[tree-ssa] estimate_num_insns_1 fix

[Richard Henderson]

On Mon, Dec 15, 2003 at 11:37:19PM +0100, Jan Hubicka wrote:
> > On Sun, Dec 14, 2003 at 12:36:37AM +0100, Jan Hubicka wrote:
> > > ! 	if (TREE_TYPE (x) == VOIDmode)
> > > ! 	  abort ();
> > 
> > What are you actually trying to test here?
> 
> I was running into problem with MODIFY_EXPR being VOIDmode despte it's
> operand being non-VOIDmode.  This is fixed now, but it seemed like good
> idea to prevent this from happening (such expressions accounts as 0
> then).

Well at the moment you're comparing tree == machine_mode.

It accidentally works because VOIDmode == 0; a deficiency
in our front end that we don't warn for this, IMO.  I think
we should be doing something more like C++ and suppress the
pointer vs integer comparison warning only for literal 0.


r~