Details
-
Bug
-
Status: Closed (View Workflow)
-
Blocker
-
Resolution: Fixed
-
11.4.4
-
LibreELEC with gcc15-20241117
Description
Potential fix available at https://github.com/MariaDB/server/pull/3686
Attachments
Issue Links
- is duplicated by
-
-
- Closed
-
Activity
I believe I was able to ascertain why the void check is failing. This from CMakeConfigureLog.yaml
kind: "try_compile-v1"
|
backtrace:
|
- "/var/media/DATA/home-rudi/LibreELEC.tv/build.LibreELEC-Generic.x86_64-13.0-devel/toolchain/share/cmake-3.31/Modules/Internal/CheckSourceCompiles.cmake:108 (try_compile)"
|
- "/var/media/DATA/home-rudi/LibreELEC.tv/build.LibreELEC-Generic.x86_64-13.0-devel/toolchain/share/cmake-3.31/Modules/CheckCSourceCompiles.cmake:58 (cmake_check_source_compiles)"
|
- "configure.cmake:678 (CHECK_C_SOURCE_COMPILES)"
|
- "CMakeLists.txt:396 (INCLUDE)"
|
checks:
|
- "Performing Test SIGNAL_RETURN_TYPE_IS_VOID"
|
directories:
|
source: "/var/media/DATA/home-rudi/LibreELEC.tv/build.LibreELEC-Generic.x86_64-13.0-devel/build/mariadb-11.4.4/.x86_64-libreelec-linux-gnu/CMakeFiles/CMakeScratch/TryCompile-RsDXgI"
|
binary: "/var/media/DATA/home-rudi/LibreELEC.tv/build.LibreELEC-Generic.x86_64-13.0-devel/build/mariadb-11.4.4/.x86_64-libreelec-linux-gnu/CMakeFiles/CMakeScratch/TryCompile-RsDXgI"
|
cmakeVariables:
|
CMAKE_C_FLAGS: "-march=x86-64-v3 -Wall -pipe -O2 -fomit-frame-pointer -DNDEBUG"
|
CMAKE_MODULE_PATH: "/var/media/DATA/home-rudi/LibreELEC.tv/build.LibreELEC-Generic.x86_64-13.0-devel/build/mariadb-11.4.4/cmake;/var/media/DATA/home-rudi/LibreELEC.tv/build.LibreELEC-Generic.x86_64-13.0-devel/build/mariadb-11.4.4/cmake/Internal/CPack"
|
buildResult:
|
variable: "SIGNAL_RETURN_TYPE_IS_VOID"
|
cached: true
|
stdout: |
|
Change Dir: '/var/media/DATA/home-rudi/LibreELEC.tv/build.LibreELEC-Generic.x86_64-13.0-devel/build/mariadb-11.4.4/.x86_64-libreelec-linux-gnu/CMakeFiles/CMakeScratch/TryCompile-RsDXgI'
|
|
Run Build Command(s): /var/media/DATA/home-rudi/LibreELEC.tv/build.LibreELEC-Generic.x86_64-13.0-devel/toolchain/bin/ninja -v cmTC_7156c
|
[1/2] /var/media/DATA/home-rudi/LibreELEC.tv/build.LibreELEC-Generic.x86_64-13.0-devel/toolchain/bin/x86_64-libreelec-linux-gnu-gcc -DPACKAGE=test -DSIGNAL_RETURN_TYPE_IS_VOID -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE=1 -D_LARGEFILE_SOURCE -D_LARGE_FILES -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -march=x86-64-v3 -Wall -pipe -O2 -fomit-frame-pointer -DNDEBUG -o CMakeFiles/cmTC_7156c.dir/src.c.o -c /var/media/DATA/home-rudi/LibreELEC.tv/build.LibreELEC-Generic.x86_64-13.0-devel/build/mariadb-11.4.4/.x86_64-libreelec-linux-gnu/CMakeFiles/CMakeScratch/TryCompile-RsDXgI/src.c
|
FAILED: CMakeFiles/cmTC_7156c.dir/src.c.o
|
