http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45967
Richard Guenther <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |ASSIGNED
Last reconfirmed| |2010.10.11 15:03:59
AssignedTo|unassigned at gcc dot |rguenth at gcc dot gnu.org
|gnu.org |
Target Milestone|--- |4.5.2
Summary|gcc-4.5.x optimizes code |[4.5/4.6 Regression]
|with side-effects away |gcc-4.5.x optimizes code
| |with side-effects away
Ever Confirmed|0 |1
--- Comment #1 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-10-11
15:03:59 UTC ---
I think I have seen a dup of this bug. The bug is in points-to analysis
which doesn't grok your weird way of initializing a pointer temporary.
So what goes on is that you have
int *pointer;
initialize-pointer-by-pieces
--*pointer;
and points-to analysis doesn't get that initialize-pointer-by-pieces
alters the points-to set of pointer and thus we end up with pointer
pointing to nothing (which makes --*pointer removed by dead code
elimination).
The bug is also latent in 4.4 but there for a pointer that points to
nothing we simply assume it points to anything (which hides bugs).
PR45623 was similar, but I am thinking of another dup.
I will give this mess a little more thought.
