http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58455
--- Comment #2 from Roland Dreier <rbd at debian dot org> ---
Arg, I don't seem to be able to avoid breaking my test cases while minimizing
them. So for my first test case here the issue seems to be that -Og is the
only level that *correctly* warns, which is surprising and I guess is a bug in
it's own right!
Sorry for the confusion.
However the test case I was originally working with and then broke is the
following:
$ cat y.cpp
struct B
{
int offset;
bool is_loaded(unsigned char *p1)
{
if (offset) {
*p1 = offset;
return true;
}
return false;
}
};
unsigned char match(struct B ref)
{
do {
unsigned char ref_offset;
if (!ref.is_loaded(&ref_offset) || false)
continue;
return ref_offset;
} while (false);
return 0;
}
$ for o in Og O0 O1 O2 O3; do echo == $o ==; gcc -Wall -$o -Werror -c ~/y.cpp;
done
== Og ==
/home/roland/y.cpp: In function ‘unsigned char match(B)’:
/home/roland/y.cpp:17:31: error: ‘ref_offset’ may be used uninitialized in this
function [-Werror=maybe-uninitialized]
unsigned char ref_offset;
^
cc1plus: all warnings being treated as errors
== O0 ==
== O1 ==
== O2 ==
== O3 ==
in that case is_loaded returns true when it sets the pointer; if it returns
false we continue, fall out of the loop and don't touch the pointer value. So
I'm pretty sure this case is correctly classified as spurious.
