Merge meson 1.12.1 - #29
Merged
rgommers merged 682 commits intoSep 29, 2026
Merged
Conversation
This means that mypy will yell if someone tries to mutate one of these values.
…bclassing The code I came up with works with mypy for versions of Python >= 3.12, but for 3.11 and 3.10 it still complains that _BASE_CLASS isn't a static type. The problem is that in those versions `pathlib.Path` is a factory instead of a type. But we can side step that because this hack exists for Windows and only, and just directly inherit the WindowsPath
Drop the manual list of modules, and just add the mesonbuild/ directory. This adds __init__.py, _pathlib.py and mesonlib.py, which were previously not included. _pathlib.py needs workarounds for Python versions up to 3.11. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Upon further testing with a more complex project, I realised that Nasm wants newlines and does NOT accept quotes+spaces for splitting arguments. So for proper escaping this must be done with a custom generator à la GNOME module, with a CustomTarget. This reverts commit 173d760.
It requires a special flag `-@foo.rsp` and its contents must be separated with newlines. Both must be done externally to ninjabackend as the latter can only handle newlines in input objects, not in arguments.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
[Paolo: add test]
normal compilers does not rely on the position of this argument, but MSVC-like compilers treat `/link` and any argument after that as linker arguments. MSVC-like compiler's linker_to_compiler_args inject `/link` before largs, resulting it being forwarded to link.exe not the compiler driver. fixes cross compiling with cc = clang-cl and c_ld = lld-link.
this function does not pass enough args from the compiler object to find windows sdk without relying on environment variables. including a file from the same directory removes the dependency on stdlib.
…ixin clang-cl supports C++11 natively, so CPP11AsCPP14Mixin is not needed. Override get_option_std_args to use /clang:-std=c++N for all standards except c++latest/vc++latest which still require /std:c++latest. Add c++23 and vc++23 to the supported standards list. Fixes mesonbuild#15845
Fuchsia is a microkernel-like OS that takes many cues from Plan9. Co-Authored-by: Dylan Baker <dylan@pnwbakers.com>
If doing cc.get_define() with MSVC, for a symbol that isn't defined, then get_define() can end up returning sentinel_undef with a trailing space, as an actual defined string. This happens if MSVC is called with -std:c11 (or -Zc:preprocessor, to opt in to the newer, standards conformant preprocessor mode) and the test source file is written with Unix newlines. (This is observed if cross compiling with MSVC wrapped in Wine on Unix, but probably would appear the same if Meson was running in Cygwin/msys2 Python as well.) This issue has been reported upstream to MSVC at [1]. [1] https://developercommunity.visualstudio.com/t/Inconsistent-whitespace-with-standard-pr/11023343
Targets created from functions such as `install_symlink` were not automatically tagged as "runtime" when installed under sbindir like it does when installed under bindir.
I have two signing keys. One is an older rsa4096 one, the newer ed25519 key I created when joining the Gentoo Developers team. Both keys are still valid, and the newer key is signed by the older key. When tagging release 1.11.1, I mistakenly uploaded assets signed by the newer key, which should not actually be a big deal but was confusing since it was not on this page. Update the docs so that people know this key is acceptable. Proofs for the new key: - cross-sig from the old to new key - recorded by https://github.com/eli-schwartz.gpg as a known key for me - recorded by https://codeberg.org/eschwartz.gpg as a known key for me - this git commit adding the new key to the docs page is signed by the old key - my online identity as a Gentoo Developer is associated with this key as observable via: https://wiki.gentoo.org/wiki/User:Eschwartz https://www.gentoo.org/inside-gentoo/developers/#eschwartz Fixes: mesonbuild#15901
The rc helper script uses flags specific for MSVC, which clang does not understand: > [19/1419] Compiling Windows resource subprojects/ffmpeg/windows_compile_resources_0_avfilterres.rc > clang: error: no such file or directory: '/showIncludes' > clang: error: no such file or directory: '/EP' > clang: error: no such file or directory: '/nologo' > clang: error: no such file or directory: '/DRC_INVOKED' > clang: error: no such file or directory: '/Tcsubprojects/ffmpeg/avfilterres.rc' > clang: error: no input files
Currently, `meson format --recursive` only recurse meson.build files that are reachable via a `subdir` function. Add `--subprojects` flag to also recurse subprojects.
As visible from generate_pbx_group(), folder_children is a PbxArray. Use the same variable names in write_group_target_entry() for clarity. Fixes: mesonbuild#15905 Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add a clippy-json target to the ninja script. This is useful for rust-analyzer, which requires an "external check command" to function when not using cargo. Also we add a "clippy-json-prereq" target, and invoke it in the clippy.py script. The prereq target tries to build as much of the project as possible with "-k0".
include_directories() can point to a build_subdir that a custom_target() generated into, so the subdir must exist in the build tree at setup time. Create a new function to abstract os.makedirs() and use it validate_build_subdir(). Fixes: mesonbuild#15902 Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This reverts commit f7758b9. Deduplicating OpenMP linker arguments is in principle a good idea, but it breaks -Xpreprocessor -fopenmp because deduplication has no idea of which options have an argument in a separate entry of the array. Fixes: mesonbuild#15764 Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These were used by the old argument checking infrastructure, which has all been removed.
detect_subsystem(system) is just a simple transformation if system; do not require setting machine_info.subsystem at all costs, and if it is unset just assume that it is okay since the system has already been matched. Regression introduced by commit 61c5859 ("envconfig: Cannot run executables if subsystem doesn't match", 2026-02-04). Fixes: mesonbuild#15840 Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Instead of adding the relaxation universally, just comment out the raise statement. This allows the CMake module case (where the relaxation is added intentionally) to not warn. Add a test case that the warning happens, though. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The get_all_linked_targets() method is incomplete, in that it does not check other targets that provide objects for the target. Because information about objects is not available in Build, move the function to Backend. Note that this drops the lru_cache; it is not necessary because the function is called in two places and each of them runs at most once per target. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> (cherry picked from commit 88c5b5d)
The get_all_linked_targets() method is incomplete, in that it does not check other targets that provide objects for the target. While ninjabackend.py does use `flatten_object_list` on the target itself, it does not check any link_with or link_whole targets. And because the static library in a `BothLibraries` object might actually have no sources and have just an `ExtractedObjects` entry instead, using `link_with: foo.get_static_lib()` in a Fortran program is broken. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> (cherry picked from commit bb575c2)
This removes duplicate code from the ninja backend. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> (cherry picked from commit 495f6d7)
Because extract_objects() will only look for sources, and not for anything passed in `objects` for the target, it is by definition non-recursive. Only extract_all_objects() can return recursive ExtractedObjects. This ensures that all the recursive calls of _flatten_object_list() receive exactly the same objlist argument for a given target, instead of also getting recursive calls with empty objlists. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
_flatten_object_list can reach the same target through multiple ExtractObjects, which causes an exponential blowup in the number of recursive calls. Introduce a set of visited dependencies (which has to be returned anyway) and use it to ensure that each distinct target is visited exactly once. Depth-first order (aka post-order) ensures that each target using extract_all_objects(recursive: true) is visited after the target providing the objects, so perform a DFS and flatten each target one by one: later visits will find the result of earlier visits already.
Debian/Ubuntu have changed their packaging such that libclang-dev no longer includes libclang-rt-dev. We require the libraries in libclang-rt-dev to run some of our tests. Signed-off-by: Tristan Partin <tristan@partin.io> Part-of: mesonbuild#16165 Closes: mesonbuild#16165 Signed-off-by: Sam James <sam@gentoo.org> (cherry picked from commit 34767ed)
(cherry picked from commit f3b78ee)
Signed-off-by: Tristan Partin <tristan@partin.io> (cherry picked from commit 570963b)
We would just run a loop of `x = y while x exists { x = f'{y}0' }`
which... means if `f'{y}0` exists we'll loop forever.
(cherry picked from commit 4d5484c)
There is no need to pass link-only arguments where we are only compiling. This has led to real world issues: Clang will emit a warning when unused arguments are passed on the command line: clang: warning: argument unused during compilation: '-L/tmp/opencode' [-Wunused-command-line-argument] This is a bit of a hard warning to trigger. The following is required: 1. Clang or Clang-derived compiler (clang-cl) 2. A cross build 3. No exe wrapper configured 4. c_link_args is non-empty When (2) and (3) hold, the compile mode becomes CompileCheckMode.COMPILE, which means we pass -c to clang's sanity check command. When -c is passed to clang, it ignores linker arguments like -L. Signed-off-by: Tristan Partin <tristan@partin.io> Part-of: mesonbuild#16142 Closes: mesonbuild#16142 Signed-off-by: Sam James <sam@gentoo.org> (cherry picked from commit c9bb03c)
Commit 300d7c3 ("compilers: add link always args if CompileCheckMode.LINK") made has_function() and friends pass the linker's always args, so that e.g. -Zomf is not missing on OS/2 when a dependency is used. Before changing where those arguments are added, add a test. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Commit c9bb03c ("Stop passing link arguments to compile-only sanity checks", 2026-08-28) dropped link arguments from compile-only checks for CLikeCompiler. Introduce Compiler._sanity_check_mode() to move this logic up to Compiler and so that it applies also to the super() call of CLikeCompiler._sanity_check_compile_args(). Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Commit 797decc ("compilers: pass external args to sanity check") made Compiler._sanity_check_compile_args() add {c,cpp}_args and {c,cpp}_link_args, assuming the extra copies on the CLike path were harmless because "arglist deduplicates them", but the sanity check builds a plain list so nothing is deduplicated. Furthermore, the copy of linker arguments in s_largs would end up in front of the /link that linker_to_compiler_args() inserts for MSVC-style compilers. For example, with c_link_args=['/SUBSYSTEM:CONSOLE'] the probe became cl ... /c /SUBSYSTEM:CONSOLE /link /SUBSYSTEM:CONSOLE (the first from super()._sanity_check_compile_args, the second from self._get_basic_compiler_args) instead of cl ... /c /link /SUBSYSTEM:CONSOLE Stop using super()._sanity_check_compile_args() to remove the duplication while keeping control of the position of the linker arguments. Note that get_always_args() was also duplicated because _get_basic_compiler_args() returns it via get_compiler_args_for_mode(), so drop it as well. The effect is to remove the copy of get_always_args() that was previously in front of the output argument, and only keep the one after the source file, but this is harmless. Fixes: mesonbuild#16111 Fixes: 797decc ("compilers: pass external args to sanity check") Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Right now they are passed twice, but this will be fixed. Add a partial test for now. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Commit 300d7c3 ("compilers: add link always args if CompileCheckMode.LINK", 2026-01-07) added get_linker_always_args() to get_compiler_args_for_mode(), but that function returns *compiler* arguments, and it is applied twice to a compile check: once by CLikeCompiler._get_basic_compiler_args() and once by Compiler.compile(). For MSVC this puts linker-only flags on the compiler command line twice *and* in the wrong place: cl ... /release /nologo ... /release /nologo ... /link foo.lib Add them to the linker argument list that is passed through linker_to_compiler_args() instead of going through get_compiler_args_for_mode(). While that appears twice (in build_wrapper_args() and in the sanity check), it is not necessary anymore to add linker always args in find_library() now that build_wrapper_args() adds them. Note that this does not include the Ninja backend which has a call to get_linker_always_args() of its own. Fixes: mesonbuild#16086 Fixes: 300d7c3 ("compilers: add link always args if CompileCheckMode.LINK", 2026-01-07) Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Commit decc3e assumed the sanity check goes through CompilerArgs, but it does not. Add to _sanity_check_compile_args() the missing call to DCompiler.unix_args_to_native(). Fixes: mesonbuild#16115 Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Avoid a repeated get_value_for() idiom. Rust wraps the linker arguments with -Clink-arg as usual; this is a bug fix for cc.links(), cc.has_function() and friends which previously received the raw arguments. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
rustc's sanity check passes no external args at all. Fix it for consistency with other languages and to ensure that failures are detected early. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
…ses or mixins This is indicated as a problem in comments, and it causes problem with recent changes to the sanity check command lines. Get -fpermissive from a separate method, and implement it where it was used before. Fixes: mesonbuild#16203 Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Match what GNU-like compiler does, helping with c++NN std. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Package seems to be unavailable for the gccx86ninja job's architecture.
Member
Author
|
Are tests expected to pass? I see failures like |
Member
|
Thanks @mattip. You can validate this by opening a draft PR to the main repo that updates the vendored CI here can be ignored, it has never passed. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The merge of 1.12.1 into main-numpymeson-1.11 worked cleanly. I tried this locally and it seems to build NumPy. What is the correct way to validate that everything is correct?
Replaces #28