[GHC] #12845: -XPartialTypeSignatures results in unbound variables

16 Nov 2016

      #12845: -XPartialTypeSignatures results in unbound variables
-------------------------------------+-------------------------------------
           Reporter:  crockeea       |             Owner:
               Type:  bug            |            Status:  new
           Priority:  normal         |         Milestone:
          Component:  Compiler       |           Version:  8.0.1
           Keywords:                 |  Operating System:  Unknown/Multiple
       Architecture:                 |   Type of failure:  GHC rejects
  Unknown/Multiple                   |  valid program
          Test Case:                 |        Blocked By:
           Blocking:                 |   Related Tickets:
Differential Rev(s):                 |         Wiki Page:
-------------------------------------+-------------------------------------
 The following code fails to compile, but I think it should.

 {{{
 {-# LANGUAGE DataKinds             #-}
 {-# LANGUAGE PartialTypeSignatures #-}
 {-# LANGUAGE ScopedTypeVariables   #-}
 {-# LANGUAGE TypeFamilies          #-}
 {-# LANGUAGE TypeOperators         #-}

 import Data.Proxy

 data Foo (m :: Bool)

 type family Head (xs :: [(Bool, Bool)]) where Head (x ': xs) = x

 type family Bar (x :: Bool) (y :: Bool) :: Bool

 -- to trigger the bug, r and r' cannot *both* appear on the RHS
 broken :: forall r r' rngs . ('(r,r') ~ Head rngs, Bar r r' ~ 'True, _)
   => Foo r -> Proxy rngs -> ()
 broken x _ = let y = requireBar x :: Foo r' in ()

 requireBar :: (Bar m m' ~ 'True) => Foo m -> Foo m'
 requireBar = undefined

 }}}

 The error is:

 {{{
...
ghci UnboundVar.hs
 GHCi, version 8.0.1: http://www.haskell.org/ghc/  :? for help
 [1 of 1] Compiling Main             ( UnboundVar.hs, interpreted )
UnboundVar.hs:18:1: error:
     • Could not deduce: Bar r r'0 ~ 'True
       from the context: ('(r, r') ~ Head rngs,
                          Bar r r' ~ 'True,
                          Bar r r' ~ 'True)
         bound by the inferred type for ‘broken’:
                    ('(r, r') ~ Head rngs, Bar r r' ~ 'True, Bar r r' ~
 'True) =>
                    Foo r -> Proxy rngs -> ()
         at UnboundVar.hs:18:1-49
       The type variable ‘r'0’ is ambiguous
     • When checking that the inferred type
         broken :: forall (r :: Bool) (r' :: Bool) (rngs :: [(Bool,
 Bool)]).
                   Bar r r' ~ 'True =>
                   Foo r -> Proxy rngs -> ()
       is as general as its (partial) signature
         broken :: forall (r :: Bool) (r' :: Bool) (rngs :: [(Bool,
 Bool)]).
                   ('(r, r') ~ Head rngs, Bar r r' ~ 'True, Bar r r' ~
 'True) =>
                   Foo r -> Proxy rngs -> ()
 }}}

 If I remove the partial type signature (i.e., remove the `_` constraint
 from `broken`), then GHC happily accepts the code. I'm not sure why adding
 a `_` introduces ambiguity; it seems GHC is forgetting that `r'` is bound
 by the `forall` (and that `-XScopedTypeVariables` is enabled).

--
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/12845
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler