Subject: Re: type "synonyms"?
From: Craig Chambers (chambers@cs.washington.edu)
Date: Tue Feb 06 2001 - 18:14:27 PST
Keunwoo Lee wrote:
>
> OK, so I have a Cecil type signature like this:
>
> method lower_op(node:UnaryOpNode, op@:ConvertUnaryOp,
> analysis:LoweringAnalysis,
> graph:AnalysisGraph[IREdge,TopoAnalysisPriority]
> ):TransformAnalysisAction[GenericLoweringAnalysisInfo,
> AnalysisGraph[IREdge,
> TopoAnalysisPriority]];
>
> Needless to say, this is a monster. However, this slight shortcut:
>
> method lower_op(node:UnaryOpNode, op@:ConvertUnaryOp,
> analysis:LoweringAnalysis,
> graph:`G <= AnalysisGraph[IREdge, TopoAnalysisPriority]
> ):TransformAnalysisAction[GenericLoweringAnalysisInfo, G]];
>
> doesn't typecheck. What, technically, is the difference? Is my
> declaration wrong? I would like to save myself at least that small amount
> of visual clutter.
G is bound to some unknown subtype of AG[IRE, TAP], and is not equal to it. So
the return type is different, and TAA hasn't been declared to be covariant in
its second parameter, so the second return type isn't a subtype of the first.
>
> Oh, as a usability issue, it would be nice if we could declare a type
> synonym and save it for use across many declarations (perhaps in a
> module), so I could say:
>
> type MyGraph renames AnalysisGraph[IREdge,TopoAnalysisPriority];
>
> Even a simple lexical substitution would be fine for this. A regular
> "subtypes" declaration don't really do the right thing, and we don't have
> circular subtyping relations...
>
> ~k.lee
Yes, type synonyms would be a very nice thing, which I've long wanted. There
was an early design reason not to have them (one ought to define explicit
subtypes for complicated types, since there's some semantic entity being
manipulated that deserves its own name), but that was before we got practical
experience with how long these sorts of types would get and how intricate the
various subtyping relationships would get (preventing easy substitution of a
subtype declaration).
-- Craig
This archive was generated by hypermail 2b25 : Tue Feb 06 2001 - 18:14:33 PST