https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103699
--- Comment #8 from Petr <kobalicek.petr at gmail dot com> ---
My only problem is that A returns a different value compared to B, C, and D:
uint32_t test_u32_a() {
char array[16] {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
writeU64be(array + 6, 0xAABBCCDDEEFF1213);
return readU32be(array + 7);
}
uint32_t test_u32_b() {
static char array[16] {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
writeU64be(array + 6, 0xAABBCCDDEEFF1213);
return readU32be(array + 7);
}
uint32_t test_u32_c() {
thread_local char array[16] {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
writeU64be(array + 6, 0xAABBCCDDEEFF1213);
return readU32be(array + 7);
}
uint32_t test_u32_d(char array[16]) {
writeU64be(array + 6, 0xAABBCCDDEEFF1213);
return readU32be(array + 7);
}
And when you compile this, you would actually see that ALL functions evaluate
to a constant (because it's known what the output will be), but only in A case
the constant is different (of course because B, C, D have side effects):
test_u32_a():
mov eax, 117967114
ret
test_u32_b():
movabs rax, 1374442237689904042
mov QWORD PTR test_u32_b()::array[rip+6], rax
mov eax, -1144201746
ret
test_u32_c():
movabs rax, 1374442237689904042
mov QWORD PTR fs:test_u32_c()::array@tpoff+6, rax
mov eax, -1144201746
ret
test_u32_d(char*):
movabs rax, 1374442237689904042
mov QWORD PTR [rdi+6], rax
mov eax, -1144201746
ret
So yeah, we can talk about breaking strict aliasing here, but it's just
inconsistent. I would just expect all functions return the same value.