Filed upstream: https://github.com/libfuse/libfuse/pull/545

If they merge, I will apply the change in Fedora.

Jeff, there are issues here.

Upstream uses Travis to build test, and when Travis builds it through Ubuntu Bionic with GCC 7.5.0, the meson check reports:

Message: Compiler supports symver attribute

However, the build has issues:

../../../home/travis/build/libfuse/libfuse/lib/fuse_loop_mt.c:309:1: warning: ‘symver’ attribute directive ignored [-Wattributes]

example/hello_ll@exe/hello_ll.c.o: In function `main':
582/tmp/libfuse-build-VwANJo/build-gcc/../../../home/travis/build/libfuse/libfuse/example/hello_ll.c:213: undefined reference to `fuse_session_loop_mt'
583collect2: error: ld returned 1 exit status

The original test code in meson.build just throws a warning on symver:

origtest.c:1:1: warning: ‘symver’ attribute directive ignored [-Wattributes]
 __attribute__((symver ("get@@TEST_0"))) int get_4() {
 ^~~~~~~~~~~~~

But because it compiles...

I think this might be a better test:

int main(void) {
return 0;
}


clang and gcc have __has_attribute, so they'll be okay here. another compiler which doesn't have __has_attribute is probably safe to use the fallback asm logic (and if not, it was broken before).

Thoughts?

I've amended this to be:


#if defined __has_attribute
# if !__has_attribute (symver)
# error symver attribute not supported
# endif
#else
#error __has_attribute not defined, assume we do not have symver
#endif

int main(void) {
return 0;
}


Thanks. I ran into something similar with the autoconf test. GCC tries to be "helpful" and only warn on the unrecognized attribute :( So yea, the amended version is better.

I'm absolutely comfortable waiting on upstream for this -- there's no pressing need to get LTO enabled for fuse3 in F33 -- I'm looking at this more in the rawhide flowing into ELN and F34 contexts.