https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116723
--- Comment #2 from Ye Xiong <bouncy12578 at gmail dot com> ---
(In reply to Andrew Pinski from comment #1)
> (p.f0 != (-18))
>
> For uint16_t, this is always true and is an always infinite loop. While for
> int16_t it is not and the loop is removed because you supplied -ffinite-loop.
>
> -ffinite-loop does NOT remove always infinite loops by design and is what is
> required by the C++ standard even.
>
> So invalid by design.
However, if swap the two types in the union (in case1.c) like:
case3.c
union U0 {
int32_t f0;
int16_t f1;
} g;
int func_1(union U0 p)
{
p.f1 = 0;
for (p.f0 = 1; (p.f0 != (-18)); p.f0 = (int32_t) (3 - p.f0));
return 1;
}
Although p.f0 is a signed int, it will outputs "1" (with -O3 and
-ffinite-loops).
Even weirder, only change the type of p1 from int16_t to int32_t the basis of
case3.c:
case4.c
union U0 {
int32_t f0;
int32_t f1;
} g;
int func_1(union U0 p)
{
p.f1 = 0;
for (p.f0 = 1; (p.f0 != (-18)); p.f0 = (int32_t) (3 - p.f0));
return 1;
}
It will be seen as an infinite loop again.
