Vendor import of xz 5.4.5 (trimmed)vendor/xz/5.4.5

33 files changed, 1295 insertions, 168 deletions

diff --git a/ChangeLog b/ChangeLog
index 64c79dbce0bc..4dd0978ef313 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,4 +1,870 @@
-commit c1e396a9ac1c1c28ce4ede5cbadb955c516477bc
+commit 49053c0a649f4c8bd2b8d97ce915f401fbc0f3d9
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-10-31 22:30:29 +0800
+
+    Bump version and soname for 5.4.5.
+
+ src/liblzma/Makefile.am        | 2 +-
+ src/liblzma/api/lzma/version.h | 2 +-
+ 2 files changed, 2 insertions(+), 2 deletions(-)
+
+commit 84c0cfc556287628df871703672879e530d0391f
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-11-01 20:18:30 +0800
+
+    Add NEWS for 5.4.5.
+
+ NEWS | 74 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+ 1 file changed, 74 insertions(+)
+
+commit d90ed84db9770712e2421e170076b43bda9b64a7
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-10-31 21:41:09 +0200
+
+    liblzma: Fix compilation of fastpos_tablegen.c.
+    
+    The macro lzma_attr_visibility_hidden has to be defined to make
+    fastpos.h usable. The visibility attribute is irrelevant to
+    fastpos_tablegen.c so simply #define the macro to an empty value.
+    
+    fastpos_tablegen.c is never built by the included build systems
+    and so the problem wasn't noticed earlier. It's just a standalone
+    program for generating fastpos_table.c.
+    
+    Fixes: https://github.com/tukaani-project/xz/pull/69
+    Thanks to GitHub user Jamaika1.
+
+ src/liblzma/lzma/fastpos_tablegen.c | 2 ++
+ 1 file changed, 2 insertions(+)
+
+commit 9b1268538b0b2c6c0a121f95165de65fc71ad23c
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-10-31 21:51:40 +0800
+
+    Build: Fix text wrapping in an output message.
+
+ configure.ac | 9 +++++----
+ 1 file changed, 5 insertions(+), 4 deletions(-)
+
+commit 068ee436f4a8a706125ef43e8228b30001b1554e
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-10-22 17:59:11 +0300
+
+    liblzma: Use lzma_always_inline in memcmplen.h.
+
+ src/liblzma/common/memcmplen.h | 3 +--
+ 1 file changed, 1 insertion(+), 2 deletions(-)
+
+commit 6cdf0a7b7974baf58c1fd20ec3278f3b84ae56e5
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-10-30 17:43:03 +0200
+
+    liblzma: #define lzma_always_inline in common.h.
+
+ src/liblzma/common/common.h | 17 +++++++++++++++++
+ 1 file changed, 17 insertions(+)
+
+commit 33daad3961a4f07f3902b40f13e823e6e43e85da
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-10-22 17:15:32 +0300
+
+    liblzma: Use lzma_attr_visibility_hidden on private extern declarations.
+    
+    These variables are internal to liblzma and not exposed in the API.
+
+ src/liblzma/check/check.h                | 7 +++++++
+ src/liblzma/common/stream_flags_common.h | 3 +++
+ src/liblzma/lz/lz_encoder_hash.h         | 1 +
+ src/liblzma/lzma/fastpos.h               | 1 +
+ src/liblzma/rangecoder/price.h           | 1 +
+ 5 files changed, 13 insertions(+)
+
+commit 6961a5ac7df178bfc2b7a181c40575847bc3035f
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-10-22 17:08:39 +0300
+
+    liblzma: #define lzma_attr_visibility_hidden in common.h.
+    
+    In ELF shared libs:
+    
+    -fvisibility=hidden affects definitions of symbols but not
+    declarations.[*] This doesn't affect direct calls to functions
+    inside liblzma as a linker can replace a call to lzma_foo@plt
+    with a call directly to lzma_foo when -fvisibility=hidden is used.
+    
+    [*] It has to be like this because otherwise every installed
+        header file would need to explictly set the symbol visibility
+        to default.
+    
+    When accessing extern variables that aren't defined in the
+    same translation unit, compiler assumes that the variable has
+    the default visibility and thus indirection is needed. Unlike
+    function calls, linker cannot optimize this.
+    
+    Using __attribute__((__visibility__("hidden"))) with the extern
+    variable declarations tells the compiler that indirection isn't
+    needed because the definition is in the same shared library.
+    
+    About 15+ years ago, someone told me that it would be good if
+    the CRC tables would be defined in the same translation unit
+    as the C code of the CRC functions. While I understood that it
+    could help a tiny amount, I didn't want to change the code because
+    a separate translation unit for the CRC tables was needed for the
+    x86 assembly code anyway. But when visibility attributes are
+    supported, simply marking the extern declaration with the
+    hidden attribute will get identical result. When there are only
+    a few affected variables, this is trivial to do. I wish I had
+    understood this back then already.
+
+ src/liblzma/common/common.h | 11 +++++++++++
+ 1 file changed, 11 insertions(+)
+
+commit 5b9e16764905d06fa8e8339ba185ddfee304e5fb
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-30 22:54:28 +0300
+
+    liblzma: Refer to MinGW-w64 instead of MinGW in the API headers.
+    
+    MinGW (formely a MinGW.org Project, later the MinGW.OSDN Project
+    at <https://osdn.net/projects/mingw/>) has GCC 9.2.0 as the
+    most recent GCC package (released 2021-02-02). The project might
+    still be alive but majority of people have switched to MinGW-w64.
+    Thus it seems clearer to refer to MinGW-w64 in our API headers too.
+    Building with MinGW is likely to still work but I haven't tested it
+    in the recent years.
+
+ src/liblzma/api/lzma.h         | 4 ++--
+ src/liblzma/api/lzma/version.h | 2 +-
+ 2 files changed, 3 insertions(+), 3 deletions(-)
+
+commit 36fabdbe67c8a8fbdc3ac695a91fc443a1328cc4
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-27 00:58:17 +0300
+
+    CMake: Use -D_FILE_OFFSET_BITS=64 if (and only if) needed.
+    
+    A CMake option LARGE_FILE_SUPPORT is created if and only if
+    -D_FILE_OFFSET_BITS=64 affects sizeof(off_t).
+    
+    This is needed on many 32-bit platforms and even with 64-bit builds
+    with MinGW-w64 to get support for files larger than 2 GiB.
+
+ CMakeLists.txt                        |  7 ++++-
+ cmake/tuklib_large_file_support.cmake | 52 +++++++++++++++++++++++++++++++++++
+ 2 files changed, 58 insertions(+), 1 deletion(-)
+
+commit 989c8c354cbd2d20fbae4a432a3e31f5bc1cb9bf
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-30 02:14:25 +0300
+
+    CMake: Generate and install liblzma.pc if not using MSVC.
+    
+    Autotools based build uses -pthread and thus adds it to Libs.private
+    in liblzma.pc. CMake doesn't use -pthread at all if pthread functions
+    are available in libc so Libs.private doesn't get -pthread either.
+
+ CMakeLists.txt | 21 +++++++++++++++++++++
+ 1 file changed, 21 insertions(+)
+
+commit 983f3b458dc79c5976a4237fdfe4f8079f8d8830
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-30 01:13:13 +0300
+
+    CMake: Rearrange the PACKAGE_ variables.
+    
+    The windres workaround now replaces spaces with \x20 so
+    the package name isn't repeated.
+    
+    These changes will help with creation of liblzma.pc.
+
+ CMakeLists.txt | 26 +++++++++++++++-----------
+ 1 file changed, 15 insertions(+), 11 deletions(-)
+
+commit 4083c8e9501a48934a5fb563d2c3ce2ae143cd27
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-29 20:46:11 +0300
+
+    liblzma: Add Cflags.private to liblzma.pc.in for MSYS2.
+    
+    It properly adds -DLZMA_API_STATIC when compiling code that
+    will be linked against static liblzma. Having it there on
+    systems other than Windows does no harm.
+    
+    See: https://www.msys2.org/docs/pkgconfig/
+
+ src/liblzma/liblzma.pc.in | 1 +
+ 1 file changed, 1 insertion(+)
+
+commit 661549ecb7a9b136d72a01c137d9776c75d52d51
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-27 22:46:20 +0300
+
+    CMake: Create liblzma.def when building liblzma.dll with MinGW-w64.
+
+ CMakeLists.txt              | 20 ++++++++++++++++++++
+ cmake/remove-ordinals.cmake | 26 ++++++++++++++++++++++++++
+ 2 files changed, 46 insertions(+)
+
+commit 0e546eb4da05c52b7d257e5bd85e15c51c4d86a3
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-10-26 21:44:42 +0300
+
+    CMake: Change one CMAKE_CURRENT_SOURCE_DIR to CMAKE_CURRENT_LIST_DIR.
+    
+    In this case they have identical values.
+
+ CMakeLists.txt | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+commit da4d04e4d6e199d28b58bd2e0df4e120c52dd5d7
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-10-01 19:10:57 +0300
+
+    CMake/Windows: Fix the import library filename.
+    
+    Both PREFIX and IMPORT_PERFIX have to be set to "" to get
+    liblzma.dll and liblzma.dll.a.
+
+ CMakeLists.txt | 1 +
+ 1 file changed, 1 insertion(+)
+
+commit 007558a358c48a0175cc8d47d11798d7967282ab
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-10-11 19:47:44 +0300
+
+    CMake: Don't shadow the cache entry ENABLE_THREADS with a normal variable.
+    
+    Using set(ENABLE_THREADS "posix") is confusing because it sets
+    a new normal variable and leaves the cache entry with the same
+    name unchanged. The intent wasn't to change the cache entry so
+    this switches to a different variable name.
+
+ CMakeLists.txt | 10 +++++++---
+ 1 file changed, 7 insertions(+), 3 deletions(-)
+
+commit 7d01de67ee3dd76cfc12c23220e2e4cdc59708f1
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-10-09 21:12:31 +0300
+
+    CMake: Edit threading related messages.
+    
+    It's mostly to change from "thread method" to "threading method".
+
+ CMakeLists.txt | 19 ++++++++++---------
+ 1 file changed, 10 insertions(+), 9 deletions(-)
+
+commit f8edcf3da689aad4b21e139197725450f2c456a0
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-10-09 20:59:24 +0300
+
+    CMake: Use FATAL_ERROR if user-supplied options aren't understood.
+    
+    This way typos are caught quickly and compounding error messages
+    are avoided (a single typo could cause more than one error).
+    
+    This keeps using SEND_ERROR when the system is lacking a feature
+    (like threading library or sandboxing method). This way the whole
+    configuration log will be generated in case someone wishes to
+    report a problem upstream.
+
+ CMakeLists.txt | 28 ++++++++++++++--------------
+ 1 file changed, 14 insertions(+), 14 deletions(-)
+
+commit 1695021e4a233a9388ddd428654c1447f0ea3bfb
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-10-19 16:09:01 +0800
+
+    CMake: Add ALLOW_CLMUL_CRC option to enable/disable CLMUL.
+    
+    The option is enabled by default, but will only be visible to a user
