| > ...which would be very useful, but would probably have unpleasant
| > consequences for type inference...
|
| To my mind, this is not a credible objection. The horse has
| already bolted; there's no point in trying to shut the stable
| door. The existing post-Hindley-Milner aspects of Haskell's
| type system are too useful for type inference to be worthy of
| such veneration.
I agree with this. The goal of 100% type inference is a red herring.
This point of view is articulated at some length in a draft paper
Mark Shields and I have written, on scoped type variables
http://research.microsoft.com/~simonpj/papers/scoped-tyvars
Type inference is still terrifically useful to "fill in the gaps"
between
programmer-supplied type signatures. The trouble with full type
lambdas is that they make unification essentially impossible,
and that has a huge impact on how many type annotations you need.
So I'm completely in agreement with moving towards allowing the
programmer to supply more explicit type information; but I have not
yet seen a proposal that supports anything like full type lambdas
in a system that looks anything like an extension of Haskell,
and that does not require quite burdensome type annotations.
But I may well simply be ignorant.
Simon
_______________________________________________
Haskell mailing list
[EMAIL PROTECTED]
http://www.haskell.org/mailman/listinfo/haskell
