#16146: Trivial partial type signature kills type inference in the presence of GADTs -------------------------------------+------------------------------------- Reporter: goldfire | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.6.3 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by goldfire): Another interesting example has cropped up. Re-consider {{{#!hs wurble :: _ wurble x = x }}} This is accepted, because the type `_` allows for implicit generalization. But what about {{{#!hs wurble2 :: forall. _ wurble2 x = x }}} To my surprise, this is accepted too, with `wurble2 :: forall a. a -> a`. Yet, I would have expected my `forall.` to say that there are no quantified variables in the type of `wurble2`. So, maybe another design is this: `foo :: _` can be generalized. (This is at odds with the fact that `foo` cannot have constraints.) But if there is an explicit `forall` in type of `foo`, we appeal to the forall-or-nothing rule and say that the variables quantified in the `forall` are precisely the quantified variables; no more can be inferred. Then, we can suppress quantification in `forall. _`. One advantage here is that this new design is backward compatible (though I'm not terribly worried about that here). -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/16146#comment:8 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler