Subject: Re: desugaring issue
From: Vassily Litvinov (vass@cs.washington.edu)
Date: Sun Jan 28 2001 - 18:48:51 PST
I agree that constraints should be "inherited" automatically. There are
some implementation issues here. Semantically one needs to make sure that
constraints on Analysis[I,G], in turn, are not "inherited" from
constraints on ScopeAnalysis[G], as that would create a circular
dependence. (Giving semantics to a circular dependence is complicated and
I'd like to avoid it.)
Todd and Jonathan, thanks for the feedback!
Vass
On Sun, 28 Jan 2001, Jonathan Aldrich wrote:
> > forall `G <= AnalysisGraph[E, P],
> > `E <= IREdge, `P <= AnalysisPriority[P]:
> > object ScopeAnalysis[G]
> > isa Analysis[GenericScopeAnalysisInfo,G];
> >
> > Semantically it is essential that all of `G, `E, and `P, together
> > with all the constraints, should be part of both the "object" and
> > "isa" desugared declarations.
>
> Hmm...that makes sense. However, if there is already a constraint on the
> second parameter of Analysis, it seems like that should automatically be
> "inherited" by ScopeAnalysis. That is,
>
>
> -- somewhere else
> forall `G <= AnalysisGraph[E, P],
> `E <= IREdge, `P <= AnalysisPriority[P], `I <= Something :
> object Analysis[I,G];
>
>
> -- inherits all the constraints on G, E, and P from Analysis
> object ScopeAnalysis[G] isa Analysis[GenericScopeAnalysisInfo,G];
>
>
> To me, that would be the intuitive semantics, and one which naturally lets
> you avoid excessive explicit constraints. Since I'm not a types expert,
> I'm probably missing some subtlety, but that's what seems natural to me as
> a user.
>
> Jonathan :-)
>
>
This archive was generated by hypermail 2b25 : Sun Jan 28 2001 - 18:48:56 PST