Explicit namespaces doesn't enforce namespaces

[mpickering]

It seems to me that the type keyword should permanently enforce that a RdrName lives in a particular namespace. Currently it only acts as a syntactic marker so that type constructors which looks like variables can be included in export lists.

The following example is accepted by GHC. As A is parsed as a type constructor but later put into the data constructor namespace.

{-# LANGUAGE ExplicitNamespaces #-}
{-# LANGUAGE TypeFamilies #-}
module M ( T (type A) ) where

data T = A

class Foo a where
  type A a
  foo :: a -> Int

The fix is to track uses of ExplicitNamespaces in the AST and then use this information in the renamer.

[mpickering]

Simon says that we should just reject the use of explicit namespace qualifiers in the subcomponents of the export item in the parser.

[ruperthorlick]

Replying to mpickering:

Simon says that we should just reject the use of explicit namespace qualifiers in the subcomponents of the export item in the parser.

This would work and be a fairly easy change, however I stumled across the following in DynFlags.hs:

-- We turn this on so that we can export associated type
-- type synonyms in subordinates (e.g. MyClass(type AssocType))
, (LangExt.TypeFamilies,     turnOn, LangExt.ExplicitNamespaces)

So for the above they might want to export Foo (type A)

Does this mean we need something more complicated?