Fix build issues for LLVM based toolchains. #1513
Conversation
|
|
| if is_libcxx | ||
| message('Detected libcxx, not appending -static-libstc++ to linker arguments') | ||
| else | ||
| link_args+=['-static-libstdc++'] |
There was a problem hiding this comment.
@flightlessmango can't this be left for packagers / users to add with meson with -D cpp_link_args when/if needed ?
There was a problem hiding this comment.
Maybe a case where its enabled by default and you have to disable it yourself for the cases where its not desirable (libcxx / no static libstdc++ available).
src/meson.build
Outdated
| objects: mangohud_static_lib.extract_all_objects(), | ||
| link_with: mangohud_static_lib, | ||
| link_args : link_args, | ||
| link_args : [link_args, '-Wl,--version-script,@0@'.format(join_paths(meson.current_source_dir(), 'mangohud.version'))], |
There was a problem hiding this comment.
This is a lot of duplication, should probably be a variable
There was a problem hiding this comment.
Not sure about how to do that nicer, I have unique files for all libs and mangohub_opengl uses a different directory as its generated at build time.
Could also break it into several lines like in the example given in this meson issue about improving mesons handling for these.
parona-source
force-pushed
the
llvm-build
branch
from
ae83e08 to
6c6f4ea
Compare
parona-source
force-pushed
the
llvm-build
branch
from
6c6f4ea to
c768be5
Compare
LLD is stricter with version scripts and will complain about nonexisting symbols in version files. Bump meson reguirement due to use of compiler_args in preprocess. Signed-off-by: Alfred Wingate <parona@protonmail.com>
Clang has a long standing bug where it will try to mix c++ libraries instead of erroring out early or transparently linking against libcxx statically instead. Signed-off-by: Alfred Wingate <parona@protonmail.com>
LLVM alternative C++ standard library is happier if it can hide its own
symbols first without preprocessor macros complicating things.
FAILED: amdgpu.p/tests_test_amdgpu.cpp.o
clang++ -Iamdgpu.p -I. -I.. -I../include -I../subprojects/spdlog-1.14.1/include -Isubprojects/implot-0.16 -I../subprojects/implot-0.16 -Isubprojects/imgui-1.89.9 -I../subprojects/imgui-1.89.9 -I../subprojects/imgui-1.89.9/backends -fdiagnostics-color=always -D_FILE_OFFSET_BITS=64 -Wall -Winvalid-pch -Wextra -std=c++14 -O3 -Werror=missing-declarations -Werror=return-type -Wno-unused-parameter -Qunused-arguments -fno-math-errno -fno-trapping-math -Wno-non-virtual-dtor -Wno-missing-field-initializers -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS '-DPACKAGE_VERSION="v0.7.2"' -DSPDLOG_ACTIVE_LEVEL=SPDLOG_LEVEL_DEBUG -D_GNU_SOURCE -DHAVE_PTHREAD -DUSE_GCC_ATOMIC_BUILTINS -DHAVE_TIMESPEC_GET -DHAVE___BUILTIN_BSWAP32 -DHAVE___BUILTIN_BSWAP64 -DHAVE___BUILTIN_CLZ -DHAVE___BUILTIN_CLZLL -DHAVE___BUILTIN_CTZ -DHAVE___BUILTIN_EXPECT -DHAVE___BUILTIN_FFS -DHAVE___BUILTIN_FFSLL -DHAVE___BUILTIN_POPCOUNT -DHAVE___BUILTIN_POPCOUNTLL -DHAVE___BUILTIN_UNREACHABLE -O2 -pipe '-DIMGUI_API=__attribute__((visibility("default")))' -pthread -DSPDLOG_COMPILED_LIB -DTEST_ONLY -MD -MQ amdgpu.p/tests_test_amdgpu.cpp.o -MF amdgpu.p/tests_test_amdgpu.cpp.o.d -o amdgpu.p/tests_test_amdgpu.cpp.o -c ../tests/test_amdgpu.cpp
In file included from ../tests/test_amdgpu.cpp:9:
In file included from ../tests/../src/amdgpu.h:6:
In file included from ../tests/../src/overlay_params.h:6:
In file included from /usr/include/c++/v1/vector:3023:
In file included from /usr/include/c++/v1/locale:212:
/usr/include/c++/v1/ios:343:30: error: expected parameter declarator
343 | _LIBCPP_HIDE_FROM_ABI bool fail() const;
| ^
/usr/include/cmocka.h:1616:22: note: expanded from macro 'fail'
1616 | #define fail() _fail(__FILE__, __LINE__)
| ^
<scratch space>:2:1: note: expanded from here
2 | "/usr/include/c++/v1/ios"
Signed-off-by: Alfred Wingate <parona@protonmail.com>
Signed-off-by: Alfred Wingate <parona@protonmail.com>
This means you don't have to use explicit compiler_args for cpp.preprocess which then means the meson requirement can be lowered. Signed-off-by: Alfred Wingate <parona@protonmail.com>
parona-source
force-pushed
the
llvm-build
branch
from
c768be5 to
0a23a14
Compare
|
Used the c preprocessor to preprocess the version file instead. Hopefully that is more appealing. |
|
@parona-source there is any change that needs to be done (besides the conflicts), because I been using this patch (with the following use flags (X dbus mangoapp mangohudctl wayland), without any problems. The verion I using is this one: @AdelKS, might be a good ideia to solve the conflicts any apply this patch to mangohud ebuilds? Edit: Here is a rebased patch if anyone needs: rebased-1513-24d88a.patch |
4 participants
Keeping it a draft as this would in its current state break builds for Windows. I'm unsure what is the intent with mangohud_opengl and what symbols should be visible in it for Windows.With LLVM toolchain I mean clang as the compiler, lld is the linker and libcxx as the c++ stdlib.
I didn't touch mangohud_shim, but its visible symbols are shown as well. It could also be explicitly specified like mangohud_opengl.
Current HEAD built with GCC/ld.bfd