https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108770
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
Last reconfirmed|2023-02-13 00:00:00 |2025-08-21
Status|UNCONFIRMED |WAITING
--- Comment #12 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Teylu from comment #11)
> (In reply to Andrew Pinski from comment #10)
> > Ok, with const maybe the compiler could see that but then it needs to record
> > that somehow but there is no way currently.
> >
>
> Sure, and, to me, it isn't a problem that it isn't currently
> fixable/resolvable, I just didn't think it was correct to close this case as
> "FIXED" when actually it isn't.
>
> In fact, you could argue that, when you turn on `diagnostics-show-context`,
> gcc is a "worse" because it "doubling down" on a false positive ...
But it is only a false positive because you the programmer has knowledge of
arr.
>
> Is it obvious to anyone what changed between 11.x and 12.1 to make this
> appear?
Just better optimization and jump threading.
If you had:
```
extern void put(int i);
int check_idx(int i) {
if (i > 1)
put(i);
return i;
}
extern const char *arr[2];
void init() {
int i = 0;
while (arr[check_idx(i)] != 0) {
if (arr[check_idx(i)]) {}
i++;
}
}
```
There would be no way for GCC to know that arr has a nullptr.
Do you have the original source where this is extracted from instead of a
simpler example? Because I suspect this simplified example might be reduced way
too much in the first place.
