https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63518
Martin Sebor <msebor at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed|2019-02-03 00:00:00 |2021-3-29
Known to fail| |10.2.0, 11.0, 4.8.4, 4.9.4,
| |5.5.0, 6.4.0, 7.2.0, 8.3.0,
| |9.1.0
--- Comment #4 from Martin Sebor <msebor at gcc dot gnu.org> ---
This is a similar problem to (although not quite the same as) the one we have
been discussing in pr60488. I have changed the test case to C to shorten the
IL. Here's the C test case and the IL from a GCC instrumented to print it just
as the pass runs. At each point of a variable's use, the warning looks for
prior statements that might set its value. If it finds one, it doesn't
trigger. In foo() the first such prior statement is the call to
setTimeout(&t). In bar(), there is no such prior statement, and so there the
warning does trigger.
I don't see how to fix this except by running the warning in the front end or
just before Gimplification when all the arguments to the wait() call statement
are still in the IL. Once the code is Gimplified the arguments are already
broken out of the call. This isn't an option today for all the reasons that
have been discussed over the years but might perhaps be doable if/when the
__builtin_warning() idea comes to fruition. With it in place, the warning
could be scheduled to be issued early on, false positives weeded out by running
the optimizers as usual, and then issued if the __builtin_warning() call was
still left in the IL. This would have to be done under the control of the same
predicate analysis as the one in tree-ssa-uninit.c today (which is why I'm
factoring it out into a standalone reusable component).
$ cat x.c && gcc -S -Wall x.c
void wait (int, _Bool);
void wait2 (_Bool, int);
_Bool setTimeout (int*);
void foo (void)
{
int t;
wait (t, setTimeout (&t));
}
void bar (void)
{
int t;
wait2 (setTimeout (&t), t);
}
void foo ()
{
int t;
_Bool _1;
int _2;
int t.0_3;
<bb 2> :
# .MEM_5 = VDEF <.MEM_4(D)>
_1 = setTimeout (&t); <<< address of t escapes first
_2 = (int) _1;
# VUSE <.MEM_5>
t.0_3 = t; <<< missing warning
# .MEM_6 = VDEF <.MEM_5>
wait (t.0_3, _2);
# .MEM_7 = VDEF <.MEM_6>
t ={v} {CLOBBER};
# VUSE <.MEM_7>
return;
}
void bar ()
{
int t;
int t.1_1;
_Bool _2;
int _3;
<bb 2> :
# VUSE <.MEM_4(D)>
t.1_1 = t; <<< -Wuninitialized
# .MEM_5 = VDEF <.MEM_4(D)>
_2 = setTimeout (&t); <<< address of t escapes second
_3 = (int) _2;
# .MEM_6 = VDEF <.MEM_5>
wait2 (_3, t.1_1);
# .MEM_7 = VDEF <.MEM_6>
t ={v} {CLOBBER};
# VUSE <.MEM_7>
return;
}
_1 = setTimeout (&t); <<< address of t escapes
_2 = (int) _1;
t.0_3 = t; <<< missing warning
wait (t.0_3, _2);
t ={v} {CLOBBER};
return;
}
void bar ()
{
int t;
int t.1_1;
_Bool _2;
int _3;
<bb 2> :
t.1_1 = t; <<< -Wuninitialized
_2 = setTimeout (&t); <<< address of t escapes
_3 = (int) _2;
wait2 (_3, t.1_1);
t ={v} {CLOBBER};
return;
}
x.c: In function ‘bar’:
x.c:15:3: warning: ‘t’ is used uninitialized [-Wuninitialized]
15 | wait2 (setTimeout (&t), t);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
x.c:14:7: note: ‘t’ declared here
14 | int t;
| ^
