https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110267
Bug ID: 110267
Summary: Bogus warning "function may return address of local
variable"
Product: gcc
Version: 12.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: nullplanwichmann at web dot de
Target Milestone: ---
Host: x86_64-pc-linux-gnu
Target: x86_64-pc-linux-gnu
Build: x86_64-pc-linux-gnu
Minimal test case:
extern char *strdup(const char *);
extern int fill(char *);
char *do_something(char *buf)
{
char tmp[10];
if (!buf)
buf = tmp;
int rv = fill(buf);
if (rv < 0) return 0;
if (buf[0] != '/')
return 0;
return buf == tmp? strdup(buf) : buf;
}
Compile with -O2 -Wall. All GCC versions I have tried (10.2.1-6 from an old
Debian installation, 12.2.0 from current Debian stable, and a development
snapshot I compiled from git this morning) warn:
tmp.c: In function ‘do_something’:
tmp.c:13:1: warning: function may return address of local variable
[-Wreturn-local-addr]
13 | }
| ^
tmp.c:5:8: note: declared here
5 | char tmp[10];
The warning is of course wrong: Whenever buf == tmp, the function returns the
strdup() of tmp, not tmp itself. I was resigned to assume that gcc cannot
identify that this is the case, but I noticed that if the if-selection in lines
10-11 is removed, then the warning disappears. The warning only occurs when
optimizing at level s, 2, or 3.
Additionally, the old gcc version 10.2.1-6 mangled the warning output, it would
write out:
cc1: warning: function may return address of local variable
[-Wreturn-local-addr]
tmp.c:5:8: note: declared here
5 | char tmp[10];
As if that was some internal warning. That seems to be fixed in the newer
versions, though.