[Bug middle-end/88518] Function call defeats -Wuninitialized
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88518
Martin Sebor changed:
What|Removed |Added
Status|UNCONFIRMED |RESOLVED
Last reconfirmed||2021-4-6
Resolution|--- |DUPLICATE
CC||msebor at gcc dot gnu.org
--- Comment #6 from Martin Sebor ---
Confirmed as a duplicate of pr60488 (which itself is likely a duplicate of some
older bug). A variable whose address escapes is not considered by the warning.
It sees the following IL:
;; Function f (f, funcdef_no=0, decl_uid=1947, cgraph_uid=1, symbol_order=0)
void f ()
{
long unsigned int i;
long unsigned int i.0_1;
long unsigned int _2;
[local count: 1073741824]:
# .MEM_4 = VDEF <.MEM_3(D)>
h ();
# VUSE <.MEM_4>
i.0_1 = i;<<< i loaded from memory and not considered
_2 = i.0_1 + 1; <<< missing -Wuninitialized
# .MEM_5 = VDEF <.MEM_4>
i = _2;
# .MEM_6 = VDEF <.MEM_5>
g ();
# .MEM_7 = VDEF <.MEM_6>
i ={v} {CLOBBER};
# VUSE <.MEM_7>
return;
}
The pointless load persists until the assembly:
f:
.LFB0:
.cfi_startproc
subq$24, %rsp
.cfi_def_cfa_offset 32
callh
leaq8(%rsp), %rdi <<< i loaded from memory
addq$1, 8(%rsp) <<< i++
callg
addq$24, %rsp
.cfi_def_cfa_offset 8
ret
.cfi_endproc
*** This bug has been marked as a duplicate of bug 60488 ***
[Bug middle-end/88518] Function call defeats -Wuninitialized
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88518
--- Comment #5 from Eric Gallager ---
(In reply to Matthew Wilcox from comment #2)
> Thanks! What I actually want to do is annotate g() to the effect that it
> reads the pointed-to variable before it writes it. IOW, I want to write
> something like:
>
> void g(unsigned long __attribute__((read_before_write)) *p);
> void h(void);
>
> void f(void)
> {
> unsigned long i;
>
> h();
> g();
> }
>
> and have gcc emit a warning that i is used uninitialised. I tried adding i
> = i prior to the call of g() through macro trickery, but was surprised that
> it didn't trigger, and simplified down to this test-case.
>
> Any chance I could get that attribute or something like it?
The object access attributes that Martin Sebor added recently might work
somewhat along these lines, but I haven't actually tested yet...
[Bug middle-end/88518] Function call defeats -Wuninitialized
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88518 --- Comment #4 from Eric Gallager --- Note that if h() has __attribute__((pure)) or __attribute__((const)) on it, you then get the -Wuninitialized warning as expected. Of course, then you also get a warning from -Wattributes about pure or const being used on a function returning void, too...
[Bug middle-end/88518] Function call defeats -Wuninitialized
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88518 Richard Biener changed: What|Removed |Added Severity|normal |enhancement --- Comment #3 from Richard Biener --- Not so easy.
[Bug middle-end/88518] Function call defeats -Wuninitialized
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88518
--- Comment #2 from Matthew Wilcox ---
Thanks! What I actually want to do is annotate g() to the effect that it reads
the pointed-to variable before it writes it. IOW, I want to write something
like:
void g(unsigned long __attribute__((read_before_write)) *p);
void h(void);
void f(void)
{
unsigned long i;
h();
g();
}
and have gcc emit a warning that i is used uninitialised. I tried adding i = i
prior to the call of g() through macro trickery, but was surprised that it
didn't trigger, and simplified down to this test-case.
Any chance I could get that attribute or something like it?
[Bug middle-end/88518] Function call defeats -Wuninitialized
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88518 Andrew Pinski changed: What|Removed |Added Keywords||diagnostic --- Comment #1 from Andrew Pinski --- I think there is a dup of this bug already and one which descibes what need to be done. Basically escape analysis is not flow sensative. That means GCC is currently assuming that i escapes at the very first function call, h, rather later on when g is called.
