Re: Fix tree-optimization/17273

[Diego Novillo <dnovillo at redhat dot com>]

On Fri, 2004-09-17 at 14:47, Andrew Pinski wrote:

> I don't like we are working around a bug because it is considered too
> late/invasive for 4.0, remember we are early into stage3 which means
> that we can have invasive stuff iff it fixes bugs.
> 
I don't like it either, but c'est la vie.

The work I'm doing is still largely experimental.  It lengthens the
optimization pipeline quite significantly and has us calling
execute_ssa_ccp() and execute_copy_prop() very often.  They become
global entry points for any pass that needs to clean things up.  So,
they have to be _fast_.

It will also impact DOM very significantly, mostly reducing it to just
the jump threader.  So, I need to coordinate with Jeff to make sure I'm
not weakening the jump threader in any way.

My time right now will be mostly tied up to existing bugs for 4.0.  So,
I don't know how much time I can dedicate to this cleanup effort.  I
could probably create a branch and then based on what we get, decided
whether to move safe cleanups into mainline.

In the meantime, this fix is the bit that was missing.  If we ever hit
the ICE condition I just introduced, we can decide whether we should
scan the block from the start looking for the statement so that we can
feed a proper statement pointer to fold_stmt.


Diego.