the project files will be removed when CMake support is good enough. Thus, please test the CMake version and help fix possible issues. To build also the command line tools: - MinGW-w64 + MSYS (32-bit and 64-bit x86): This is used for building the official binary packages for Windows. There is windows/build.bash to ease packaging XZ Utils with MinGW(-w64) + MSYS into a redistributable .zip or .7z file. See windows/INSTALL-MinGW.txt for more information. - MinGW + MSYS (32-bit x86): I haven't recently tested this. - Cygwin 1.7.35 and later: NOTE that using XZ Utils >= 5.2.0 under Cygwin older than 1.7.35 can lead to DATA LOSS! If you must use an old Cygwin version, stick to XZ Utils 5.0.x which is safe under older Cygwin versions. You can check the Cygwin version with the command "cygcheck -V". It may be possible to build liblzma with other toolchains too, but that will probably require writing a separate makefile. Building the command line tools with non-GNU toolchains will be harder than building only liblzma. Even if liblzma is built with MinGW(-w64), the resulting DLL can be used by other compilers and linkers, including MSVC. See windows/README-Windows.txt for details. 1.2.8. DOS There is a Makefile in the "dos" directory to build XZ Utils on DOS using DJGPP. Support for long file names (LFN) is needed at build time but the resulting xz.exe works without LFN support too. See dos/INSTALL.txt and dos/README.txt for more information. 1.2.9. z/OS To build XZ Utils on z/OS UNIX System Services using xlc, pass these options to the configure script: CC='xlc -qhaltonmsg=CCN3296' CPPFLAS='-D_UNIX03_THREADS -D_XOPEN_SOURCE=600'. The first makes xlc throw an error if a header file is missing, which is required to make the tests in configure work. The CPPFLAGS are needed to get pthread support (some other CPPFLAGS may work too; if there are problems, try -D_UNIX95_THREADS instead of -D_UNIX03_THREADS). test_scripts.sh in "make check" will fail even if the scripts actually work because the test data includes compressed files with US-ASCII text. No other tests should fail. If test_files.sh fails, check that the included .xz test files weren't affected by EBCDIC conversion. XZ Utils doesn't have code to detect the amount of physical RAM and number of CPU cores on z/OS. 1.3. Adding support for new platforms If you have written patches to make XZ Utils to work on previously unsupported platform, please send the patches to me! I will consider including them to the official version. It's nice to minimize the need of third-party patching. One exception: Don't request or send patches to change the whole source package to C89. I find C99 substantially nicer to write and maintain. However, the public library headers must be in C89 to avoid frustrating those who maintain programs, which are strictly in C89 or C++. 2. configure options -------------------- In most cases, the defaults are what you want. Many of the options below are useful only when building a size-optimized version of liblzma or command line tools. --enable-encoders=LIST --disable-encoders Specify a comma-separated LIST of filter encoders to build. See "./configure --help" for exact list of available filter encoders. The default is to build all supported encoders. If LIST is empty or --disable-encoders is used, no filter encoders will be built and also the code shared between encoders will be omitted. Disabling encoders will remove some symbols from the liblzma ABI, so this option should be used only when it is known to not cause problems. --enable-decoders=LIST --disable-decoders This is like --enable-encoders but for decoders. The default is to build all supported decoders. --enable-match-finders=LIST liblzma includes two categories of match finders: hash chains and binary trees. Hash chains (hc3 and hc4) are quite fast but they don't provide the best compression ratio. Binary trees (bt2, bt3 and bt4) give excellent compression ratio, but they are slower and need more memory than hash chains. You need to enable at least one match finder to build the LZMA1 or LZMA2 filter encoders. Usually hash chains are used only in the fast mode, while binary trees are used to when the best compression ratio is wanted. The default is to build all the match finders if LZMA1 or LZMA2 filter encoders are being built. --enable-checks=LIST liblzma support multiple integrity checks. CRC32 is mandatory, and cannot be omitted. See "./configure --help" for exact list of available integrity check types. liblzma and the command line tools can decompress files which use unsupported integrity check type, but naturally the file integrity cannot be verified in that case. Disabling