https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122462
Bug ID: 122462
Summary: -fwrapv-pointer suppresses optimization but not
-Waddress warning in pointer arithmetic with NULL
Product: gcc
Version: 14.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: jiangxuezhi2 at huawei dot com
Target Milestone: ---
I encountered an inconsistency between the behavior of the -fwrapv-pointer
option and the -Waddress warning. The -fwrapv-pointer option changes the
generated assembly code (as expected), but the -Waddress warning is still
issued based on standard C semantics, which seems incorrect when the option is
active.
gcc -O3 -Wall test.c -o test
~~~c
extern char *pNext;
int kkk() {
return !!(pNext + 1 == (void *)0);
}
~~~
Assembly Output:
kkk:
xor eax, eax
ret
output:
<source>: In function 'kkk':
<source>:4:25: warning: the comparison will always evaluate as 'false' for the
pointer operand in 'pNext + 1' must not be NULL [-Waddress]
4 | return !!(pNext + 1 == (void *)0);
| ^~
however,we add option -fwrapv-pointer
Assembly Output:
kkk:
xor eax, eax
cmp QWORD PTR pNext[rip], -1
sete al
ret
The generated assembly correctly performs a comparison (checking if pNext is
-1), indicating that the pointer wrap-around semantics are respected due to
-fwrapv-pointer.But the output is same as before, like:
<source>: In function 'kkk':
<source>:4:25: warning: the comparison will always evaluate as 'false' for the
pointer operand in 'pNext + 1' must not be NULL [-Waddress]
4 | return !!(pNext + 1 == (void *)0);
Additional Information：
This warning occurs due to enhanced -Waddress checks introduced in GCC 12 (as
mentioned in the discussion of PR 102103
