Version 49 (modified by chak, 11 years ago) (diff)


Type Functions: Implementation Status

Back to TypeFunctions.


  • Complete TcTyClsDecls.tcIdxTyInstDecl1 for TyData: (1) probably need to generate a new name; (2) produce a TyCon, which requires to extend the result type, and (3) produce an equality axiom. The produced TyCon need to be entered into the global environment, including all its implicitTyThings, by the caller, namely TcInstDecls.tcInstDecls1.

Parsing and Renaming

Todo (high-level):

  1. Make the kind specification optional in family declarations. (How to do the representation in the data/newtype case is not entirely clear, as we use the presence of the kind signature at the moment to identify family declarations.)
  2. Parse and rename equality constraints in signatures.
  3. Defaults for associated type synonyms. (Having both a kind signature and vanilla synonym is problematic as in RnNames.getLocalDeclBinders its hard to see that not both of them are defining declarations, which leads to a multiple declarations error. Defaults are quite different from vanilla synonyms anyway, as they usually have tyvars on their rhs that do not occur on the lhs.)


  • Parsing and renaming of kind signatures (toplevel and in classes).
  • Parsing and renaming of indexed types declarations (toplevel and in classes).
  • Using new syntax with family and instance on top level.

Type Checking

Todo (low-level):

  • RHS of a type instance must be a tau type.
  • In an AT definition, no argument variable may be repeated.
  • Check that the arguments of AT instances coincide with the respective instance arguments of their class. This might be a bit more tricky if we want to allow that they can vary syntactically before expansion of type synonyms.
  • Check that each class instance has a definition for every AT and conversely that that all defined associated types are, in fact, part of the class - with the exception of associated synonyms with a default definition. (Do this in the type checker - GHC does the corresponding checks for methods in the type checker, too.)
  • Families declared as an AT, may not receive toplevel type instances.
  • Check that patterns of type indexes don't contain type functions.
  • For each case scrutinising an associated data type, check that all constructors have been defined in a single instance. (Maybe we can just extend the existing check that ensures that case expressions don't mix constructors of different data types.)
  • Construct InstInfo for type equation in tcIdxTyInstDecl1.

Todo (high-level):

  1. Type checking in the presence of associated data types.
  2. Type checking in the presence of associated synonym defaults. (Default AT synonyms are only allowed for ATs defined in the same class.)
  3. Type checking of type functions (and hence, associated type synonyms); forget about iso for the moment.
  4. Type check functional dependencies as type functions.


  • Kind and type checking of kind signatures.
  • Kind and type checking of instance declarations of indexed types.


Todo (low-level):

  • Derivings on an associated data type declaration need to be inherited by all definitions of that data type in instances.

Todo (high-level):

  1. Desugar indexed data types.
  2. Extend interface files.
    • How do we exactly want to represent ATs in interface files?
      • SPJ pointed out that instances are maintained in InstEnv.InstEnv with different values for the home packages and others. The definitions of ATs may have to be maintained in a similar way, as they are also incrementally collected during compiling a program.
      • IfaceInst contains the instance declaration information for interfaces.
    • Export and import lists: The name lists that may appear at class imports and exports can now also contain type names, which is tricky as data type names can carry a list of data constructors.
    • At the moment, we add as the parent name of the data constructors of associated data types defined in instances, the new name for the data type constructor, which is different from that of the data type constructor in the class (also their source representation is the same). We may need to fix that during renaming. (We can't easily fix it in getLocalDeclBinders, where the names of the data constructors are made, as we don't have the means to get at the right class at that point.)
  3. Desugar type functions and equality constraints.


  • Representation of kind signatures as TyCon.TyCons.
  • Extension of Class.Class by associated TyCons.



  • Convert AT.hs to tests in the testsuite.