https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105810
--- Comment #8 from cqwrteur <unlvsur at live dot com> --- (In reply to Jonathan Wakely from comment #7) > Right, but that function doesn't even exist today, so there's nothing to > merge today. This change adds new symbols (one for every assertion in the > library) which must be assembled and then merged by the linker. > > I'm just not sure why you're pointing this out in comment 4 like it's an > advantage of the change. Of course there's only one copy of the function, > otherwise it would be an ODR violation, or a linker error. But it's still > one more function for every assertion in the library. the thing is that. if I ca(In reply to Jonathan Wakely from comment #7) > Right, but that function doesn't even exist today, so there's nothing to > merge today. This change adds new symbols (one for every assertion in the > library) which must be assembled and then merged by the linker. > > I'm just not sure why you're pointing this out in comment 4 like it's an > advantage of the change. Of course there's only one copy of the function, > otherwise it would be an ODR violation, or a linker error. But it's still > one more function for every assertion in the library. The thing is that: https://godbolt.org/z/85M1PE9Te The compiler has to generate subq $8, %rsp leaq .LC0(%rip), %rcx leaq .LC1(%rip), %rdx movl $278, %esi leaq .LC2(%rip), %rdi call std::__glibcxx_assert_fail(char const*, int, char const*, char const*)@PLT for every function when we use them. while for the 2nd case https://godbolt.org/z/TjP7YxYcT will only generate once for the assertion part.
