[GHC] #10607: Auto derive from top to bottom

[GHC]

#10607: Auto derive from top to bottom
-------------------------------------+-------------------------------------
        Reporter:  songzh            |                Owner:
            Type:  feature request   |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  7.11
      Resolution:                    |             Keywords:  deriving,
                                     |  typeclass, auto
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
-------------------------------------+-------------------------------------

Comment (by songzh):

 Replying to [comment:5 goldfire]:
 > Yes -- TH struggles a bit with instance lookup and such. Implementing
 new querying features shouldn't be hard though, if you can suggest an API.
 >
 > Why do you need fancy querying capabilities for your use case? Where
 does `reifyInstances` fail for you? As for type synonyms, you can check
 out the `th-expand-syns` package. `th-desugar` also does (limited) type
 family expansion (among other, larger features).

 Replying to [comment:5 goldfire]:
 > Yes -- TH struggles a bit with instance lookup and such. Implementing
 new querying features shouldn't be hard though, if you can suggest an API.
 >
 > Why do you need fancy querying capabilities for your use case? Where
 does `reifyInstances` fail for you? As for type synonyms, you can check
 out the `th-expand-syns` package. `th-desugar` also does (limited) type
 family expansion (among other, larger features).

 Hi Richard
 I tried to implement a prototype of this top down deriving scheme with
 standalone deriving. The project is on
 https://github.com/HaskellZhangSong/TopdownDerive

 However, I have 3 problems about it.

 1. It forces me to use `KindSignatures`, `ConstraintKind` and other
 extension which should not be needed.(please see `TopDownDeriveTest.hs`
 file)

 2. For type synonym, I want to do it without using TypeSynonymInstance,
 but I am not sure how to get the arity of a type constructor. For example:
 `type T a = (a,a,a,a,a)` I need to generate (Eq a , Eq b, Eq c ,Eq d, Eq
 e) => (a,b,c,d,e).

 3. When I derive some instances that based on Generic class like Binary or
 FromJSON it gives me a type error:

 {{{
     Could not deduce (G.Generic a)
       arising from a use of `binary-0.7.6.1:Data.Binary.Class.$gdmget'
     from the context (B.Binary a)
       bound by the instance declaration
 }}}

 while I think `deriving instance (Binary a ,Binary b) => Binary (A a b)`
 should work fine. Why do I have to write `deriving instance (B.Binary a,
 G.Generic a) => (B.Binary (B a))`? And could you give me some suggestions
 to solve it.

 Thanks

 Song

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10607#comment:6>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler