Ticket #1823 (closed bug: fixed)
GADTs and scoped type variables don't work right
| Reported by: | simonpj | Owned by: | chak |
|---|---|---|---|
| Priority: | normal | Milestone: | 6.10 branch |
| Component: | Compiler (Type checker) | Version: | 6.8.1 |
| Keywords: | Cc: | ganesh@… | |
| Operating System: | Unknown/Multiple | Architecture: | Unknown/Multiple |
| Type of failure: | Difficulty: | Unknown | |
| Test Case: | gadt/scoped.hs | Blocked By: | |
| Blocking: | Related Tickets: |
Description
In this program, which is also in the testsuite as gadt/scoped.hs, all three definitions of g should work
{-# OPTIONS_GHC -XGADTs -XScopedTypeVariables -XPatternSignatures #-}
module GADT where
data C x y where
C :: a -> C a a
data D x y where
D :: C b c -> D a c
------- All these should be ok
-- Rejected!
g1 :: forall x y . C x y -> ()
-- C (..) :: C x y
-- Inside match on C, x=y
g1 (C (p :: y)) = ()
-- OK!
g2 :: forall x y . C x y -> ()
-- C (..) :: C x y
-- Inside match on C, x=y
g2 (C (p :: x)) = ()
-- Rejected!
g3 :: forall x y . D x y -> ()
-- D (..) :: D x y
-- C (..) :: C sk y
-- sk = y
-- p :: sk
g3 (D (C (p :: y))) = ()
But they don't. The reason is that the "refinement" is applied to the environment only after checking an entire pattern; but the refinement is needed during checking the pattern if scoped type variables are to work right.
I don't propose to fix this, because it'll come out in the wash when we get rid of type refinements in favour of equality constraints.
Change History
Note: See
TracTickets for help on using
tickets.
