[Bug pch/13689] GGC PCH breaks aliasing

["dberlin at dberlin dot org" <gcc-bugzilla at gcc dot gnu dot org>]

------- Additional Comments From dberlin at dberlin dot org  2004-01-15 02:16 -------
Subject: Re:  GGC PCH breaks aliasing


On Jan 14, 2004, at 9:09 PM, pinskia at gcc dot gnu dot org wrote:

>
> ------- Additional Comments From pinskia at gcc dot gnu dot org  
> 2004-01-15 02:09 -------
> no it is not as it still causes wrong-code in certain cases.
>

As stated, doing that would just hide the bug, not fix it.
The bug is actually caused by PCH, *not* Jan's changes.
We will still create alias sets the same way, we are just storing them 
different, so that the queries just don't trip anything in GCC (because 
of lack of asserts that alias set values we expect to be in the splay 
tree are in the splay tree) that lets you know there is a bug.
It's still there, and still potential for wrong code exists.

Reverting Jan's change is the absolute *wrong* thing to do. It removes 
a speedup, AND papers over the bug, instead of actually fixing the 
problem.
and it doesn't fix *anything*.

The bug is PCH not storing the actual alias sets, when  the types it 
*does* store are using them.
So that we end up with trees referencing alias set 7 or 9 that when the 
PCH is reloaded, do not exist.
With splay trees, we would look these up, see nothing, and assume 
everything was fine anyway, when it wasn't.
Since Jan uses varrays, when we try to access the out-of-bounds 
element, we get a check error.

In short, this is not a regression caused by Jan's patch, it's a latent 
bug caused by PCH that Jan's patch simply uncovered.




-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13689