integrity checks may remove some symbols from the liblzma ABI, so this option should be used only when it is known to not cause problems. --enable-external-sha256 Try to use SHA-256 code from the operating system libc or similar base system libraries. This doesn't try to use OpenSSL or libgcrypt or such libraries. The reasons to use this option: - It makes liblzma slightly smaller. - It might improve SHA-256 speed if the implementation in the operating is very good (but see below). External SHA-256 is disabled by default for two reasons: - On some operating systems the symbol names of the SHA-256 functions conflict with OpenSSL's libcrypto. This causes weird problems such as decompression errors if an application is linked against both liblzma and libcrypto. This problem affects at least FreeBSD 10 and older and MINIX 3.3.0 and older, but other OSes that provide a function "SHA256_Init" might also be affected. FreeBSD 11 has the problem fixed. NetBSD had the problem but it was fixed it in 2009 already. OpenBSD uses "SHA256Init" and thus never had a conflict with libcrypto. - The SHA-256 code in liblzma is faster than the SHA-256 code provided by some operating systems. If you are curious, build two copies of xz (internal and external SHA-256) and compare the decompression (xz --test) times: dd if=/dev/zero bs=1024k count=1024 \ | xz -v -0 -Csha256 > foo.xz time xz --test foo.xz --disable-microlzma Don't build MicroLZMA encoder and decoder. This omits lzma_microlzma_encoder() and lzma_microlzma_decoder() API functions from liblzma. These functions are needed by specific applications only. They were written for erofs-utils but they may be used by others too. --disable-lzip-decoder Disable decompression support for .lz (lzip) files. This omits the API function lzma_lzip_decoder() from liblzma and .lz support from the xz tool. --disable-xz --disable-xzdec --disable-lzmadec --disable-lzmainfo Don't build and install the command line tool mentioned in the option name. NOTE: Disabling xz will skip some tests in "make check". NOTE: If xzdec is disabled and lzmadec is left enabled, a dangling man page symlink lzmadec.1 -> xzdec.1 is created. --disable-lzma-links Don't create symlinks for LZMA Utils compatibility. This includes lzma, unlzma, and lzcat. If scripts are installed, also lzdiff, lzcmp, lzgrep, lzegrep, lzfgrep, lzmore, and lzless will be omitted if this option is used. --disable-scripts Don't install the scripts xzdiff, xzgrep, xzmore, xzless, and their symlinks. --disable-doc Don't install the documentation files to $docdir (often /usr/doc/xz or /usr/local/doc/xz). Man pages will still be installed. The $docdir can be changed with --docdir=DIR. --disable-assembler This disables CRC32 and CRC64 assembly code on 32-bit x86. This option currently does nothing on other architectures (not even on x86-64). The 32-bit x86 assembly is position-independent code which is suitable for use in shared libraries and position-independent executables. It uses only i386 instructions but the code is optimized for i686 class CPUs. If you are compiling liblzma exclusively for pre-i686 systems, you may want to disable the assembler code. --disable-clmul-crc Disable the use carryless multiplication for CRC calculation even if compiler support for it is detected. The code uses runtime detection of SSSE3, SSE4.1, and CLMUL instructions on x86. On 32-bit x86 this currently is used only if --disable-assembler is used (this might be fixed in the future). The code works on E2K too. If using compiler options that unconditionally allow the required extensions (-msse4.1 -mpclmul) then runtime detection isn't used and the generic code is omitted. --enable-unaligned-access Allow liblzma to use unaligned memory access for 16-bit, 32-bit, and 64-bit loads and stores. This should be enabled only when the hardware supports this, that is, when unaligned access is fast. Some operating system kernels emulate unaligned access, which is extremely slow. This option shouldn't be used on systems that rely on such emulation. Unaligned access is enabled by default on x86, x86-64, big endian PowerPC, some ARM, and some ARM64 systems. --enable-unsafe-type-punning This enables use of code like uint8_t *buf8 = ...; *(uint32_t *)buf8 = ...; which violates strict aliasing rules and may result in broken code. There should be no need to use this option with recent GCC or Clang versions on any arch as just as fast code can be generated in a safe way too (using __builtin_assume_aligned + memcpy). However, this option might improve performance in some other cases, especially with old compilers (for example, GCC 3 and early 4.x on x86, GCC < 6 on ARMv6 and