+    listing cache variables or using a CMake GUI application if the
+    immintrin.h header file is found.
+    
+    This mirrors our Autotools build --disable-clmul-crc functionality.
+
+ CMakeLists.txt | 40 +++++++++++++++++++++++-----------------
+ 1 file changed, 23 insertions(+), 17 deletions(-)
+
+commit 5056bc51071d1a07097c5667a0d5bd85242e31b9
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-10-14 17:56:59 +0300
+
+    tuklib_integer: Revise unaligned reads and writes on strict-align archs.
+    
+    In XZ Utils context this doesn't matter much because
+    unaligned reads and writes aren't used in hot code
+    when TUKLIB_FAST_UNALIGNED_ACCESS isn't #defined.
+
+ src/common/tuklib_integer.h | 256 ++++++++++++++++++++++++++++++++------------
+ 1 file changed, 189 insertions(+), 67 deletions(-)
+
+commit 9e14743ee5ba79181076bc33952245d5b18fbc58
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-23 02:21:49 +0300
+
+    tuklib_integer: Add missing write64be and write64le fallback functions.
+
+ src/common/tuklib_integer.h | 34 ++++++++++++++++++++++++++++++++++
+ 1 file changed, 34 insertions(+)
+
+commit 4cc91ceb3992ef4f51302b56178c3b2c2aeaaaad
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-10-12 20:12:18 +0800
+
+    Build: Update visibility.m4 from Gnulib.
+    
+    Updating from version 6 -> 8 from upstream. Declarations for variables
+    and function bodies were added to avoid unnecessary failures with
+    -Werror.
+
+ m4/visibility.m4 | 9 +++++++--
+ 1 file changed, 7 insertions(+), 2 deletions(-)
+
+commit 1824a6007cb1c8d5d7abcc7bf649148bc06fa72c
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-10-06 19:36:35 +0300
+
+    Update THANKS.
+
+ THANKS | 1 +
+ 1 file changed, 1 insertion(+)
+
+commit 8fdc71a27d07b10a3da52432432e080b6d577642
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-09-29 20:14:39 +0800
+
+    CMake: Rename xz and man page symlink custom targets.
+    
+    The Ninja Generator for CMake cannot have a custom target and its
+    BYPRODUCTS have the same name. This has prevented Ninja builds on
+    Unix-like systems since the xz symlinks were introduced in
+    80a1a8bb838842a2be343bd88ad1462c21c5e2c9.
+
+ CMakeLists.txt | 6 +++---
+ 1 file changed, 3 insertions(+), 3 deletions(-)
+
+commit 38171492ded6426ddf53d0c200fa8c93fcd02a60
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-27 19:54:35 +0300
+
+    CMake: Fix Windows build with Clang/LLVM 17.
+    
+    llvm-windres 17.0.0 has more accurate emulation of GNU windres, so
+    the hack for GNU windres must now be used with llvm-windres too.
+    
+    LLVM 16.0.6 has the old behavior and there likely won't be more
+    16.x releases. So we can simply check for >= 17.0.0.
+    
+    The workaround must not be used with Clang that is acting in
+    MSVC mode. This checks for the known environments that need
+    the workaround instead of using "NOT MSVC".
+    
+    See also:
+    https://github.com/llvm/llvm-project/commit/2bcc0fdc58a220cb9921b47ec8a32c85f2511a47
+
+ CMakeLists.txt | 26 ++++++++++++++------------
+ 1 file changed, 14 insertions(+), 12 deletions(-)
+
+commit 1bce6fe48334b5df33d0487a9cbe41950122230e
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-09-27 00:02:11 +0800
+
+    liblzma: Avoid compiler warning without creating extra symbol.
+    
+    When the generic fast crc64 method is used, then we omit
+    lzma_crc64_table[][].
+    
+    The C standards don't allow an empty translation unit which can be
+    avoided by declaring something, without exporting any symbols.
+
+ src/liblzma/check/crc64_table.c | 6 ++----
+ 1 file changed, 2 insertions(+), 4 deletions(-)
+
+commit dce95a593e6cd9779110aa1e314abd8b35c75f6b
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-26 17:24:15 +0300
+
+    Build: Update the comment about -Werror usage in checks.
+
+ configure.ac | 10 ++++++++--
+ 1 file changed, 8 insertions(+), 2 deletions(-)
+
+commit f3c32762ae309afa2fe330e7fb397acfdedc4d37
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-26 13:51:31 +0300
+
+    Build: Fix underquoted AC_LANG_SOURCE.
+    
+    It made no practical difference in this case.
+
+ configure.ac | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+commit 7dd57f2f2c8fde93fa42b4dbf6d9860717723b41
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-26 13:14:37 +0300
+
+    Build: Silence Autoconf warning.
+    
+    There was a use of AC_COMPILE_IFELSE that didn't use
+    AC_LANG_SOURCE and Autoconf warned about this. The omission
+    had been intentional but it turned out that this didn't do
+    what I thought it would.
+    
+    Autoconf 2.71 manual gives an impression that AC_LANG_SOURCE
+    inserts all #defines that have been made with AC_DEFINE so
+    far (confdefs.h). The idea was that omitting AC_LANG_SOURCE
+    would mean that only the exact code included in the
+    AC_COMPILE_IFELSE call would be compiled.
+    
+    With C programs this is not true: the #defines get added without
+    AC_LANG_SOURCE too. There seems to be no neat way to avoid this.
+    Thus, with the C language at least, adding AC_LANG_SOURCE makes
+    no other difference than silencing a warning from Autoconf. The
+    generated "configure" remains identical. (Docs of AC_LANG_CONFTEST
+    say that the #defines have been inserted since Autoconf 2.63b and
+    that AC_COMPILE_IFELSE uses AC_LANG_CONFTEST. So the behavior is
+    documented if one also reads the docs of macros that one isn't
+    calling directly.)
+    
+    Any extra code, including #defines, can cause problems for
+    these two tests because these tests must use -Werror.
+    CC=clang CFLAGS=-Weverything is the most extreme example.
+    It enables -Wreserved-macro-identifier which warns about
+    It's possible to write a test file that passes -Weverything but
+    it becomes impossible when Autoconf inserts confdefs.h.
+    
+    So this commit adds AC_LANG_SOURCE to silence Autoconf warnings.
+    A different solution is needed for -Werror tests.
+
+ configure.ac | 5 ++---
+ 1 file changed, 2 insertions(+), 3 deletions(-)
+
+commit edec253e418562f3164a01ecc8805295fa022efa
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-09-26 00:47:26 +0800
+
+    Build: Remove Gnulib dependency from tests.
+    
+    The tests do not use any Gnulib replacements so they do not need to link
+    libgnu.a or have /lib in the include path.
+
+ tests/Makefile.am | 7 +------
+ 1 file changed, 1 insertion(+), 6 deletions(-)
+
+commit 46cb133ce7360496eecca1255b364c05f0205855
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-09-26 00:43:43 +0800
+
+    CMake: Remove /lib from tests include path.
+    
+    The tests never included anything from /lib, so this was not needed.
+
+ CMakeLists.txt | 1 -
+ 1 file changed, 1 deletion(-)
+
+commit 4ae13cfe0dedb8ddc3cf9ded8cd1ac09361b3bd1
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-24 16:32:32 +0300
+
+    sysdefs.h: Update the comment about __USE_MINGW_ANSI_STDIO.
+
+ src/common/sysdefs.h | 10 +++++++++-
+ 1 file changed, 9 insertions(+), 1 deletion(-)
+
+commit 660c8c29e57d30dbd5009ef1f0ec1bbe195ccef6
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-22 02:33:29 +0300
+
+    xz: Windows: Don't (de)compress to special files like "con" or "nul".
+    
+    Before this commit, the following writes "foo" to the
+    console and deletes the input file:
+    
+        echo foo | xz > con_xz
+        xz --suffix=_xz --decompress con_xz
+    
+    It cannot happen without --suffix because names like con.xz
+    are also special and so attempting to decompress con.xz
+    (or compress con to con.xz) will already fail when opening
+    the input file.
+    
+    Similar thing is possible when compressing. The following
+    writes to "nul" and the input file "n" is deleted.
+    
+        echo foo | xz > n
+        xz --suffix=ul n
+    
+    Now xz checks if the destination is a special file before
+    continuing. DOS/DJGPP version had a check for this but
+    Windows (and OS/2) didn't.
+
+ src/xz/file_io.c | 35 ++++++++++++++++++++++++++++-------
+ 1 file changed, 28 insertions(+), 7 deletions(-)
+
+commit b7ce6e80786fc0c08ed129e8ee262ea96a5473a1
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-21 20:42:52 +0300
+
+    CMake: Wrap two overlong lines that are possible to wrap.
+
+ CMakeLists.txt | 6 ++++--
+ 1 file changed, 4 insertions(+), 2 deletions(-)
+
+commit 1595f454d5c8257c668cccd6a86dd68175d5c430
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-21 20:36:31 +0300
+
+    CMake: Add a comment about threads on Cygwin.
+
+ CMakeLists.txt | 1 +
+ 1 file changed, 1 insertion(+)
+
+commit 5be6275f19784cdd5a954f0188045c8ff4934d54
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-12 21:12:34 +0300
+
+    CMake: Bump maximum policy version to 3.27.
+    
+    There are several new policies. CMP0149 may affect the Windows SDK
+    version that CMake will choose by default. The new behavior is more
+    predictable, always choosing the latest SDK version by default.
+    
+    The other new policies shouldn't affect this package.
+
+ CMakeLists.txt | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+commit e515643d7524851d1eb7dab73453e26d8521324c
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-08 19:08:57 +0300
+
+    Doxygen: Add more C macro names to PREDEFINED.
+
+ doxygen/Doxyfile | 7 +++++--
+ 1 file changed, 5 insertions(+), 2 deletions(-)
+
+commit e3478ae4f36cd06522a2fef023860893f068434d
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-11 18:47:26 +0300
+
+    liblzma: Move a few __attribute__ uses in function declarations.
+    
+    The API headers have many attributes but these were left
+    as is for now.
+
+ src/liblzma/common/common.c    | 6 ++++--
+ src/liblzma/common/common.h    | 8 ++++----
+ src/liblzma/common/memcmplen.h | 3 ++-
+ 3 files changed, 10 insertions(+), 7 deletions(-)
+
+commit b71b8922ef3971e5ccffd1e213888d44abe21d11
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-11 19:03:35 +0300
+
+    xz, xzdec, lzmainfo: Use tuklib_attr_noreturn.
+    
+    For compatibility with C23's [[noreturn]], tuklib_attr_noreturn
+    must be at the beginning of declaration (before "extern" or
+    "static", and even before any GNU C's __attribute__).
+    
+    This commit also moves all other function attributes to
+    the beginning of function declarations. "extern" is kept
+    at the beginning of a line so the attributes are listed on
+    separate lines before "extern" or "static".
+
+ src/lzmainfo/lzmainfo.c |  6 ++++--
+ src/xz/coder.c          |  3 ++-
+ src/xz/hardware.h       |  3 ++-
+ src/xz/message.h        | 30 +++++++++++++++++-------------
+ src/xz/options.c        |  3 ++-
+ src/xz/util.h           |  8 ++++----
+ src/xzdec/xzdec.c       |  9 ++++++---
+ 7 files changed, 37 insertions(+), 25 deletions(-)
+
+commit 359e5c6cb128dab64ea6070d21d1c240f96cea6b
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-11 18:53:31 +0300
+
+    Remove incorrect uses of __attribute__((__malloc__)).
+    
+    xrealloc() is obviously incorrect, modern GCC docs even
+    mention realloc() as an example where this attribute
+    cannot be used.
+    
+    liblzma's lzma_alloc() and lzma_alloc_zero() would be
+    correct uses most of the time but custom allocators
+    may use a memory pool or otherwise hold the pointer
+    so aliasing issues could happen in theory.
+    
+    The xstrdup() case likely was correct but I removed it anyway.
+    Now there are no __malloc__ attributes left in the code.
+    The allocations aren't in hot paths so this should make
+    no practical difference.
+
+ src/liblzma/common/common.c | 4 ++--
+ src/liblzma/common/common.h | 4 ++--
+ src/xz/util.h               | 4 ++--
+ 3 files changed, 6 insertions(+), 6 deletions(-)
+
+commit 589b4cba22fccb1dbc919df5d134aefb2b5a6b01
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-19 14:03:45 +0300
+
+    Update THANKS.
+
+ THANKS | 1 +
+ 1 file changed, 1 insertion(+)
+
+commit 43728ed2267e921fbdfa699ee1e91b105ab0e98b
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-14 16:35:46 +0300
+
+    Update THANKS.
+
+ THANKS | 1 +
+ 1 file changed, 1 insertion(+)
+
+commit caf00e0988ba47842cfd93dfbb17f7d30120d6e7
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-09-14 16:34:07 +0300
+
+    liblzma: Mark crc64_clmul() with __attribute__((__no_sanitize_address__)).
+    
+    Thanks to Agostino Sarubbo.
+    Fixes: https://github.com/tukaani-project/xz/issues/62
+
+ src/liblzma/check/crc64_fast.c | 8 ++++++++
+ 1 file changed, 8 insertions(+)
+
+commit a70e96d2da761b8b3a77bf14e08002d871e5950b
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-09-12 22:36:12 +0800
+
+    CMake: Fix time.h checks not running on second CMake run.
+    
+    If CMake was configured more than once, HAVE_CLOCK_GETTIME and
+    HAVE_CLOCK_MONOTONIC would not be set as compile definitions. The check
+    for librt being needed to provide HAVE_CLOCK_GETTIME was also
+    simplified.
+
+ CMakeLists.txt | 18 ++++++++++--------
+ 1 file changed, 10 insertions(+), 8 deletions(-)
+
+commit d5275d83bd2a9701c5feb8666785007c074b1359
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-09-12 22:34:06 +0800
+
+    CMake: Fix unconditionally defining HAVE_CLOCK_MONOTONIC.
+    
+    If HAVE_CLOCK_GETTIME was defined, then HAVE_CLOCK_MONOTONIC was always
+    added as a compile definition even if the check for it failed.
+
+ CMakeLists.txt | 8 +++-----
+ 1 file changed, 3 insertions(+), 5 deletions(-)
+
+commit 1f6e7c68fbdeeaa9482fc77de090be63d90912fd
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-08-31 19:50:05 +0300
+
+    xz: Refactor thousand separator detection and disable it on MSVC.
+    
+    Now the two variations of the format strings are created with
+    a macro, and the whole detection code can be easily disabled
+    on platforms where thousand separator formatting is known to
+    not work (MSVC has no support, and on DJGPP 2.05 it can have
+    problems in some cases).
+
+ src/xz/util.c | 89 ++++++++++++++++++++++++++++++-----------------------------
+ 1 file changed, 45 insertions(+), 44 deletions(-)
+
+commit ef71f83973a20cc28a3221f85681922026ea33f5
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-08-31 18:14:43 +0300
+
+    xz: Fix a too relaxed assertion and remove uses of SSIZE_MAX.
+    
+    SSIZE_MAX isn't readily available on MSVC. Removing it means
+    that there is one thing less to worry when porting to MSVC.
+
+ src/xz/file_io.c | 5 ++---
+ src/xz/file_io.h | 4 ++--
+ 2 files changed, 4 insertions(+), 5 deletions(-)
+
+commit cf8ba7c3a89e37736b926dfbe85dffeff725db47
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-08-28 23:14:45 +0800
+
+    Tests: Improve invalid unpadded size check in test_lzma_index_append().
+    
+    This check was extended to test the code added to fix a failing assert
+    in 68bda971bb8b666a009331455fcedb4e18d837a4.
+
+ tests/test_index.c | 26 +++++++++++++++++++++++---
+ 1 file changed, 23 insertions(+), 3 deletions(-)
+
+commit 4a4180ce74788e97e90b9aab579bfd7c6dce3f59
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-08-28 21:54:41 +0800
+
+    Tests: Improve comments in test_index.c.
+
+ tests/test_index.c | 6 +++---
+ 1 file changed, 3 insertions(+), 3 deletions(-)
+
+commit 4b23b84b89e39a5117e16f66c3b01db4f08ed3e7
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-08-28 21:52:54 +0800
+
+    Update THANKS.
+
+ THANKS | 1 +
+ 1 file changed, 1 insertion(+)
+
+commit 773f1e8622cb1465df528cb16a749517650acd93
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-08-28 21:50:16 +0800
+
+    liblzma: Update assert in vli_ceil4().
+    
+    The argument to vli_ceil4() should always guarantee the return value
+    is also a valid lzma_vli. Thus the highest three valid lzma_vli values
+    are invalid arguments. All uses of the function ensure this so the
+    assert is updated to match this.
+
+ src/liblzma/common/index.h | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+commit 68bda971bb8b666a009331455fcedb4e18d837a4
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-08-28 21:31:25 +0800
+
+    liblzma: Add overflow check for Unpadded size in lzma_index_append().
+    
+    This was not a security bug since there was no path to overflow
+    UINT64_MAX in lzma_index_append() or when it calls index_file_size().
+    The bug was discovered by a failing assert() in vli_ceil4() when called
+    from index_file_size() when unpadded_sum (the sum of the compressed size
+    of current Stream and the unpadded_size parameter) exceeds LZMA_VLI_MAX.
+    
+    Previously, the unpadded_size parameter was checked to be not greater
+    than UNPADDED_SIZE_MAX, but no check was done once compressed_base was
+    added.
+    
+    This could not have caused an integer overflow in index_file_size() when
+    called by lzma_index_append(). The calculation for file_size breaks down
+    into the sum of:
+    
+    - Compressed base from all previous Streams
+    - 2 * LZMA_STREAM_HEADER_SIZE (size of the current Streams header and
+      footer)
+    - stream_padding (can be set by lzma_index_stream_padding())
+    - Compressed base from the current Stream
+    - Unpadded size (parameter to lzma_index_append())
+    
+    The sum of everything except for Unpadded size must be less than
+    LZMA_VLI_MAX. This is guarenteed by overflow checks in the functions
+    that can set these values including lzma_index_stream_padding(),
+    lzma_index_append(), and lzma_index_cat(). The maximum value for
+    Unpadded size is enforced by lzma_index_append() to be less than or
+    equal UNPADDED_SIZE_MAX. Thus, the sum cannot exceed UINT64_MAX since
+    LZMA_VLI_MAX is half of UINT64_MAX.
+    
+    Thanks to Joona Kannisto for reporting this.
+
+ src/liblzma/common/index.c | 6 ++++++
+ 1 file changed, 6 insertions(+)
+
+commit b41bb79c602481d7ea93d65f5b3e3f08dc54233b
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-08-28 22:18:29 +0800
+
+    Translations: Update the Esperanto translation.
+
+ po/eo.po | 47 +++++++++++++++++++++++++++++------------------
+ 1 file changed, 29 insertions(+), 18 deletions(-)
+
+commit 6614e6d4bf8e2b5af6eb73930148e0ffc8d2265a
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-08-09 20:55:36 +0800
+
+    Docs: Update INSTALL for --enable-threads method win95.
+    
+    The Autotools build allows win95 threads and --enable-small together now
+    if the compiler supports __attribute__((__constructor__)).
+
+ INSTALL | 6 ++++--
+ 1 file changed, 4 insertions(+), 2 deletions(-)
+
+commit bfb623ad96fa6f1dbc0c560403c4296e3c8e26c9
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-08-09 20:54:15 +0800
+
+    CMake: Conditionally allow win95 threads and --enable-small.
+
+ CMakeLists.txt | 27 +++++++++++++++++++--------
+ 1 file changed, 19 insertions(+), 8 deletions(-)
+
+commit e919ebb29ac9f5270cd7176a39d0d3b4cea875a9
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-08-09 20:35:16 +0800
+
+    Build: Conditionally allow win95 threads and --enable-small.
+    
+    When the compiler supports __attribute__((__constructor__))
+    mythread_once() is never used, even with --enable-small. A configuration
+    with win95 threads and --enable-small will compile and be thread safe so
+    it can be allowed.
+    
+    This isn't a very common configuration since MSVC does not support
+    __attribute__((__constructor__)), but MINGW32 and CLANG32 environments
+    for MSYS2 can use win95 threads and have
+    __attribute__((__constructor__)) support.
+
+ configure.ac | 21 +++++++++++++--------
+ 1 file changed, 13 insertions(+), 8 deletions(-)
+
+commit c0c0cd4a483a672b66a13761583bc4f84d86d501
+Author: Jamaika1 <lukaszcz18@wp.pl>
+Date:   2023-08-08 14:07:59 +0200
+
+    mythread.h: Fix typo error in Vista threads mythread_once().
+    
+    The "once_" variable was accidentally referred to as just "once". This
+    prevented building with Vista threads when
+    HAVE_FUNC_ATTRIBUTE_CONSTRUCTOR was not defined.
+
+ src/common/mythread.h | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+commit d93fbefcc48a8737fdf5678ce66d1c1d605752a0
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-08-03 20:10:21 +0800
+
+    Tests: Style fixes to test_lzip_decoder.c.
+
+ tests/test_lzip_decoder.c | 36 ++++++++++++++++++++++++------------
+ 1 file changed, 24 insertions(+), 12 deletions(-)
+
+commit 65981d8e45741fd1502e007609469e1d60312e69
+Author: Jia Tan <jiat0218@gmail.com>
+Date:   2023-08-03 15:56:20 +0800
+
+    Translations: Update the Chinese (simplified) translation.
+
+ po/zh_CN.po | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+commit a108ed589171d683c34238a87e358b87f69e39a0
+Author: Lasse Collin <lasse.collin@tukaani.org>
+Date:   2023-08-02 17:15:12 +0300
+
+    xz: Omit an empty paragraph on the man page.
+
+ src/xz/xz.1 | 1 -
+ 1 file changed, 1 deletion(-)
+
+commit 03c51c5c08fe3579d8bbc8eea8251a1128001330
 Author: Jia Tan <jiat0218@gmail.com>
 Date:   2023-08-02 20:32:20 +0800
 
@@ -8,7 +874,7 @@ Date:   2023-08-02 20:32:20 +0800
  src/liblzma/api/lzma/version.h | 2 +-
  2 files changed, 2 insertions(+), 2 deletions(-)
 
-commit 7d266d25ae323a2dc5f2e254c991ef84b997adad
+commit d7fa3f1b58a77f79b1a4e12452418b5555632e18
 Author: Jia Tan <jiat0218@gmail.com>
 Date:   2023-08-02 20:30:07 +0800
 
diff --git a/THANKS b/THANKS
index cf7c59c844c5..0206af938f1a 100644
--- a/THANKS
+++ b/THANKS
@@ -19,6 +19,7 @@ has been important. :-) In alphabetical order:
   - Jakub Bogusz
   - Adam Borowski
   - Maarten Bosmans
+  - Lukas Braune
   - Benjamin Buch
   - Trent W. Buck
   - Kevin R. Bulgrien
@@ -64,6 +65,7 @@ has been important. :-) In alphabetical order:
   - Jouk Jansen
   - Jun I Jin
   - Kiyoshi Kanazawa
+  - Joona Kannisto
   - Per Øyvind Karlsen
   - Iouri Kharon
   - Thomas Klausner
@@ -127,6 +129,7 @@ has been important. :-) In alphabetical order:
   - Torsten Rupp
   - Stephen Sachs
   - Jukka Salmi
+  - Agostino Sarubbo
   - Alexandre Sauvé
   - Benno Schulenberg
   - Andreas Schwab
@@ -138,6 +141,7 @@ has been important. :-) In alphabetical order:
   - Brad Smith
   - Bruce Stark
   - Pippijn van Steenhoven
+  - Martin Storsjö
   - Jonathan Stott
   - Dan Stromberg
   - Jia Tan
diff --git a/doxygen/Doxyfile b/doxygen/Doxyfile
index 20afb528e4f4..14350cf44f64 100644
--- a/doxygen/Doxyfile
+++ b/doxygen/Doxyfile
@@ -2277,8 +2277,11 @@ INCLUDE_FILE_PATTERNS  =
 # This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
 
 PREDEFINED             = LZMA_API(type)=type \
-                         LZMA_API_IMPORT \
-                         LZMA_API_CALL=
+                         LZMA_API_IMPORT= \
+                         LZMA_API_CALL= \
+                         tuklib_attr_noreturn= \
+                         lzma_attribute(attr)= \
+                         lzma_attr_alloc_size(size)=
 
 # If the MACRO_EXPANSION and EXPAND_ONLY_PREDEF tags are set to YES then this
 # tag can be used to specify a list of macro names that should be expanded. The
diff --git a/src/common/mythread.h b/src/common/mythread.h
index 1cce50e9fd85..4495e017b290 100644
--- a/src/common/mythread.h
+++ b/src/common/mythread.h
@@ -378,7 +378,7 @@ typedef struct {
 			abort(); \
 		if (pending_) { \
 			func(); \
-			if (!InitOnceComplete(&once, 0, NULL)) \
+			if (!InitOnceComplete(&once_, 0, NULL)) \
 				abort(); \
 		} \
 	} while (0)
diff --git a/src/common/sysdefs.h b/src/common/sysdefs.h
index 97be4ee380d0..f04e45dd2147 100644
--- a/src/common/sysdefs.h
+++ b/src/common/sysdefs.h
@@ -24,7 +24,15 @@
 #	include <config.h>
 #endif
 
-// Get standard-compliant stdio functions under MinGW and MinGW-w64.
+// This #define ensures that C99 and POSIX compliant stdio functions are
+// available with MinGW-w64 (both 32-bit and 64-bit). Modern MinGW-w64 adds
+// this automatically, for example, when the compiler is in C99 (or later)
+// mode when building against msvcrt.dll. It still doesn't hurt to be explicit
+// that we always want this and #define this unconditionally.
+//
+// With Universal CRT (UCRT) this is less important because UCRT contains
+// C99-compatible stdio functions. It's still nice to #define this as UCRT
+// doesn't support the POSIX thousand separator flag in printf (like "%'u").
 #ifdef __MINGW32__
 #	define __USE_MINGW_ANSI_STDIO 1
 #endif
diff --git a/src/common/tuklib_integer.h b/src/common/tuklib_integer.h
index 24d9efb1165a..e22aa8ad83b2 100644
--- a/src/common/tuklib_integer.h
+++ b/src/common/tuklib_integer.h
@@ -195,6 +195,9 @@
 // Unaligned reads and writes //
 ////////////////////////////////
 
+// No-strict-align archs like x86-64
+// ---------------------------------
+//
 // The traditional way of casting e.g. *(const uint16_t *)uint8_pointer
 // is bad even if the uint8_pointer is properly aligned because this kind
 // of casts break strict aliasing rules and result in undefined behavior.
@@ -209,12 +212,115 @@
 // build time. A third method, casting to a packed struct, would also be
 // an option but isn't provided to keep things simpler (it's already a mess).
 // Hopefully this is flexible enough in practice.
+//
+// Some compilers on x86-64 like Clang >= 10 and GCC >= 5.1 detect that
+//
+//     buf[0] | (buf[1] << 8)
+//
+// reads a 16-bit value and can emit a single 16-bit load and produce
+// identical code than with the memcpy() method. In other cases Clang and GCC
+// produce either the same or better code with memcpy(). For example, Clang 9
+// on x86-64 can detect 32-bit load but not 16-bit load.
+//
+// MSVC uses unaligned access with the memcpy() method but emits byte-by-byte
+// code for "buf[0] | (buf[1] << 8)".
+//
+// Conclusion: The memcpy() method is the best choice when unaligned access
+// is supported.
+//
+// Strict-align archs like SPARC
+// -----------------------------
+//
+// GCC versions from around 4.x to to at least 13.2.0 produce worse code
+// from the memcpy() method than from simple byte-by-byte shift-or code
+// when reading a 32-bit integer:
+//
+//     (1) It may be constructed on stack using using four 8-bit loads,
+//         four 8-bit stores to stack, and finally one 32-bit load from stack.
+//
+//     (2) Especially with -Os, an actual memcpy() call may be emitted.
+//
+// This is true on at least on ARM, ARM64, SPARC, SPARC64, MIPS64EL, and
+// RISC-V. Of these, ARM, ARM64, and RISC-V support unaligned access in
+// some processors but not all so this is relevant only in the case when
+// GCC assumes that unaligned is not supported or -mstrict-align or
+// -mno-unaligned-access is used.
+//
+// For Clang it makes little difference. ARM64 with -O2 -mstrict-align
+// was one the very few with a minor difference: the memcpy() version
+// was one instruction longer.
+//
+// Conclusion: At least in case of GCC and Clang, byte-by-byte code is
+// the best choise for strict-align archs to do unaligned access.
+//
+// See also: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111502
+//
+// Thanks to <https://godbolt.org/> it was easy to test different compilers.
+// The following is for little endian targets:
+/*
+#include <stdint.h>
+#include <string.h>
+
+uint32_t bytes16(const uint8_t *b)
+{
+    return (uint32_t)b[0]
+        | ((uint32_t)b[1] << 8);
+}
+
+uint32_t copy16(const uint8_t *b)
+{
+    uint16_t v;
+    memcpy(&v, b, sizeof(v));
+    return v;
+}
+
+uint32_t bytes32(const uint8_t *b)
+{
+    return (uint32_t)b[0]
+        | ((uint32_t)b[1] << 8)
+        | ((uint32_t)b[2] << 16)
+        | ((uint32_t)b[3] << 24);
+}
+
+uint32_t copy32(const uint8_t *b)
+{
+    uint32_t v;
+    memcpy(&v, b, sizeof(v));
+    return v;
+}
+
+void wbytes16(uint8_t *b, uint16_t v)
+{
+    b[0] = (uint8_t)v;
+    b[1] = (uint8_t)(v >> 8);
+}
+
+void wcopy16(uint8_t *b, uint16_t v)
+{
+    memcpy(b, &v, sizeof(v));
+}
+
+void wbytes32(uint8_t *b, uint32_t v)
+{
+    b[0] = (uint8_t)v;
+    b[1] = (uint8_t)(v >> 8);
+    b[2] = (uint8_t)(v >> 16);
+    b[3] = (uint8_t)(v >> 24);
+}
+
+void wcopy32(uint8_t *b, uint32_t v)
+{
+    memcpy(b, &v, sizeof(v));
+}
+*/
+
+
+#ifdef TUKLIB_FAST_UNALIGNED_ACCESS
 
 static inline uint16_t
 read16ne(const uint8_t *buf)
 {
-#if defined(TUKLIB_FAST_UNALIGNED_ACCESS) \
-		&& defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING)
+#ifdef TUKLIB_USE_UNSAFE_TYPE_PUNNING
 	return *(const uint16_t *)buf;
 #else
 	uint16_t num;
@@ -227,8 +333,7 @@ read16ne(const uint8_t *buf)
 static inline uint32_t
 read32ne(const uint8_t *buf)
 {
-#if defined(TUKLIB_FAST_UNALIGNED_ACCESS) \
-		&& defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING)
+#ifdef TUKLIB_USE_UNSAFE_TYPE_PUNNING
 	return *(const uint32_t *)buf;
 #else
 	uint32_t num;
@@ -241,8 +346,7 @@ read32ne(const uint8_t *buf)
 static inline uint64_t
 read64ne(const uint8_t *buf)
 {
-#if defined(TUKLIB_FAST_UNALIGNED_ACCESS) \
-		&& defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING)
+#ifdef TUKLIB_USE_UNSAFE_TYPE_PUNNING
 	return *(const uint64_t *)buf;
 #else
 	uint64_t num;
@@ -255,8 +359,7 @@ read64ne(const uint8_t *buf)
 static inline void
 write16ne(uint8_t *buf, uint16_t num)
 {
-#if defined(TUKLIB_FAST_UNALIGNED_ACCESS) \
-		&& defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING)
+#ifdef TUKLIB_USE_UNSAFE_TYPE_PUNNING
 	*(uint16_t *)buf = num;
 #else
 	memcpy(buf, &num, sizeof(num));
@@ -268,8 +371,7 @@ write16ne(uint8_t *buf, uint16_t num)
 static inline void
 write32ne(uint8_t *buf, uint32_t num)
 {
-#if defined(TUKLIB_FAST_UNALIGNED_ACCESS) \
-		&& defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING)
+#ifdef TUKLIB_USE_UNSAFE_TYPE_PUNNING
 	*(uint32_t *)buf = num;
 #else
 	memcpy(buf, &num, sizeof(num));
@@ -281,8 +383,7 @@ write32ne(uint8_t *buf, uint32_t num)
 static inline void
 write64ne(uint8_t *buf, uint64_t num)
 {
-#if defined(TUKLIB_FAST_UNALIGNED_ACCESS) \
-		&& defined(TUKLIB_USE_UNSAFE_TYPE_PUNNING)
+#ifdef TUKLIB_USE_UNSAFE_TYPE_PUNNING
 	*(uint64_t *)buf = num;
 #else
 	memcpy(buf, &num, sizeof(num));
@@ -294,68 +395,122 @@ write64ne(uint8_t *buf, uint64_t num)
 static inline uint16_t
 read16be(const uint8_t *buf)
 {
-#if defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
 	uint16_t num = read16ne(buf);
 	return conv16be(num);
-#else
-	uint16_t num = ((uint16_t)buf[0] << 8) | (uint16_t)buf[1];
-	return num;
-#endif
 }
 
 
 static inline uint16_t
 read16le(const uint8_t *buf)
 {
-#if !defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
 	uint16_t num = read16ne(buf);
 	return conv16le(num);
-#else
-	uint16_t num = ((uint16_t)buf[0]) | ((uint16_t)buf[1] << 8);
-	return num;
-#endif
 }
 
 
 static inline uint32_t
 read32be(const uint8_t *buf)
 {
-#if defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
 	uint32_t num = read32ne(buf);
 	return conv32be(num);
+}
+
+
+static inline uint32_t
+read32le(const uint8_t *buf)
+{
+	uint32_t num = read32ne(buf);
+	return conv32le(num);
+}
+
+
+static inline uint64_t
+read64be(const uint8_t *buf)
+{
+	uint64_t num = read64ne(buf);
+	return conv64be(num);
+}
+
+
+static inline uint64_t
+read64le(const uint8_t *buf)
+{
+	uint64_t num = read64ne(buf);
+	return conv64le(num);
+}
+
+
+// NOTE: Possible byte swapping must be done in a macro to allow the compiler
+// to optimize byte swapping of constants when using glibc's or *BSD's
+// byte swapping macros. The actual write is done in an inline function
+// to make type checking of the buf pointer possible.
+#define write16be(buf, num) write16ne(buf, conv16be(num))
+#define write32be(buf, num) write32ne(buf, conv32be(num))
+#define write64be(buf, num) write64ne(buf, conv64be(num))
+#define write16le(buf, num) write16ne(buf, conv16le(num))
+#define write32le(buf, num) write32ne(buf, conv32le(num))
+#define write64le(buf, num) write64ne(buf, conv64le(num))
+
 #else
+
+#ifdef WORDS_BIGENDIAN
+#	define read16ne read16be
+#	define read32ne read32be
+#	define read64ne read64be
+#	define write16ne write16be
+#	define write32ne write32be
+#	define write64ne write64be
+#else
+#	define read16ne read16le
+#	define read32ne read32le
+#	define read64ne read64le
+#	define write16ne write16le
+#	define write32ne write32le
+#	define write64ne write64le
+#endif
+
+
+static inline uint16_t
+read16be(const uint8_t *buf)
+{
+	uint16_t num = ((uint16_t)buf[0] << 8) | (uint16_t)buf[1];
+	return num;
+}
+
+
+static inline uint16_t
+read16le(const uint8_t *buf)
+{
+	uint16_t num = ((uint16_t)buf[0]) | ((uint16_t)buf[1] << 8);
+	return num;
+}
+
+
+static inline uint32_t
+read32be(const uint8_t *buf)
+{
 	uint32_t num = (uint32_t)buf[0] << 24;
 	num |= (uint32_t)buf[1] << 16;
 	num |= (uint32_t)buf[2] << 8;
 	num |= (uint32_t)buf[3];
 	return num;
-#endif
 }
 
 
 static inline uint32_t
 read32le(const uint8_t *buf)
 {
-#if !defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
-	uint32_t num = read32ne(buf);
-	return conv32le(num);
-#else
 	uint32_t num = (uint32_t)buf[0];
 	num |= (uint32_t)buf[1] << 8;
 	num |= (uint32_t)buf[2] << 16;
 	num |= (uint32_t)buf[3] << 24;
 	return num;
-#endif
 }
 
 
 static inline uint64_t
 read64be(const uint8_t *buf)
 {
-#if defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
-	uint64_t num = read64ne(buf);
-	return conv64be(num);
-#else
 	uint64_t num = (uint64_t)buf[0] << 56;
 	num |= (uint64_t)buf[1] << 48;
 	num |= (uint64_t)buf[2] << 40;
@@ -365,17 +520,12 @@ read64be(const uint8_t *buf)
 	num |= (uint64_t)buf[6] << 8;
 	num |= (uint64_t)buf[7];
 	return num;
-#endif
 }
 
 
 static inline uint64_t
 read64le(const uint8_t *buf)
 {
-#if !defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
-	uint64_t num = read64ne(buf);
-	return conv64le(num);
-#else
 	uint64_t num = (uint64_t)buf[0];
 	num |= (uint64_t)buf[1] << 8;
 	num |= (uint64_t)buf[2] << 16;
@@ -385,28 +535,9 @@ read64le(const uint8_t *buf)
 	num |= (uint64_t)buf[6] << 48;
 	num |= (uint64_t)buf[7] << 56;
 	return num;
-#endif
 }
 
 
-// NOTE: Possible byte swapping must be done in a macro to allow the compiler
-// to optimize byte swapping of constants when using glibc's or *BSD's
-// byte swapping macros. The actual write is done in an inline function
-// to make type checking of the buf pointer possible.
-#if defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
-#	define write16be(buf, num) write16ne(buf, conv16be(num))
-#	define write32be(buf, num) write32ne(buf, conv32be(num))
-#	define write64be(buf, num) write64ne(buf, conv64be(num))
-#endif
-
-#if !defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
-#	define write16le(buf, num) write16ne(buf, conv16le(num))
-#	define write32le(buf, num) write32ne(buf, conv32le(num))
-#	define write64le(buf, num) write64ne(buf, conv64le(num))
-#endif
-
-
-#ifndef write16be
 static inline void
 write16be(uint8_t *buf, uint16_t num)
 {
@@ -414,10 +545,8 @@ write16be(uint8_t *buf, uint16_t num)
 	buf[1] = (uint8_t)num;
 	return;
 }
-#endif
 
 
-#ifndef write16le
 static inline void
 write16le(uint8_t *buf, uint16_t num)
 {
@@ -425,10 +554,8 @@ write16le(uint8_t *buf, uint16_t num)
 	buf[1] = (uint8_t)(num >> 8);
 	return;
 }
-#endif
 
 
-#ifndef write32be
 static inline void
 write32be(uint8_t *buf, uint32_t num)
 {
@@ -438,10 +565,8 @@ write32be(uint8_t *buf, uint32_t num)
 	buf[3] = (uint8_t)num;
 	return;
 }
-#endif
 
 
-#ifndef write32le
 static inline void
 write32le(uint8_t *buf, uint32_t num)
 {
@@ -451,6 +576,37 @@ write32le(uint8_t *buf, uint32_t num)
 	buf[3] = (uint8_t)(num >> 24);
 	return;
 }
+
+
+static inline void
+write64be(uint8_t *buf, uint64_t num)
+{
+	buf[0] = (uint8_t)(num >> 56);
+	buf[1] = (uint8_t)(num >> 48);
+	buf[2] = (uint8_t)(num >> 40);
+	buf[3] = (uint8_t)(num >> 32);
+	buf[4] = (uint8_t)(num >> 24);
+	buf[5] = (uint8_t)(num >> 16);
+	buf[6] = (uint8_t)(num >> 8);
+	buf[7] = (uint8_t)num;
+	return;
+}
+
+
+static inline void
+write64le(uint8_t *buf, uint64_t num)
+{
+	buf[0] = (uint8_t)num;
+	buf[1] = (uint8_t)(num >> 8);
+	buf[2] = (uint8_t)(num >> 16);
+	buf[3] = (uint8_t)(num >> 24);
+	buf[4] = (uint8_t)(num >> 32);
+	buf[5] = (uint8_t)(num >> 40);
+	buf[6] = (uint8_t)(num >> 48);
+	buf[7] = (uint8_t)(num >> 56);
+	return;
+}
+
 #endif
 
 
diff --git a/src/liblzma/api/lzma.h b/src/liblzma/api/lzma.h
index f38513d35a45..de12f225859f 100644
--- a/src/liblzma/api/lzma.h
+++ b/src/liblzma/api/lzma.h
@@ -182,11 +182,11 @@
  * against static liblzma on them, don't worry about LZMA_API_STATIC. That
  * is, most developers will never need to use LZMA_API_STATIC.
  *
- * The GCC variants are a special case on Windows (Cygwin and MinGW).
+ * The GCC variants are a special case on Windows (Cygwin and MinGW-w64).
  * We rely on GCC doing the right thing with its auto-import feature,
  * and thus don't use __declspec(dllimport). This way developers don't
  * need to worry about LZMA_API_STATIC. Also the calling convention is
- * omitted on Cygwin but not on MinGW.
+ * omitted on Cygwin but not on MinGW-w64.
  */
 #ifndef LZMA_API_IMPORT
 #	if !defined(LZMA_API_STATIC) && defined(_WIN32) && !defined(__GNUC__)
diff --git a/src/liblzma/api/lzma/version.h b/src/liblzma/api/lzma/version.h
index 8739d751c704..8dac38297273 100644
--- a/src/liblzma/api/lzma/version.h
+++ b/src/liblzma/api/lzma/version.h
@@ -23,7 +23,7 @@
 #define LZMA_VERSION_MINOR 4
 
 /** \brief Patch version number of the liblzma release. */
-#define LZMA_VERSION_PATCH 4
+#define LZMA_VERSION_PATCH 5
 
 /**
  * \brief Version stability marker
@@ -106,7 +106,7 @@
 		LZMA_VERSION_COMMIT)
 
 
-/* #ifndef is needed for use with windres (MinGW or Cygwin). */
+/* #ifndef is needed for use with windres (MinGW-w64 or Cygwin). */
 #ifndef LZMA_H_INTERNAL_RC
 
 /**
diff --git a/src/liblzma/check/check.h b/src/liblzma/check/check.h
index 783627bd21b9..8ae95d59019f 100644
--- a/src/liblzma/check/check.h
+++ b/src/liblzma/check/check.h
@@ -99,10 +99,17 @@ typedef struct {
 /// lzma_crc32_table[0] is needed by LZ encoder so we need to keep
 /// the array two-dimensional.
 #ifdef HAVE_SMALL
+lzma_attr_visibility_hidden
 extern uint32_t lzma_crc32_table[1][256];
+
 extern void lzma_crc32_init(void);
+
 #else
+
+lzma_attr_visibility_hidden
 extern const uint32_t lzma_crc32_table[8][256];
+
+lzma_attr_visibility_hidden
 extern const uint64_t lzma_crc64_table[4][256];
 #endif
 
diff --git a/src/liblzma/check/crc64_fast.c b/src/liblzma/check/crc64_fast.c
index e686dbd742f4..0c8622a1f367 100644
--- a/src/liblzma/check/crc64_fast.c
+++ b/src/liblzma/check/crc64_fast.c
@@ -206,6 +206,14 @@ calc_hi(uint64_t poly, uint64_t a)
 #if (defined(__GNUC__) || defined(__clang__)) && !defined(__EDG__)
 __attribute__((__target__("ssse3,sse4.1,pclmul")))
 #endif
+// The intrinsics use 16-byte-aligned reads from buf, thus they may read
+// up to 15 bytes before or after the buffer (depending on the alignment
+// of the buf argument). The values of the extra bytes are ignored.
+// This unavoidably trips -fsanitize=address so address sanitizier has
+// to be disabled for this function.
+#if lzma_has_attribute(__no_sanitize_address__)
+__attribute__((__no_sanitize_address__))
+#endif
 static uint64_t
 crc64_clmul(const uint8_t *buf, size_t size, uint64_t crc)
 {
diff --git a/src/liblzma/check/crc64_table.c b/src/liblzma/check/crc64_table.c
index 241adcd4a5cb..688e527b0fd1 100644
--- a/src/liblzma/check/crc64_table.c
+++ b/src/liblzma/check/crc64_table.c
@@ -18,10 +18,8 @@
 #if (defined(__x86_64__) && defined(__SSSE3__) \
 			&& defined(__SSE4_1__) && defined(__PCLMUL__)) \
 		|| (defined(__e2k__) && __iset__ >= 6)
-// No table needed but something has to be exported to keep some toolchains
-// happy. Also use a declaration to silence compiler warnings.
-extern const char lzma_crc64_dummy;
-const char lzma_crc64_dummy;
+// No table needed. Use a typedef to avoid an empty translation unit.
+typedef void lzma_crc64_dummy;
 
 #else
 // Having the declaration here silences clang -Wmissing-variable-declarations.
diff --git a/src/liblzma/common/common.c b/src/liblzma/common/common.c
index baad3dd84c7b..adb50d785d52 100644
--- a/src/liblzma/common/common.c
+++ b/src/liblzma/common/common.c
@@ -35,7 +35,8 @@ lzma_version_string(void)
 // Memory allocation //
 ///////////////////////
 
-extern void * lzma_attribute((__malloc__)) lzma_attr_alloc_size(1)
+lzma_attr_alloc_size(1)
+extern void *
 lzma_alloc(size_t size, const lzma_allocator *allocator)
 {
 	// Some malloc() variants return NULL if called with size == 0.
@@ -53,7 +54,8 @@ lzma_alloc(size_t size, const lzma_allocator *allocator)
 }
 
 
-extern void * lzma_attribute((__malloc__)) lzma_attr_alloc_size(1)
+lzma_attr_alloc_size(1)
+extern void *
 lzma_alloc_zero(size_t size, const lzma_allocator *allocator)
 {
 	// Some calloc() variants return NULL if called with size == 0.
diff --git a/src/liblzma/common/common.h b/src/liblzma/common/common.h
index 4d9cab53cdbb..378923e4012b 100644
--- a/src/liblzma/common/common.h
+++ b/src/liblzma/common/common.h
@@ -17,17 +17,28 @@
 #include "mythread.h"
 #include "tuklib_integer.h"
 
+// LZMA_API_EXPORT is used to mark the exported API functions.
+// It's used to define the LZMA_API macro.
+//
+// lzma_attr_visibility_hidden is used for marking *declarations* of extern
+// variables that are internal to liblzma (-fvisibility=hidden alone is
+// enough to hide the *definitions*). Such markings allow slightly more
+// efficient code to accesses those variables in ELF shared libraries.
 #if defined(_WIN32) || defined(__CYGWIN__)
 #	ifdef DLL_EXPORT
 #		define LZMA_API_EXPORT __declspec(dllexport)
 #	else
 #		define LZMA_API_EXPORT
 #	endif
+#	define lzma_attr_visibility_hidden
 // Don't use ifdef or defined() below.
 #elif HAVE_VISIBILITY
 #	define LZMA_API_EXPORT __attribute__((__visibility__("default")))
+#	define lzma_attr_visibility_hidden \
+			__attribute__((__visibility__("hidden")))
 #else
 #	define LZMA_API_EXPORT
+#	define lzma_attr_visibility_hidden
 #endif
 
 #define LZMA_API(type) LZMA_API_EXPORT type LZMA_API_CALL
@@ -87,6 +98,23 @@
 #	endif
 #endif
 
+// MSVC has __forceinline which shouldn't be combined with the inline keyword
+// (results in a warning).
+//
+// GCC 3.1 added always_inline attribute so we don't need to check
+// for __GNUC__ version. Similarly, all relevant Clang versions
+// support it (at least Clang 3.0.0 does already).
+// Other compilers might support too which also support __has_attribute
+// (Solaris Studio) so do that check too.
+#if defined(_MSC_VER)
+#	define lzma_always_inline __forceinline
+#elif defined(__GNUC__) || defined(__clang__) || defined(__INTEL_COMPILER) \
+		|| lzma_has_attribute(__always_inline__)
+#	define lzma_always_inline inline __attribute__((__always_inline__))
+#else
+#	define lzma_always_inline inline
+#endif
+
 // These allow helping the compiler in some often-executed branches, whose
 // result is almost always the same.
 #ifdef __GNUC__
@@ -297,14 +325,14 @@ struct lzma_internal_s {
 
 
 /// Allocates memory
-extern void *lzma_alloc(size_t size, const lzma_allocator *allocator)
-		lzma_attribute((__malloc__)) lzma_attr_alloc_size(1);
+lzma_attr_alloc_size(1)
+extern void *lzma_alloc(size_t size, const lzma_allocator *allocator);
 
 /// Allocates memory and zeroes it (like calloc()). This can be faster
 /// than lzma_alloc() + memzero() while being backward compatible with
 /// custom allocators.
-extern void * lzma_attribute((__malloc__)) lzma_attr_alloc_size(1)
-		lzma_alloc_zero(size_t size, const lzma_allocator *allocator);
+lzma_attr_alloc_size(1)
+extern void *lzma_alloc_zero(size_t size, const lzma_allocator *allocator);
 
 /// Frees memory
 extern void lzma_free(void *ptr, const lzma_allocator *allocator);
diff --git a/src/liblzma/common/index.c b/src/liblzma/common/index.c
index 97cc9f953fae..8a35f4398dbe 100644
--- a/src/liblzma/common/index.c
+++ b/src/liblzma/common/index.c
@@ -661,6 +661,12 @@ lzma_index_append(lzma_index *i, const lzma_allocator *allocator,
 	if (uncompressed_base + uncompressed_size > LZMA_VLI_MAX)
 		return LZMA_DATA_ERROR;
 
+	// Check that the new unpadded sum will not overflow. This is
+	// checked again in index_file_size(), but the unpadded sum is
+	// passed to vli_ceil4() which expects a valid lzma_vli value.
+	if (compressed_base + unpadded_size > UNPADDED_SIZE_MAX)
+		return LZMA_DATA_ERROR;
+
 	// Check that the file size will stay within limits.
 	if (index_file_size(s->node.compressed_base,
 			compressed_base + unpadded_size, s->record_count + 1,
diff --git a/src/liblzma/common/index.h b/src/liblzma/common/index.h
index 031efcc718f2..7b27d7004cfa 100644
--- a/src/liblzma/common/index.h
+++ b/src/liblzma/common/index.h
@@ -46,7 +46,7 @@ extern void lzma_index_prealloc(lzma_index *i, lzma_vli records);
 static inline lzma_vli
 vli_ceil4(lzma_vli vli)
 {
-	assert(vli <= LZMA_VLI_MAX);
+	assert(vli <= UNPADDED_SIZE_MAX);
 	return (vli + 3) & ~LZMA_VLI_C(3);
 }
 
diff --git a/src/liblzma/common/memcmplen.h b/src/liblzma/common/memcmplen.h
index 3c12422beaaf..99d9c519cc0d 100644
--- a/src/liblzma/common/memcmplen.h
+++ b/src/liblzma/common/memcmplen.h
@@ -49,7 +49,7 @@
 ///             It's rounded up to 2^n. This extra amount needs to be
 ///             allocated in the buffers being used. It needs to be
 ///             initialized too to keep Valgrind quiet.
-static inline uint32_t lzma_attribute((__always_inline__))
+static lzma_always_inline uint32_t
 lzma_memcmplen(const uint8_t *buf1, const uint8_t *buf2,
 		uint32_t len, uint32_t limit)
 {
diff --git a/src/liblzma/common/stream_flags_common.h b/src/liblzma/common/stream_flags_common.h
index 9f3122a3b1e5..84e96ba1ff66 100644
--- a/src/liblzma/common/stream_flags_common.h
+++ b/src/liblzma/common/stream_flags_common.h
@@ -18,7 +18,10 @@
 /// Size of the Stream Flags field
 #define LZMA_STREAM_FLAGS_SIZE 2
 