/var/media/DATA/home-rudi/LibreELEC.tv/build.LibreELEC-Generic.x86_64-13.0-devel/toolchain/bin/x86_64-libreelec-linux-gnu-gcc -DPACKAGE=test -DSIGNAL_RETURN_TYPE_IS_VOID -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE=1 -D_LARGEFILE_SOURCE -D_LARGE_FILES -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -march=x86-64-v3 -Wall -pipe -O2 -fomit-frame-pointer -DNDEBUG -o CMakeFiles/cmTC_7156c.dir/src.c.o -c /var/media/DATA/home-rudi/LibreELEC.tv/build.LibreELEC-Generic.x86_64-13.0-devel/build/mariadb-11.4.4/.x86_64-libreelec-linux-gnu/CMakeFiles/CMakeScratch/TryCompile-RsDXgI/src.c
|
src.c:9:8: error: conflicting types for 'signal'; have 'void (*(void))(void)'
|
9 | void (*signal ()) ();
|
| ^~~~~~
|
In file included from src.c:2:
|
/var/media/DATA/home-rudi/LibreELEC.tv/build.LibreELEC-Generic.x86_64-13.0-devel/toolchain/x86_64-libreelec-linux-gnu/sysroot/usr/include/signal.h:88:23: note: previous declaration of 'signal' with type 'void (*(int, void (*)(int)))(int)'
|
88 | extern __sighandler_t signal (int __sig, __sighandler_t __handler)
|
| ^~~~~~
|
ninja: build stopped: subcommand failed.
|
|
exitCode: 1
|
As a quick hack test, the following allowed a correct compile of mariadb with gcc-15
--- mariadb-11.4.4/configure.cmake 2024-10-30 14:38:12.000000000 +0000
|
+++ build/mariadb-11.4.4/configure.cmake 2024-12-10 14:58:29.378634217 +0000
|
@@ -672,27 +672,8 @@
|
SET(STACK_DIRECTION -1 CACHE INTERNAL "Stack grows direction")
|
ENDIF()
|
|
-#
|
-# Check return type of signal handlers
|
-#
|
-CHECK_C_SOURCE_COMPILES("
|
-#include <signal.h>
|
-#ifdef signal
|
-# undef signal
|
-#endif
|
-#ifdef __cplusplus
|
-extern \"C\" void (*signal (int, void (*)(int)))(int);
|
-#else
|
-void (*signal ()) ();
|
-#endif
|
-int main(int ac, char **av) {}
|
-" SIGNAL_RETURN_TYPE_IS_VOID)
|
-IF(SIGNAL_RETURN_TYPE_IS_VOID)
|
SET(RETSIGTYPE void)
|
SET(VOID_SIGHANDLER 1)
|
-ELSE(SIGNAL_RETURN_TYPE_IS_VOID)
|
- SET(RETSIGTYPE int)
|
-ENDIF(SIGNAL_RETURN_TYPE_IS_VOID)
|
|
|
CHECK_INCLUDE_FILES("time.h;sys/time.h" TIME_WITH_SYS_TIME)
|
According to https://gcc.gnu.org/gcc-15/changes.html
Note: GCC 15 has not been released yet, so this document is a work-in-progress.
and I think it's too early to think about gcc-15 portability
gcc-14 works fine. Let's wait for gcc-15 release. I couldn't find this change in behavior in the documentation or release notes, may be it's some debugging extra strictness that will go away in a release.
Indeed, even gcc-13 fails this with -std=c2x, while -std=c18 works.
So the actual reason is the new stricter C standard. And supposedly MDEV-35606 will need a separate fix too.
could you try this fix? it worked for me, but I didn't try gcc-15
there's a configure check for the correct signal handler return type. it doesn't work in your environment and we need to understand why.
Here's how you can find it out:
cmake -USIGNAL_RETURN_TYPE_IS_VOID --debug-trycompilethe last line might look differently depending on your cmake version and your make tool