[GHC] #15639: Surprising failure combining QuantifiedConstraints with Coercible

[GHC]

#15639: Surprising failure combining QuantifiedConstraints with Coercible
-------------------------------------+-------------------------------------
        Reporter:  dfeuer            |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.6.1
       Component:  Compiler (Type    |              Version:  8.5
  checker)                           |             Keywords:
      Resolution:                    |  QuantifiedConstraints
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:                    |
-------------------------------------+-------------------------------------

Comment (by dfeuer):

 Replying to [comment:5 simonpj]:
 > Can you explain more about this example?  I think that it is important
 that
 >
 > * You give a role signature for `Yeah`.  (What is the inferred
 signature?)

 The inferred signature is `type role Yeah phantom`

 > * You export the type constructor for `Yeah` but not the data
 constructor.

 Yes. The export isn't really as relevant as the import it enables; when a
 newtype constructor is in scope that changes the coercion axioms the type
 checker uses for the type.

 > But I'm a bit lost about what you expect to happen.  Can you amplify?
 Why do you think it should typecheck? It'd probably help to give the
 declaration for `Coercion` too, for those of us who do not use it daily.

 The declaration of `Coercion` is

 {{{#!hs
 data Coercion a b where
   Coercion :: Coercible a b => Coercion a b
 }}}

 The situation is

 {{{#!hs
 yeahCoercible :: ((forall a b. Coercible (Yeah a) (Yeah b)) => r) -> r
 yeahCoercible r = r

 -- Fishy2.hs

 yeah :: Coercion [Yeah a] [Yeah b]
 yeah = yeahCoercible Coercion
 }}}

 We call `yeahCoercible` with `Coercion`. For this to typecheck, we need

 {{{#!hs
 Coercion :: (forall a b. Coercible (Yeah a) (Yeah b)) => Coercion [Yeah a]
 [Yeah b]
 }}}

 Put another way, we need `Coercible (Yeah a) (Yeah b)` to imply `Coercible
 [Yeah a] [Yeah b]`. Under normal circumstances, that would follow from the
 fact that `[]`'s parameter has a representational role. But the constraint
 solver isn't figuring that out.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15639#comment:7>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler