------- Comment #3 from rguenth at gcc dot gnu dot org 2008-05-13 11:41 ------- The patch isn't really correct, but the problem is that the fact that <result> is marked addressable is causing flow-insensitive analysis to say that the incoming pointers can alias it.
Also incoming pointers don't get useful constraints for PTA - they should get &NONLOCAL (err, ok - we had that once ;)). Or we can simply prune the points-to set after the fact. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36201
