https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70301
Marc Glisse <glisse at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2016-03-19
Ever confirmed|0 |1
Severity|normal |enhancement
--- Comment #1 from Marc Glisse <glisse at gcc dot gnu.org> ---
It would make the testcase a bit more convincing if you actually used those
pointers. Just taking the address is dead code. For instance, if you try to
read *p, you will get:
D.2276 = f ();
D.2276 ={v} {CLOBBER};
_6 = MEM[(int *)&D.2276];
which we should diagnose in the middle-end (there are a couple PRs about it).
If you just call an unknown function on the pointer:
D.2264 = f ();
D.2264 ={v} {CLOBBER};
fun (&D.2264.a);
(or if you try to write to *p instead of reading) then it looks like the
middle-end cannot do anything, as it lost the information that the memory
location is gone, not just its content, and we could have valid code (running a
destructor, then placement new) that looks the same in gimple. So it does look
like the C++ FE is the only place where we could warn.
