|Version 16 (modified by dreixel, 5 years ago) (diff)|
The new Generic Deriving mechanism (ongoing work)
GHC includes a new (in 2010) mechanism to let you write generic functions. It is described in A generic deriving mechanism for Haskell, by Magalhães, Dijkstra, Jeuring and Löh. This page sketches the specifics of the implementation; we assume you have read the paper.
This mechanism replaces the previous generic classes implementation. The code is in a branch of GHC; you can get it with darcs get http://darcs.haskell.org/ghc-generic-15Feb11/ghc.
- TcDeriv.tcDeriving generates an InstInfo for each data type that fulfills the isRep0 predicate. This InstInfo is the Representable0 instance for that type, allowing it to be handled generically (by kind-* generic functions).
- The representation types and core functionality of the library live on GHC.Generics (on the ghc-prim package).
- Many names have been added as known in prelude/PrelNames
- Most of the code generation is handled by types/Generics
What already works
- Representable0 instances are automatically generated when -XGenerics is enabled.
- Remove all of the old deriving mechanism stuff
- Properly deal with fixity and isTuple information for constructors
- Generate Representable1 instances
- What about base types like , Maybe, etc.?
- Generic instances
- Add deriving as a keyword. This replaces the DERIVABLE pragma from the UHC implementation, and is attached to a default method on a class declaration.
- Change the Class definition to allow for generic defaults (in addition to standard defaults).
- Generate default instances for representable types which derive generic classes.
- For temporary testing, a file test/Main.hs is available with sample datatypes.
- Currently, in TcDeriv.genGenericRepBind we generate instances using mkLocalInstance. Is this right, or should we use mkImportedInstance instead? SLPJ: mkLocalInstance: it's as if the instance declaration was in this module, right?