Re:What is Hindley-Milner and why is it cool? (2008)
by dtech on Sunday August 20, 2017 @03:45PM
(#15059935)
> I wonder why you think so? The inferred types in a language like Haskell or OCaml are almost always exactly what the programmer would have written
I have no experience with ML-using languages but in Scala, which has inferior but still useful type inference, it is recommended to add explicit types to public functions/methods because otherwise, implementation details or errors might alter the signature.
Take for example:
data Tree = Node Leaf Leaf | Leaf Integral
makeEmptyTree x => Leaf x
I assume `Leaf` will be inferred as a type, while the programmer might have meant `Tree` to be the public API. (or vice-versa)
Re:What is Hindley-Milner and why is it cool? (2008)
by catnaroek on Sunday August 20, 2017 @03:59PM
(#15059988)
In ML, constructors aren't subtypes of a sum type. And that's a good thing.
Re:What is Hindley-Milner and why is it cool? (2008)
by tel on Monday August 21, 2017 @07:00AM
(#15063578)
This is largely to do with subtyping and it's well known that systems with user-defined nominal subtyping don't have good inference. Scala's inference is really in the middle: squarely better than C/Java and much worse than Haskell/OCaml.
I wonder why you think so? The inferred types in a language like Haskell or OCaml are almost always exactly what the programmer would have written. So long as a language has principle type inference, the worst case is that the inferred type is a bit more general than you expected.
> Also, whole-program type inference and separate compilation inherently conflict.
There is no conflict here, actually. All type variables in definitions exported from a module are generalizable. As such, no uses in other modules will further refine those types.