https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110052
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |WAITING
Last reconfirmed| |2023-05-31
Ever confirmed|0 |1
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
/* This is the original code, the local variable is superfluous */
int size = foo->size;
if (!m)
m = myrealloc(m, 256, &size);
foo->size = size;
I'm not sure we are allowed to change the &size argument to &foo->size
since it's observable whether the argument is equal to &foo->size or
not since 'foo' is global and myrealloc could have access to it
and also clobber it. Consider
myrealloc(..., int *size)
{
*size = 4;
foo->size = 0;
}
if 'foo' were global. Can you make the testcase avoid these
considerations? I think we have no pass doing this kind of transform.
Maybe sth along
int size1, size2;
foo (&size1);
size2 = size1;
bar (&size2);
of course since size1 escapes to foo() bar() might do the same as myrealloc
above and it would break passing &size1 to bar().
