Re: [PATCH][RFC] A GIMPLE and "GIMPLE type system" verifier

[Richard Guenther <rguenther at suse dot de>]

On Thu, 31 May 2007, Diego Novillo wrote:

> On 5/31/07 10:21 AM, Richard Guenther wrote:
> 
> > Comments?
> 
> Well, without having a properly defined type system for GIMPLE, I don't
> know how we can implement a verifier for it.  Seems to me like we should
> do that first.

Sure ;)  We have it implicitly defined by means of
tree_ssa_useless_type_conversion_1 (which is frontend specific) and
what tree.def says about valid operand types.  This is what I currently
verify.

> I guess that we could use some of this verification code, but if you are
> implementing against a moving target, it's going to be mighty hard to
> keep it working.

That's true, which is why I concentrate on what comes out of the
gimplifier which is a good measurement for what we get out of the
frontends.  As far as I understood we are going to keep GENERIC as it
is now and eventually go from it via the gimplifier to whatever
tuples and type system we want.

Richard.