ARMv7). --enable-small Reduce the size of liblzma by selecting smaller but semantically equivalent version of some functions, and omit precomputed lookup tables. This option tends to make liblzma slightly slower. Note that while omitting the precomputed tables makes liblzma smaller on disk, the tables are still needed at run time, and need to be computed at startup. This also means that the RAM holding the tables won't be shared between applications linked against shared liblzma. This option doesn't modify CFLAGS to tell the compiler to optimize for size. You need to add -Os or equivalent flag(s) to CFLAGS manually. --enable-assume-ram=SIZE On the most common operating systems, XZ Utils is able to detect the amount of physical memory on the system. This information is used by the options --memlimit-compress, --memlimit-decompress, and --memlimit when setting the limit to a percentage of total RAM. On some systems, there is no code to detect the amount of RAM though. Using --enable-assume-ram one can set how much memory to assume on these systems. SIZE is given as MiB. The default is 128 MiB. Feel free to send patches to add support for detecting the amount of RAM on the operating system you use. See src/common/tuklib_physmem.c for details. --enable-threads=METHOD Threading support is enabled by default so normally there is no need to specify this option. Supported values for METHOD: yes Autodetect the threading method. If none is found, configure will give an error. posix Use POSIX pthreads. This is the default except on Windows outside Cygwin. win95 Use Windows 95 compatible threads. This is compatible with Windows XP and later too. This is the default for 32-bit x86 Windows builds. Unless the compiler supports __attribute__((__constructor__)), the `win95' threading is incompatible with --enable-small. vista Use Windows Vista compatible threads. The resulting binaries won't run on Windows XP or older. This is the default for Windows excluding 32-bit x86 builds (that is, on x86-64 the default is `vista'). no Disable threading support. This is the same as using --disable-threads. NOTE: If combined with --enable-small and the compiler doesn't support __attribute__((__constructor__)), the resulting liblzma won't be thread safe, that is, if a multi-threaded application calls any liblzma functions from more than one thread, something bad may happen. --enable-sandbox=METHOD There is limited sandboxing support in the xz tool. If built with sandbox support, it's used automatically when (de)compressing exactly one file to standard output and the options --files or --files0 weren't used. This is a common use case, for example, (de)compressing .tar.xz files via GNU tar. The sandbox is also used for single-file `xz --test' or `xz --list'. Supported METHODs: auto Look for a supported sandboxing method and use it if found. If no method is found, then sandboxing isn't used. This is the default. no Disable sandboxing support. capsicum Use Capsicum (FreeBSD >= 10) for sandboxing. If no Capsicum support is found, configure will give an error. pledge Use pledge(2) (OpenBSD >= 5.9) for sandboxing. If pledge(2) isn't found, configure will give an error. --enable-symbol-versions[=VARIANT] Use symbol versioning for liblzma shared library. This is enabled by default on GNU/Linux (glibc only), other GNU-based systems, and FreeBSD. Symbol versioning is never used for static liblzma. This option is ignored when not building a shared library. Supported VARIANTs: no Disable symbol versioning. This is the same as using --disable-symbol-versions. auto Autodetect between "no", "linux", and "generic". yes Autodetect between "linux" and "generic". This forces symbol versioning to be used when building a shared library. generic Generic version is the default for FreeBSD and GNU/Linux on MicroBlaze. This is also used on GNU/Linux when building with NVIDIA HPC Compiler because the compiler doesn't support the features required for the "linux" variant below. linux Special version for GNU/Linux (glibc only). This adds a few extra symbol versions for compatibility with binaries that have been linked against a liblzma version that has been patched with "xz-5.2.2-compat-libs.patch" from RHEL/CentOS 7. That patch was used by some build tools outside of RHEL/CentOS 7 too. --enable-debug This enables the assert() macro and possibly some other run-time consistency checks. It makes the code slower, so you normally don't want to have this enabled. --enable-werror If building with GCC, make all compiler warnings an error, that abort the compilation. This may help catching bugs, and should work on most systems. This has no effect on the resulting