[GHC] #15470: Record projections with ambiguous types

[GHC]

#15470: Record projections with ambiguous types
-------------------------------------+-------------------------------------
        Reporter:  sweirich          |                Owner:  (none)
            Type:  feature request   |               Status:  new
        Priority:  normal            |            Milestone:  8.8.1
       Component:  Compiler          |              Version:  8.4.3
      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 simonpj):

 I thought that visible type applications would work too, so I tried it
 (diff below).

 Alas it turns out that we give record selectors ''nested'' forall types.
 For example from
 {{{
 data T m = MkT { fld :: forall a. m a -> m a }
 }}}
 we get the selector:
 {{{
 fld :: forall m. T m -> (forall a. m a -> m a)
 }}}
 So we can't use scoped type variables from the signature to instantiate in
 the RHS, ike this
 {{{
 fld :: forall m. T m -> (forall a. m a -> m a)
 fld (MkT x) = x @ a  -- No
 }}}
 Bother.

 Maybe we can change this assumption, and put all the foralls (and
 constraints) at the top of record selectors?  I'm not sure of the
 consequences of doing so.  The choice is made in the definition of
 `sel_ty` in `TcTyDecls.mkOneRecordSelector`.

 It affects user code.  For example, do we write `fld @IO r @Int` or `fld
 @IO @Int r`?  So it's be a breaking change.

 I'm not sure I see any other solution, alas.

 Richard, Mr VTA?

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15470#comment:2>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler