[GHC] #10343: Make Typeable track kind information better

[GHC]

#10343: Make Typeable track kind information better
-------------------------------------+-------------------------------------
        Reporter:  oerjan            |                   Owner:  goldfire
            Type:  feature request   |                  Status:  new
        Priority:  normal            |               Milestone:  7.12.1
       Component:  Compiler          |                 Version:  7.10.1
      Resolution:                    |                Keywords:
Operating System:  Unknown/Multiple  |            Architecture:
 Type of failure:  None/Unknown      |  Unknown/Multiple
      Blocked By:                    |               Test Case:  typeOf ::
 Related Tickets:  #9858             |  Typeable (a::k) => Proxy a ->
                                     |  TypeRep
                                     |                Blocking:
                                     |  Differential Revisions:
-------------------------------------+-------------------------------------

Comment (by simonpj):

 Can I make two pleas?

  1. Could we cast this discussion in the language of type-indexed type
 representations `TTypeRep`, in [wiki:Typeable]? A `TTypeRep a` is very
 like `Typeable a`, but has the very great merit of being an ordinary data
 value.  `Typeable a` is just the implicit version of it.  You can get from
 one to the other via `tTypeRep`:
 {{{
 class Typeable a where
   tTypeRep :: TTypeRep a
 }}}
   By using an ordinary data value we get away from talking about classes
 or superclasses or solvers or whatnot.  Once we know, with forensic
 precision, what `TTypeable a` should do, then we can think about how the
 solver might generate `Typeable a` dictionaries, which are simply
 containers for a `TTypeable a` value.

  2. Can we keep `TTypeRep` abstract, so instead of talking about what it
 "contains" we speak entirely in terms of what functions are available over
 it?  `decomposeFun` is one such function.  Now, for `kindRep` I think you
 are asking for a function
 {{{
 kindRep :: forall (a::k). TTypeRep a -> TTypeRep k
 }}}
   And indeed it seems plausible that we should provide such a function,
 although I have seen no examples whatsoever of why it might be useful.
 (Producing a `TypeRep` only to print it does not count; there are easier
 ways of producing that sequence of characters.)

 I'm afraid I found the summary entirely opaque "The type inference and
 runtime representation should support constructing `Typeable (T :: k)`,
 even when ``k contains kind variables that are only mentioned in other
 `Typeable` constraints not necessarily involving the type constructor `T`.
 This will allow simple composition with polymorphic type constructors like
 `Proxy` and `Typeable` to work more naturally, but also more complicated
 things."  What do you mean by "simple composition"?  What do you mean by
 "more complicated things"?

 Don't answer directly: rather, just give a sequence of examples that you
 think should work.  That will ''show'' me what "simple composition" means,
 rather than ''telling'' me!

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10343#comment:13>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler