https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109446
--- Comment #4 from Xi Ruoyao <xry111 at gcc dot gnu.org> ---
(In reply to Martin Liška from comment #3)
> The problem here is that we normally preserve memcpy calls and then
> __interceptor_memcpy is used from the run-time library. However, in this
> case the second argument of memcpy is a known constant and we convert it to:
> MEM <unsigned char[64]> [(char * {ref-all})_7] = MEM <unsigned char[64]>
> [(char * {ref-all})&txt];
>
> for such an assignment we only check the beginning and the end of the chunk
> and we miss the overflow.
It seems Clang disables this optimization and convert memcpy to __asan_memcpy
calls if -fsanitize=address used:
https://godbolt.org/z/dcfadoMYY