binaries. --enable-path-for-scripts=PREFIX If PREFIX isn't empty, PATH=PREFIX:$PATH will be set in the beginning of the scripts (xzgrep and others). The default is empty except on Solaris the default is /usr/xpg4/bin. This can be useful if the default PATH doesn't contain modern POSIX tools (as can be the case on Solaris) or if one wants to ensure that the correct xz binary is in the PATH for the scripts. Note that the latter use can break "make check" if the prefixed PATH causes a wrong xz binary (other than the one that was just built) to be used. Older xz releases support a different method for setting the PATH for the scripts. It is described in section 3.2 and is supported in this xz version too. 2.1. Static vs. dynamic linking of liblzma On 32-bit x86, linking against static liblzma can give a minor speed improvement. Static libraries on x86 are usually compiled as position-dependent code (non-PIC) and shared libraries are built as position-independent code (PIC). PIC wastes one register, which can make the code slightly slower compared to a non-PIC version. (Note that this doesn't apply to x86-64.) If you want to link xz against static liblzma, the simplest way is to pass --disable-shared to configure. If you want also shared liblzma, run configure again and run "make install" only for src/liblzma. 2.2. Optimizing xzdec and lzmadec xzdec and lzmadec are intended to be relatively small instead of optimizing for the best speed. Thus, it is a good idea to build xzdec and lzmadec separately: - To link the tools against static liblzma, pass --disable-shared to configure. - To select somewhat size-optimized variant of some things in liblzma, pass --enable-small to configure. - Tell the compiler to optimize for size instead of speed. For example, with GCC, put -Os into CFLAGS. - xzdec and lzmadec will never use multithreading capabilities of liblzma. You can avoid dependency on libpthread by passing --disable-threads to configure. - There are and will be no translated messages for xzdec and lzmadec, so it is fine to pass also --disable-nls to configure. - Only decoder code is needed, so you can speed up the build slightly by passing --disable-encoders to configure. This shouldn't affect the final size of the executables though, because the linker is able to omit the encoder code anyway. If you have no use for xzdec or lzmadec, you can disable them with --disable-xzdec and --disable-lzmadec. 3. xzgrep and other scripts --------------------------- 3.1. Dependencies POSIX shell (sh) and bunch of other standard POSIX tools are required to run the scripts. The configure script tries to find a POSIX compliant sh, but if it fails, you can force the shell by passing gl_cv_posix_shell=/path/to/posix-sh as an argument to the configure script. xzdiff (xzcmp/lzdiff/lzcmp) may use mktemp if it is available. As a fallback xzdiff will use mkdir to securely create a temporary directory. Having mktemp available is still recommended since the mkdir fallback method isn't as robust as mktemp is. The original mktemp can be found from . On GNU, most will use the mktemp program from GNU coreutils instead of the original implementation. Both mktemp versions are fine. In addition to using xz to decompress .xz files, xzgrep and xzdiff use gzip, bzip2, and lzop to support .gz, bz2, and .lzo files. 3.2. PATH The method described below is supported by older xz releases. It is supported by the current version too, but the newer --enable-path-for-scripts=PREFIX described in section 2 may be more convenient. The scripts assume that the required tools (standard POSIX utilities, mktemp, and xz) are in PATH; the scripts don't set the PATH themselves (except as described for --enable-path-for-scripts=PREFIX). Some people like this while some think this is a bug. Those in the latter group can easily patch the scripts before running the configure script by taking advantage of a placeholder line in the scripts. For example, to make the scripts prefix /usr/bin:/bin to PATH: perl -pi -e 's|^#SET_PATH.*$|PATH=/usr/bin:/bin:\$PATH|' \ src/scripts/xz*.in 4. Tests -------- The test framework can be built and run by executing "make check" in the build directory. The tests are a mix of executables and POSIX shell scripts (sh). All tests should pass if the default configuration is used. Disabling features through the configure options may cause some tests to be skipped. If any tests do not pass, see section 5.5. 4.1. Testing in parallel The tests can be run in parallel using the "-j" make option on systems that support it. For instance, "make -j4 check" will run up to four tests simultaneously. 