[GHC] #10856: Record update doesn't emit new constraints

[GHC]

#10856: Record update doesn't emit new constraints
-------------------------------------+-------------------------------------
              Reporter:  goldfire    |             Owner:
                  Type:  bug         |            Status:  new
              Priority:  normal      |         Milestone:
             Component:  Compiler    |           Version:  7.10.2
              Keywords:              |  Operating System:  Unknown/Multiple
          Architecture:              |   Type of failure:  None/Unknown
  Unknown/Multiple                   |
             Test Case:              |        Blocked By:
              Blocking:              |   Related Tickets:
Differential Revisions:              |
-------------------------------------+-------------------------------------
 I believe this should compile:

 {{{
 {-# LANGUAGE ExistentialQuantification #-}

 data Rec a b = Show a => Mk { a :: a, b :: b }

 update :: Show c => c -> Rec a b -> Rec c b
 update c r = r { a = c }
 }}}

 But it fails with

 {{{
 /Users/rae/temp/Bug.hs:6:14:
     Couldn't match type ‘a’ with ‘c’
       ‘a’ is a rigid type variable bound by
           the type signature for update :: Show c => c -> Rec a b -> Rec c
 b
           at /Users/rae/temp/Bug.hs:5:11
       ‘c’ is a rigid type variable bound by
           the type signature for update :: Show c => c -> Rec a b -> Rec c
 b
           at /Users/rae/temp/Bug.hs:5:11
     Expected type: Rec c b
       Actual type: Rec a b
     Relevant bindings include
       r :: Rec a b (bound at /Users/rae/temp/Bug.hs:6:10)
       c :: c (bound at /Users/rae/temp/Bug.hs:6:8)
       update :: c -> Rec a b -> Rec c b
         (bound at /Users/rae/temp/Bug.hs:6:1)
     In the expression: r
     In the expression: r {a = c}
 }}}

 I believe the problem has to do with the `fixed_tvs`, the tyvars that are
 most certainly shared between the initial and result types. The
 `getFixedTyVars` function always includes tyvars mentioned in constraints.
 But this is unnecessary.

 Once the `fixed_tvs` are cleaned up, then we'll also have to make sure to
 instantiate the constraints necessary to prove that the new constraints
 are satisfied. We should be careful '''not''' to emit any unchanged
 constraints, because these are provided by the GADT pattern-match that the
 desugarer produces.

 There will probably have to be an update to the `RecordUpd` constructor of
 `HsExpr` to have a place to put the new dictionaries.

 NB: This is very related to implementation trouble in Phab:D1152, and I
 imagine that patch will evolve to fix this infelicity.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10856>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler