[GHC] #7939: RHS of associated type not kind-checked

[GHC]

#7939: RHS of associated type not kind-checked
------------------------------------------+---------------------------------
  Reporter:  goldfire                     |          Owner:  goldfire        
      Type:  bug                          |         Status:  new             
  Priority:  normal                       |      Milestone:                  
 Component:  Compiler                     |        Version:  7.7             
Resolution:                               |       Keywords:                  
        Os:  Unknown/Multiple             |   Architecture:  Unknown/Multiple
   Failure:  GHC accepts invalid program  |     Difficulty:  Unknown         
  Testcase:  ghci/scripts/T7939           |      Blockedby:                  
  Blocking:                               |        Related:                  
------------------------------------------+---------------------------------

Comment(by goldfire):

 As I was looking at all of this, I wasn't sure what to do about type
 variables like `b` above. My decision to leave it to be polymorphic was
 mostly from a lack of a clearly-articulated reason why some variables are
 polymorphic and some aren't.

 However, I don't think Simon's "scope" rule really does it.

 Currently, variables are kind-polymorphic-by-default in the following
 places:
  * Class definitions
  * Datatype definitions without a full kind signature
  * Associated type declarations
  * Vanilla type synonyms

 Variables are defaulted to `*` in the following places:
  * Top-level family declarations
  * Datatype definitions with a full kind signature

 And, with the kind inference for closed type families that Simon and I
 discussed, we have a third case:
  * Type variables in closed type families are kind-polymorphic, but
 unconstrained type variables are defaulted to `*`.

 The "scope" rule doesn't quite describe this menagerie. Even excepting
 closed type families, the scope rule fails for datatype definitions. Type
 variables in a (GADT-style) datatype definition have no scope, but they
 get their constraints other ways.

 It's easy enough in the implementation to make the type variable `b` from
 above default to `*`, and that might be sensible, but it does make for a
 rather complicated rule to state.

-- 
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/7939#comment:13>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler