Emitting -undefined,error was correct,, but starting with Xcode 15 / Sonoma,
doing so triggers "ld: warning: -undefined error is deprecated". Given that
"-undefined error" is documented to be the linker's default behaviour, this
warning seems ill advised. However, it does create a lot of noise. As
"-undefined error" is the default behaviour, the least bad way to deal with
this seems to be to just not emit anything. Of course that only works as long
as nothing else injects -undefined dynamic_lookup, or such. Complain to Apple.
Fixes: https://github.com/mesonbuild/meson/issues/12450
no_warn_args is unused. Its only purpose was to implement automatic
hiding of UB in transpiled code, and it was not used at all in languages
other than C/C++ -- specifically when the C/C++ source files were
created by transpiling from vala or cython.
This was originally added for vala only, with the rationale that vala
generates bad code that has warnings. Unfortunately, the rationale was
fatally flawed. The compiler warns about a number of things, which the
user can control depending on their code (or their code generator's
code), but some of those things are absolutely critical to warn about.
In particular, GCC 14 and clang 17 are updating their defaults to warn
-- and error by default for -- invalid C code that breaks the standard,
but has been silently accepted for over 20 years "because lots of people
do it". The code in question is UB, and compilers will generate faulty
machine code that behaves erroneously and probably has a mass of CVEs
waiting to happen.
Compiler warnings are NOT safe to just... universally turn off. Compiler
warnings could be either:
- coding style lints
- threatening statements that the code is factually and behaviorally wrong
There is no magic bullet to ignore the former while respecting the
latter. And the very last thing we should ever do is pass `-w`, since
that causes ALL warnings to be disabled, even the manually added
`-Werror=XXX`.
If vala generated code creates warnings, then the vala compiler can
decrease the log level by generating better code, or by adding warning
suppression pragmas for *specific* issues, such as unused functions.
The new linker in Sonoma / Xcode 15 considers the dependency via the
initializer sufficient to pull in the library. The man page now notes:
The linker never dead strips initialization and termination routines.
They are considered "roots" of the dead strip graph.
I could not find a good way to skip only if the linker version is new
enough. Before long everyone will be using the new linker anyway...
This was encountered while looking into an issue with
https://github.com/NixOS/nixpkgs/pull/268583.
I run my Nix store on case-sensitive APFS, so the test fails due to
trying to link `-framework ldap` instead of `-framework LDAP`.
The code assumed that sysconfig.get_platform() returns "mingw" for mingw Python,
but that's no longer the case for 2.5 years now, as it now only starts with
"mingw" and contains further information like the arch and other ABI relevant things
to avoid conflicts.
This updates the detection code to the current status quo. mingw Python only documents
right now that it starts with "mingw", and none of that arch stuff, but it's
unlikely that this will change, and this looks less error prone than looking at CC.
Fixes#12547
Previously macos reported "ld: unknown option: --version" when being passed
--version, but now sometimes it reports in plural, albeit without an obvious
pattern when. To handle that, simply just check for the prefix without the :
Fixes: https://github.com/mesonbuild/meson/issues/12552
If an annotation could not be resolved, it's classified as a "missing
import" and our configuration ignored it:
```
Skipping analyzing "mesonbuild.backends": module is installed, but missing library stubs or py.typed marker
```
As far as mypy is concerned, this library may or may not exist, but it
doesn't have any typing information at all (may need to be installed
first).
We ignored this because of our docs/ and tools/ thirdparty dependencies,
but we really should not. It is trivial to install them, and then
enforce that this "just works".
By enforcing it, we also make sure typos get caught.
This is needed now that str.format() is not allowing it any more. It is
also more consistent with other objects that have that method as well,
such as build targets.
Fixes: #12406
Releases have been happening an average of once every 90 days for the
past two years (since 0.60.0). If we just look at releases since 1.0.0,
the average is over 100 days.
[why]
On Apple clang 15.0.0 linker (i.e. ld64 1015.7) giving the same rpath
multiple times raises a warning:
ld: warning: duplicate -rpath '/local/lib' ignored
This can frequently happen when linking several dependencies that all
have that rpath in e.g. pkgconfig.
[how]
Deduplicate all rpath arguments.
[note]
I'm not sure how the code handles --start/end-group, but for rpath that
should not make any difference as that is not bound to a group.
Signed-off-by: Fini Jastrow <ulf.fini.jastrow@desy.de>
Whilst working on the Reproducible Builds effort, we noticed that
meson was generates .pkgconfig files that are not reproducible.
For example, here is neatvnc's pkgconfig file when built with HEAD^1:
Name: neatvnc
Description: A Neat VNC server library
Version: 0.7.0
-Requires.private: pixman-1, aml < 0.4.0, aml >= 0.3.0, zlib, libdrm, libturbojpeg, gnutls, nettle, hogweed, gmp, gbm, libavcodec, libavfilter, libavutil
+Requires.private: pixman-1, aml >= 0.3.0, aml < 0.4.0, zlib, libdrm, libturbojpeg, gnutls, nettle, hogweed, gmp, gbm, libavcodec, libavfilter, libavutil
Libs: -L${libdir} -lneatvnc
Libs.private: -lm
Cflags: -I${includedir}
This is, ultimately, due to iterating over the contents of a set within a
DefaultDict and can thus be fixed by sorting the output immediately prior to
generating the Requires.private string.
An alternative solution would be to place the sorted(…) call a few lines
down:
return ', '.join(sorted(result))
However, this changes the expected ordering of the entire line, and many users
may be unhappy with that (alternative) change as a result. By contrast, this
commit will only enforce an ordering when there are multiple version
requirements (eg. a lower and a higher version requirement, ie. a version
range). It will, additionally, order them with the lower part of the range
first.
This was originally filed (with a slightly different patch) by myself in
the the Debian bug tracker <https://bugs.debian.org/1056117>.
Signed-off-by: Chris Lamb <lamby@debian.org>
EmbedManifest seems to default to true, which creates a default manifest based
on other parameters (likewise defaults) and makes it impossible to supply your
own with CREATEPROCESS_MANIFEST_RESOURCE_ID. There is value to being able to do
this and no value to the default one, so this should be disabled.
* unity builds: correct integer ceiling division
* edge case failure with unity builds:
- static archive bar that gets installed, that links with another static
archive foo that does not get installed
- the number of files in static archive foo is divisible by unity_size
would yield an error with ninja:
ninja: error: 'subprojects/foo/src/libfoo.a.p/meson-generated_foo-unity1.cpp.o', needed by 'src/libbar.a', missing and no known rule to make it
* unity builds: test for build failure when #files is divisible by unity_size
It was previously impossible to do this:
```
dep.get_pkgconfig_variable(
'foo',
define_variable: ['prefix', '/usr', 'datadir', '/usr/share'],
)
```
since get_pkgconfig_variable mandated exactly two (if any) arguments.
However, you could do this:
```
dep.get_variable(
'foo',
pkgconfig_define: ['prefix', '/usr', 'datadir', '/usr/share'],
)
```
It would silently do the wrong thing, by defining "prefix" as
`/usr=datadir=/usr/share`, which might not "matter" if only datadir was
used in the "foo" variable as the unmodified value might be adequate.
The actual intention of anyone writing such a meson.build is that they
aren't sure whether the .pc file uses ${prefix} or ${datadir} (or which
one gets used, might have changed between versions of that .pc file,
even).
A recent refactor made this into a hard error, which broke some projects
that were doing this and inadvertently depending on some .pc file that
only used the second variable. (This was "fine" since the result was
essentially meaningful, and even resulted in behavior identical to the
intended behavior if both projects were installed into the same prefix
-- in which case there's nothing to remap.)
Re-allow this. There are two ways we could re-allow this:
- ignore it with a warning
- add a new feature to allow actually doing this
Since the use case which triggered this bug actually has a pretty good
reason to want to do this, it makes sense to add the new feature.
Fixes https://bugs.gentoo.org/916576
Fixes https://github.com/containers/bubblewrap/issues/609
This time we have a case where people are passing non-objects by using
them as str | File, which we never warned about and silently accepted.
If it was passed via custom_target outputs we *would* error out,
interestingly enough. At the backend layer, we just pass them directly
to the linker... which is valid, if we misdetected what's a valid linker
input or people just used funny names. In particular, the mingw
toolchain allows passing a *.def file directly, and some people are
doing that.
If we do want to allow this, we should do it consistently. For now, just
follow the current theme of what's expected, but do so by warning
instead of fatally erroring, for cases where users were able to do it in
the past.
In the case r1 -> s1 -> s2 where s1 and s2 are uninstalled C static
libraries, the libs1.a rule does not depend on libs2.a. That means that
r1 rule must depend on both s1 and s2.
We haven't actually verified that these kwargs are equal to what we had
before, and should probably revert the entire series. But I have
multiple reports in the wild of projects that no longer build because of
`install: [true, false, get_option('foobar')]` which was always
incorrect and always equal to just dropping values all over the floor
and treating it the same as "bool(value) == True".
Special case this particular typed kwarg and allow it with a sternly
worded warning that it was always wrong and should never ever ever be
done.
Fixes: https://bugs.gentoo.org/917118
Fixes: http://qa-logs.debian.net/2023/11/11/rhythmbox_3.4.7-1_unstable_meson-exp.log
Thanks to the Gentoo Tinderbox project, and Lucas Nussbaum of the Debian
project.
Pass link arguments directly down to linker by using `-C link-args=`
instead of letting rustc/linker resolve `-l` arguments. This solves
problems with e.g. +verbatim not being portable. Note that we also pass
`-l` args as `-Clink-args=-l` because rustc would otherwise reorder
arguments and put `-lstdc++` before `-Clink-args=libfoo++.a`.
However, when building a rlib/staticlib we should still use `-l`
arguments because that allows rustc to bundle static libraries we
link-whole. In that case, since there is no platform specific dynamic
linker, +verbatim works.
This also fix installed staticlib that now bundle uninstalled static
libraries it links to (recursively). This is done by putting them all
into self.link_whole_targets instead of putting their objects into
self.objects, and let rustc do the bundling. This has the extra
advantage that rustc can bundle static libries built with CustomTarget.
Disable bundling in all other cases, otherwise we could end up with
duplicated objects in static libraries, in diamond dependency graph
case.
Fixes: #12484
Andres Freund