#7939: RHS of associated type not kind-checked ------------------------------------------+--------------------------------- Reporter: goldfire | Owner: 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): I tend to agree that generalizing the result kind (without an explicit annotation) is a step too far. In general, I'm happy with the current story that type family arguments and results default to {{{*}}}. Because instances aren't provided at the same place as the family declaration, any sort of kind inference amounts to some level of spooky action at a distance. Note that the kind inference for datatypes looks only at the declaration of the constructors, not any uses of the type. Similarly, the kind inference for classes only looks at the methods/other members of the class, not any instances. For consistency with those cases, I would expect for type families either (everything defaults to {{{*}}}) or (everything defaults to fresh kind variables). The former gives better errors in the common case, so I favor that one. -- Ticket URL: http://hackage.haskell.org/trac/ghc/ticket/7939#comment:7 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler