Created attachment 50697 [details]
Minimal test case code.

Bug occurs in GCC 11.1.0, but none of the 10.x releases.

It appears that GCC is now inferring a size of 0 when doing reinterpret_cast<void*>(ConstantInteger); when doing std::memcpy/std::memset to/from the result pointers, bogus warnings are emitted about reading/writing to regions of zero size.

My target is an embedded system with a fixed memory layout; I have been using constexpr uintptr_t/size_ts's to describe the memory regions, and correspondingly calls to set or copy memory regions are now emitting bogus warnings.

I have made an example minimal test case here (also attached): https://godbolt.org/z/WPaGY8eaz

Relevant errors (compiling with -O -Werror):

void StringopOverread() {
    // error: 'void* memset(void*, int, size_t)' writing 16 bytes into a region of size 0 overflows the destination [-Werror=stringop-overflow=]
    std::memset(reinterpret_cast<void *>(0xCAFEBABE), 0xCC, 0x10);
}

void StringopOverflow2(const void *src) {
    // error: 'void* memcpy(void*, const void*, size_t)' writing 16 bytes into a region of size 0 overflows the destination [-Werror=stringop-overflow=]
    std::memcpy(reinterpret_cast<void *>(0xCAFEBABE), src, 0x10);
}

void StringopOverread(void *dst) {
    // error: 'void* memcpy(void*, const void*, size_t)' reading 16 bytes from a region of size 0 [-Werror=stringop-overread]
    std::memcpy(dst, reinterpret_cast<void *>(0xCAFEBABE), 0x10);
}