+lzma_attr_visibility_hidden
 extern const uint8_t lzma_header_magic[6];
+
+lzma_attr_visibility_hidden
 extern const uint8_t lzma_footer_magic[2];
 
 
diff --git a/src/liblzma/liblzma.pc.in b/src/liblzma/liblzma.pc.in
index 9fa489115a0a..d077cb77135a 100644
--- a/src/liblzma/liblzma.pc.in
+++ b/src/liblzma/liblzma.pc.in
@@ -15,5 +15,6 @@ Description: General purpose data compression library
 URL: @PACKAGE_URL@
 Version: @PACKAGE_VERSION@
 Cflags: -I${includedir}
+Cflags.private: -DLZMA_API_STATIC
 Libs: -L${libdir} -llzma
 Libs.private: @PTHREAD_CFLAGS@ @LIBS@
diff --git a/src/liblzma/lz/lz_encoder_hash.h b/src/liblzma/lz/lz_encoder_hash.h
index fb15c58146db..4d9971ae6a5d 100644
--- a/src/liblzma/lz/lz_encoder_hash.h
+++ b/src/liblzma/lz/lz_encoder_hash.h
@@ -17,6 +17,7 @@
 	// This is to make liblzma produce the same output on big endian
 	// systems that it does on little endian systems. lz_encoder.c
 	// takes care of including the actual table.
+	lzma_attr_visibility_hidden
 	extern const uint32_t lzma_lz_hash_table[256];
 #	define hash_table lzma_lz_hash_table
 #else
diff --git a/src/liblzma/lzma/fastpos.h b/src/liblzma/lzma/fastpos.h
index cba442c27e5c..dbeb16f7e31a 100644
--- a/src/liblzma/lzma/fastpos.h
+++ b/src/liblzma/lzma/fastpos.h
@@ -91,6 +91,7 @@ get_dist_slot_2(uint32_t dist)
 
 #define FASTPOS_BITS 13
 
+lzma_attr_visibility_hidden
 extern const uint8_t lzma_fastpos[1 << FASTPOS_BITS];
 
 
diff --git a/src/liblzma/lzma/fastpos_tablegen.c b/src/liblzma/lzma/fastpos_tablegen.c
index d4484c82d0b5..57ed15039b27 100644
--- a/src/liblzma/lzma/fastpos_tablegen.c
+++ b/src/liblzma/lzma/fastpos_tablegen.c
@@ -13,6 +13,8 @@
 
 #include <inttypes.h>
 #include <stdio.h>
+
+#define lzma_attr_visibility_hidden
 #include "fastpos.h"
 
 
diff --git a/src/liblzma/rangecoder/price.h b/src/liblzma/rangecoder/price.h
index 8ae02ca7474e..45dbbbb20cef 100644
--- a/src/liblzma/rangecoder/price.h
+++ b/src/liblzma/rangecoder/price.h
@@ -22,6 +22,7 @@
 
 
 /// Lookup table for the inline functions defined in this file.
+lzma_attr_visibility_hidden
 extern const uint8_t lzma_rc_prices[RC_PRICE_TABLE_SIZE];
 
 
diff --git a/src/lzmainfo/lzmainfo.c b/src/lzmainfo/lzmainfo.c
index b0ccdfb430ec..71e62958ad67 100644
--- a/src/lzmainfo/lzmainfo.c
+++ b/src/lzmainfo/lzmainfo.c
@@ -26,7 +26,8 @@
 #endif
 
 
-static void lzma_attribute((__noreturn__))
+tuklib_attr_noreturn
+static void
 help(void)
 {
 	printf(
@@ -45,7 +46,8 @@ _("Usage: %s [--help] [--version] [FILE]...\n"
 }
 
 
-static void lzma_attribute((__noreturn__))
+tuklib_attr_noreturn
+static void
 version(void)
 {
 	puts("lzmainfo (" PACKAGE_NAME ") " LZMA_VERSION_STRING);
diff --git a/src/xz/coder.c b/src/xz/coder.c
index 589ec072e63c..91d40ed2bb7b 100644
--- a/src/xz/coder.c
+++ b/src/xz/coder.c
@@ -128,7 +128,8 @@ coder_add_filter(lzma_vli id, void *options)
 }
 
 
-static void lzma_attribute((__noreturn__))
+tuklib_attr_noreturn
+static void
 memlimit_too_small(uint64_t memory_usage)
 {
 	message(V_ERROR, _("Memory usage limit is too low for the given "
diff --git a/src/xz/file_io.c b/src/xz/file_io.c
index a181b531258f..28280293ef39 100644
--- a/src/xz/file_io.c
+++ b/src/xz/file_io.c
@@ -944,20 +944,41 @@ io_open_dest_real(file_pair *pair)
 		}
 	}
 
-#ifndef TUKLIB_DOSLIKE
-	// dest_st isn't used on DOS-like systems except as a dummy
-	// argument to io_unlink(), so don't fstat() on such systems.
 	if (fstat(pair->dest_fd, &pair->dest_st)) {
 		// If fstat() really fails, we have a safe fallback here.
-#	if defined(__VMS)
+#if defined(__VMS)
 		pair->dest_st.st_ino[0] = 0;
 		pair->dest_st.st_ino[1] = 0;
 		pair->dest_st.st_ino[2] = 0;
-#	else
+#else
 		pair->dest_st.st_dev = 0;
 		pair->dest_st.st_ino = 0;
-#	endif
-	} else if (try_sparse && opt_mode == MODE_DECOMPRESS) {
+#endif
+	}
+#if defined(TUKLIB_DOSLIKE) && !defined(__DJGPP__)
+	// Check that the output file is a regular file. We open with O_EXCL
+	// but that doesn't prevent open()/_open() on Windows from opening
+	// files like "con" or "nul".
+	//
+	// With DJGPP this check is done with stat() even before opening
+	// the output file. That method or a variant of it doesn't work on
+	// Windows because on Windows stat()/_stat64() sets st.st_mode so
+	// that S_ISREG(st.st_mode) will be true even for special files.
+	// With fstat()/_fstat64() it works.
+	else if (pair->dest_fd != STDOUT_FILENO
+			&& !S_ISREG(pair->dest_st.st_mode)) {
+		message_error("%s: Destination is not a regular file",
+				pair->dest_name);
+
+		// dest_fd needs to be reset to -1 to keep io_close() working.
+		(void)close(pair->dest_fd);
+		pair->dest_fd = -1;
+
+		free(pair->dest_name);
+		return true;
+	}
+#elif !defined(TUKLIB_DOSLIKE)
+	else if (try_sparse && opt_mode == MODE_DECOMPRESS) {
 		// When writing to standard output, we need to be extra
 		// careful:
 		//  - It may be connected to something else than
@@ -1157,8 +1178,7 @@ io_fix_src_pos(file_pair *pair, size_t rewind_size)
 extern size_t
 io_read(file_pair *pair, io_buf *buf, size_t size)
 {
-	// We use small buffers here.
-	assert(size < SSIZE_MAX);
+	assert(size <= IO_BUFFER_SIZE);
 
 	size_t pos = 0;
 
@@ -1285,7 +1305,7 @@ is_sparse(const io_buf *buf)
 static bool
 io_write_buf(file_pair *pair, const uint8_t *buf, size_t size)
 {
-	assert(size < SSIZE_MAX);
+	assert(size <= IO_BUFFER_SIZE);
 
 	while (size > 0) {
 		const ssize_t amount = write(pair->dest_fd, buf, size);
diff --git a/src/xz/file_io.h b/src/xz/file_io.h
index 8a9e33630db3..6992efa4f86c 100644
--- a/src/xz/file_io.h
+++ b/src/xz/file_io.h
@@ -118,7 +118,7 @@ extern void io_close(file_pair *pair, bool success);
 ///
 /// \param      pair    File pair having the source file open for reading
 /// \param      buf     Destination buffer to hold the read data
-/// \param      size    Size of the buffer; assumed be smaller than SSIZE_MAX
+/// \param      size    Size of the buffer; must be at most IO_BUFFER_SIZE
 ///
 /// \return     On success, number of bytes read is returned. On end of
 ///             file zero is returned and pair->src_eof set to true.
@@ -172,7 +172,7 @@ extern bool io_pread(file_pair *pair, io_buf *buf, size_t size, uint64_t pos);
 ///
 /// \param      pair    File pair having the destination file open for writing
 /// \param      buf     Buffer containing the data to be written
-/// \param      size    Size of the buffer; assumed be smaller than SSIZE_MAX
+/// \param      size    Size of the buffer; must be at most IO_BUFFER_SIZE
 ///
 /// \return     On success, zero is returned. On error, -1 is returned
 ///             and error message printed.
diff --git a/src/xz/hardware.h b/src/xz/hardware.h
index 2bb3d7ba9493..a67b26efecb8 100644
--- a/src/xz/hardware.h
+++ b/src/xz/hardware.h
@@ -71,4 +71,5 @@ extern bool hardware_memlimit_mtenc_is_default(void);
 extern uint64_t hardware_memlimit_mtdec_get(void);
 
 /// Display the amount of RAM and memory usage limits and exit.
-extern void hardware_memlimit_show(void) lzma_attribute((__noreturn__));
+tuklib_attr_noreturn
+extern void hardware_memlimit_show(void);
diff --git a/src/xz/message.h b/src/xz/message.h
index b264f82153a3..f608ec75dffa 100644
--- a/src/xz/message.h
+++ b/src/xz/message.h
@@ -44,42 +44,44 @@ extern enum message_verbosity message_verbosity_get(void);
 /// \brief      Print a message if verbosity level is at least "verbosity"
 ///
 /// This doesn't touch the exit status.
-extern void message(enum message_verbosity verbosity, const char *fmt, ...)
-		lzma_attribute((__format__(__printf__, 2, 3)));
+lzma_attribute((__format__(__printf__, 2, 3)))
+extern void message(enum message_verbosity verbosity, const char *fmt, ...);
 
 
 /// \brief      Prints a warning and possibly sets exit status
 ///
 /// The message is printed only if verbosity level is at least V_WARNING.
 /// The exit status is set to WARNING unless it was already at ERROR.
-extern void message_warning(const char *fmt, ...)
-		lzma_attribute((__format__(__printf__, 1, 2)));
+lzma_attribute((__format__(__printf__, 1, 2)))
+extern void message_warning(const char *fmt, ...);
 
 
 /// \brief      Prints an error message and sets exit status
 ///
 /// The message is printed only if verbosity level is at least V_ERROR.
 /// The exit status is set to ERROR.
-extern void message_error(const char *fmt, ...)
-		lzma_attribute((__format__(__printf__, 1, 2)));
+lzma_attribute((__format__(__printf__, 1, 2)))
+extern void message_error(const char *fmt, ...);
 
