optimizer bug, latest version of egcs ( 0 && 0 => 1 )

[Richard Earnshaw]

law@cygnus.com said:
> Thanks for tracking this down!  I've got a half-dozen or so bugs on my
> queue which look like they may be the same underlying problem.  The
> basic idea behind your fix seems sound to me.

> Here's the patch I'm playing with -- I'll run it through the rest of
> my "gcc losts cast" bugs tomorrow. 

Functionally, I think our two patches are identical, since type is only 
set at the one place (near the top of the while loop).  So I guess it's 
down to which is the better of the two places to put this.

I think that after one of the patches is applied, the setting of orig_type 
in the NOP_EXPR case is now redundant (orig_type cannot now be null unless 
type is also null -- in which case there would be an even bigger problem 
:-)