Re: [tree-ssa] verify cond_expr condition type

[Jan Hubicka <jh at suse dot cz>]

> On Sat, 17 Jan 2004 23:53:32 +0100, Jan Hubicka <hubicka@ucw.cz> wrote:
> 
> >> As discussed re one of law's dominator changes.  Fortunately my fears
> >> were unfounded and bootstrap and check passes with this installed.
> >
> > Nice, but would be perhaps better to go for my verify_gimple_grammar
> > patch?  That should be place for such a checks to go
> > (it don't do this check in current incarnation)
> 
> It makes sense for me to keep the logic about what is or is not fully
> simplified all in the same place, i.e. the gimplifier.  I've had a patch to
> add a speculative mode to the gimplifier in the works for a while now;
> sorry I haven't finished it yet.  I'll definitely finish it up this week.
OK, it makes sense.  On the other hand, it seemed to me that it would
actually make sense to duplicate the logic just in order to have some
double checking that there is no stubble bug in gimplifier itself and
that it may perhaps make sense to move the comments downwards to the
verifiers to make it ieasier to keep these in sync too, but this is not
critical.  The duplication in between CFG builder and verify_flow_info
in the past did made relatively good job catching bugs on the both
sides...

Will be possible to then call the gimplifier from verify_stmt and get
some sane error messages out the output?

Plus the stmt checking if done in between each pass has serious overhead
(about 10%) that needs to be taken into account.

Said that, I don't really worry about what implementation will go into
official tree as long as we will have some :)
Honza
> 
> Jason