4.2. Cross compiling The tests can be built without running them: make check TESTS= The TESTS variable is the list of tests you wish to run. Leaving it empty will compile the tests without running any. If the tests are copied to a target machine to execute, the test data files in the directory tests/files must also be copied. The tests search for the data files using the environment variable $srcdir, expecting to find the data files under $srcdir/files/. If $srcdir isn't set then it defaults to the current directory. The shell script tests can be copied from the source directory to the target machine to execute. In addition to the test files, these tests will expect the following relative file paths to execute properly: ./create_compress_files ../config.h ../src/xz/xz ../src/xzdec/xzdec ../src/scripts/xzdiff ../src/scripts/xzgrep 5. Troubleshooting ------------------ 5.1. "No C99 compiler was found." You need a C99 compiler to build XZ Utils. If the configure script cannot find a C99 compiler and you think you have such a compiler installed, set the compiler command by passing CC=/path/to/c99 as an argument to the configure script. If you get this error even when you think your compiler supports C99, you can override the test by passing ac_cv_prog_cc_c99= as an argument to the configure script. The test for C99 compiler is not perfect (and it is not as easy to make it perfect as it sounds), so sometimes this may be needed. You will get a compile error if your compiler doesn't support enough C99. 5.2. "No POSIX conforming shell (sh) was found." xzgrep and other scripts need a shell that (roughly) conforms to POSIX. The configure script tries to find such a shell. If it fails, you can force the shell to be used by passing gl_cv_posix_shell=/path/to/posix-sh as an argument to the configure script. Alternatively you can omit the installation of scripts and this error by passing --disable-scripts to configure. 5.3. configure works but build fails at crc32_x86.S The easy fix is to pass --disable-assembler to the configure script. The configure script determines if assembler code can be used by looking at the configure triplet; there is currently no check if the assembler code can actually actually be built. The x86 assembler code should work on x86 GNU/Linux, *BSDs, Solaris, Darwin, MinGW, Cygwin, and DJGPP. On other x86 systems, there may be problems and the assembler code may need to be disabled with the configure option. If you get this error when building for x86-64, you have specified or the configure script has misguessed your architecture. Pass the correct configure triplet using the --build=CPU-COMPANY-SYSTEM option (see INSTALL.generic). 5.4. Lots of warnings about symbol visibility On some systems where symbol visibility isn't supported, GCC may still accept the visibility options and attributes, which will make configure think that visibility is supported. This will result in many compiler warnings. You can avoid the warnings by forcing the visibility support off by passing gl_cv_cc_visibility=no as an argument to the configure script. This has no effect on the resulting binaries, but fewer warnings looks nicer and may allow using --enable-werror. 5.5. "make check" fails If the other tests pass but test_scripts.sh fails, then the problem is in the scripts in src/scripts. Comparing the contents of tests/xzgrep_test_output to tests/xzgrep_expected_output might give a good idea about problems in xzgrep. One possibility is that some tools are missing from the current PATH or the tools lack support for some POSIX features. This can happen at least on Solaris where the tools in /bin may be ancient but good enough tools are available in /usr/xpg4/bin or /usr/xpg6/bin. For possible fixes, see --enable-path-for-scripts=PREFIX in section 2 and the older alternative method described in section 3.2 of this file. If tests other than test_scripts.sh fail, a likely reason is that libtool links the test programs against an installed version of liblzma instead of the version that was just built. This is obviously a bug which seems to happen on some platforms. A workaround is to uninstall the old liblzma versions first. If the problem isn't any of those described above, then it's likely a bug in XZ Utils or in the compiler. See the platform-specific notes in this file for possible known problems. Please report a bug if you cannot solve the problem. See README for contact information. 5.6. liblzma.so (or similar) not found when running xz If you installed the package with "make install" and get an error about liblzma.so (or a similarly named file) being missing, try running "ldconfig" to update the run-time linker cache (if your operating system has such a command).