 
 /// \brief      Prints an error message and exits with EXIT_ERROR
 ///
 /// The message is printed only if verbosity level is at least V_ERROR.
-extern void message_fatal(const char *fmt, ...)
-		lzma_attribute((__format__(__printf__, 1, 2)))
-		lzma_attribute((__noreturn__));
+tuklib_attr_noreturn
+lzma_attribute((__format__(__printf__, 1, 2)))
+extern void message_fatal(const char *fmt, ...);
 
 
 /// Print an error message that an internal error occurred and exit with
 /// EXIT_ERROR.
-extern void message_bug(void) lzma_attribute((__noreturn__));
+tuklib_attr_noreturn
+extern void message_bug(void);
 
 
 /// Print a message that establishing signal handlers failed, and exit with
 /// exit status ERROR.
-extern void message_signal_handler(void) lzma_attribute((__noreturn__));
+tuklib_attr_noreturn
+extern void message_signal_handler(void);
 
 
 /// Convert lzma_ret to a string.
@@ -100,11 +102,13 @@ extern void message_try_help(void);
 
 
 /// Prints the version number to stdout and exits with exit status SUCCESS.
-extern void message_version(void) lzma_attribute((__noreturn__));
+tuklib_attr_noreturn
+extern void message_version(void);
 
 
 /// Print the help message.
-extern void message_help(bool long_help) lzma_attribute((__noreturn__));
+tuklib_attr_noreturn
+extern void message_help(bool long_help);
 
 
 /// \brief      Set the total number of files to be processed
diff --git a/src/xz/options.c b/src/xz/options.c
index f46621311d92..4d5e899ccdc1 100644
--- a/src/xz/options.c
+++ b/src/xz/options.c
@@ -241,7 +241,8 @@ enum {
 };
 
 
-static void lzma_attribute((__noreturn__))
+tuklib_attr_noreturn
+static void
 error_lzma_preset(const char *valuestr)
 {
 	message_fatal(_("Unsupported LZMA1/LZMA2 preset: %s"), valuestr);
diff --git a/src/xz/util.c b/src/xz/util.c
index 9f9a8fb024e4..6ab4c2d776ce 100644
--- a/src/xz/util.c
+++ b/src/xz/util.c
@@ -17,9 +17,45 @@
 /// Buffers for uint64_to_str() and uint64_to_nicestr()
 static char bufs[4][128];
 
-/// Thousand separator support in uint64_to_str() and uint64_to_nicestr()
+
+// Thousand separator support in uint64_to_str() and uint64_to_nicestr():
+//
+// DJGPP 2.05 added support for thousands separators but it's broken
+// at least under WinXP with Finnish locale that uses a non-breaking space
+// as the thousands separator. Workaround by disabling thousands separators
+// for DJGPP builds.
+//
+// MSVC doesn't support thousand separators.
+#if defined(__DJGPP__) || defined(_MSC_VER)
+#	define FORMAT_THOUSAND_SEP(prefix, suffix) prefix suffix
+#	define check_thousand_sep(slot) do { } while (0)
+#else
+#	define FORMAT_THOUSAND_SEP(prefix, suffix) ((thousand == WORKS) \
+			? prefix "'" suffix \
+			: prefix suffix)
+
 static enum { UNKNOWN, WORKS, BROKEN } thousand = UNKNOWN;
 
+/// Check if thousands separator is supported. Run-time checking is easiest
+/// because it seems to be sometimes lacking even on a POSIXish system.
+/// Note that trying to use thousands separators when snprintf() doesn't
+/// support them results in undefined behavior. This just has happened to
+/// work well enough in practice.
+///
+/// This must be called before using the FORMAT_THOUSAND_SEP macro.
+static void
+check_thousand_sep(uint32_t slot)
+{
+	if (thousand == UNKNOWN) {
+		bufs[slot][0] = '\0';
+		snprintf(bufs[slot], sizeof(bufs[slot]), "%'u", 1U);
+		thousand = bufs[slot][0] == '1' ? WORKS : BROKEN;
+	}
+
+	return;
+}
+#endif
+
 
 extern void *
 xrealloc(void *ptr, size_t size)
@@ -142,31 +178,6 @@ round_up_to_mib(uint64_t n)
 }
 
 
-/// Check if thousands separator is supported. Run-time checking is easiest
-/// because it seems to be sometimes lacking even on a POSIXish system.
-/// Note that trying to use thousands separators when snprintf() doesn't
-/// support them results in undefined behavior. This just has happened to
-/// work well enough in practice.
-///
-/// DJGPP 2.05 added support for thousands separators but it's broken
-/// at least under WinXP with Finnish locale that uses a non-breaking space
-/// as the thousands separator. Workaround by disabling thousands separators
-/// for DJGPP builds.
-static void
-check_thousand_sep(uint32_t slot)
-{
-	if (thousand == UNKNOWN) {
-		bufs[slot][0] = '\0';
-#ifndef __DJGPP__
-		snprintf(bufs[slot], sizeof(bufs[slot]), "%'u", 1U);
-#endif
-		thousand = bufs[slot][0] == '1' ? WORKS : BROKEN;
-	}
-
-	return;
-}
-
-
 extern const char *
 uint64_to_str(uint64_t value, uint32_t slot)
 {
@@ -174,10 +185,8 @@ uint64_to_str(uint64_t value, uint32_t slot)
 
 	check_thousand_sep(slot);
 
-	if (thousand == WORKS)
-		snprintf(bufs[slot], sizeof(bufs[slot]), "%'" PRIu64, value);
-	else
-		snprintf(bufs[slot], sizeof(bufs[slot]), "%" PRIu64, value);
+	snprintf(bufs[slot], sizeof(bufs[slot]),
+			FORMAT_THOUSAND_SEP("%", PRIu64), value);
 
 	return bufs[slot];
 }
@@ -201,10 +210,8 @@ uint64_to_nicestr(uint64_t value, enum nicestr_unit unit_min,
 	if ((unit_min == NICESTR_B && value < 10000)
 			|| unit_max == NICESTR_B) {
 		// The value is shown as bytes.
-		if (thousand == WORKS)
-			my_snprintf(&pos, &left, "%'u", (unsigned int)value);
-		else
-			my_snprintf(&pos, &left, "%u", (unsigned int)value);
+		my_snprintf(&pos, &left, FORMAT_THOUSAND_SEP("%", "u"),
+				(unsigned int)value);
 	} else {
 		// Scale the value to a nicer unit. Unless unit_min and
 		// unit_max limit us, we will show at most five significant
@@ -215,21 +222,15 @@ uint64_to_nicestr(uint64_t value, enum nicestr_unit unit_min,
 			++unit;
 		} while (unit < unit_min || (d > 9999.9 && unit < unit_max));
 
-		if (thousand == WORKS)
-			my_snprintf(&pos, &left, "%'.1f", d);
-		else
-			my_snprintf(&pos, &left, "%.1f", d);
+		my_snprintf(&pos, &left, FORMAT_THOUSAND_SEP("%", ".1f"), d);
 	}
 
 	static const char suffix[5][4] = { "B", "KiB", "MiB", "GiB", "TiB" };
 	my_snprintf(&pos, &left, " %s", suffix[unit]);
 
-	if (always_also_bytes && value >= 10000) {
-		if (thousand == WORKS)
-			snprintf(pos, left, " (%'" PRIu64 " B)", value);
-		else
-			snprintf(pos, left, " (%" PRIu64 " B)", value);
-	}
+	if (always_also_bytes && value >= 10000)
+		snprintf(pos, left, FORMAT_THOUSAND_SEP(" (%", PRIu64 " B)"),
+				value);
 
 	return bufs[slot];
 }
diff --git a/src/xz/util.h b/src/xz/util.h
index 4a536f52ea4e..6d7e1481863b 100644
--- a/src/xz/util.h
+++ b/src/xz/util.h
@@ -19,12 +19,12 @@
 
 
 /// \brief      Safe realloc() that never returns NULL
-extern void *xrealloc(void *ptr, size_t size)
-		lzma_attribute((__malloc__)) lzma_attr_alloc_size(2);
+lzma_attr_alloc_size(2)
+extern void *xrealloc(void *ptr, size_t size);
 
 
 /// \brief      Safe strdup() that never returns NULL
-extern char *xstrdup(const char *src) lzma_attribute((__malloc__));
+extern char *xstrdup(const char *src);
 
 
 /// \brief      Fancy version of strtoull()
@@ -101,8 +101,8 @@ extern const char *uint64_to_nicestr(uint64_t value,
 ///
 /// A maximum of *left bytes is written starting from *pos. *pos and *left
 /// are updated accordingly.
-extern void my_snprintf(char **pos, size_t *left, const char *fmt, ...)
-		lzma_attribute((__format__(__printf__, 3, 4)));
+lzma_attribute((__format__(__printf__, 3, 4)))
+extern void my_snprintf(char **pos, size_t *left, const char *fmt, ...);
 
 
 /// \brief      Test if stdin is a terminal
diff --git a/src/xz/xz.1 b/src/xz/xz.1
index 8e85a17e920e..73ca6efd6aff 100644
--- a/src/xz/xz.1
+++ b/src/xz/xz.1
@@ -1160,7 +1160,6 @@ to
 resets the
 .I limit
 to the default system-specific value.
-.IP ""
 .TP
 \fB\-M\fR \fIlimit\fR, \fB\-\-memlimit=\fIlimit\fR, \fB\-\-memory=\fIlimit
 This is equivalent to specifying
diff --git a/src/xzdec/xzdec.c b/src/xzdec/xzdec.c
index c1bd19986424..556c548d05ad 100644
--- a/src/xzdec/xzdec.c
+++ b/src/xzdec/xzdec.c
@@ -40,7 +40,8 @@
 static int display_errors = 2;
 
 
-static void lzma_attribute((__format__(__printf__, 1, 2)))
+lzma_attribute((__format__(__printf__, 1, 2)))
+static void
 my_errorf(const char *fmt, ...)
 {
 	va_list ap;
@@ -57,7 +58,8 @@ my_errorf(const char *fmt, ...)
 }
 
 
-static void lzma_attribute((__noreturn__))
+tuklib_attr_noreturn
+static void
 help(void)
 {
 	printf(
@@ -81,7 +83,8 @@ PACKAGE_NAME " home page: <" PACKAGE_URL ">\n", progname);
 }
 
 
-static void lzma_attribute((__noreturn__))
+tuklib_attr_noreturn
+static void
 version(void)
 {
 	printf(TOOL_FORMAT "dec (" PACKAGE_NAME ") " LZMA_VERSION_STRING "\n"