https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96564
Bug ID: 96564
Summary: New maybe use of uninitialized variable warning since
GCC >10
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: stefansf at linux dot ibm.com
Target Milestone: ---
Consider the following MWE:
extern void* malloc (long unsigned int);
void fun (unsigned *x) {
unsigned *a = malloc (*x);
if (a == 0)
return;
if (a != x) // (A)
*a = *x;
*x = *a;
}
If compiled with GCC 10, then no warning is emitted.
If compiled with GCC HEAD (currently 84005b8abf9), then the following warning
is emitted:
gcc -W -c -O2 mwe.c
mwe.c: In function 'fun':
mwe.c:8:8: warning: '*(a)' may be used uninitialized [-Wmaybe-uninitialized]
8 | *x = *a;
| ^~
Rational why this example is strictly speaking fine:
1) Assume x is a valid pointer. Then malloc will either return a nullpointer
and we return, or a pointer to a fresh object which address is different from
any other existing object. Thus (A) always evaluates to true which means *a is
initialized.
2) Assume x is an invalid pointer. Then dereferencing x prior to the call to
malloc already results in UB.
Thus the only case in which condition (A) may evaluate to false is when x is an
invalid pointer (e.g. previously malloc'd and then free'd such that a further
call to malloc returns the very same address rendering (A) false) results in
UB.
Since this is a maybe warning I'm wondering whether this is considered a bug or
is acceptable. Any thoughts?
