https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116494
--- Comment #1 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to Архипов Дмитрий from comment #0)
> When compiled with -std=c++20 -O3 -Wall -Werror this code:
>
> #include <string>
>
> bool foo(std::string s1)
> {
> s1.replace(1, 1, s1.data(), s1.size());
> return s1.size();
> }
>
> fails with a warning about incorrect use of __builtin_memcpy. As far as I
> can see, inside _M_replace the call to _M_disjunct paradoxically determines
> that s1.data() is not contained in s1's buffer. Because of that _M_replace
> then uses the wrong branch and uses memcpy, when it shouldn't have.
No, it doesn't. It's just a warning during compilation. The branch is not taken
under those conditions.
