[GHC] #11008: Difficulties around inferring exotic contexts (was: GND with Type Families)

[GHC]

#11008: Difficulties around inferring exotic contexts
-------------------------------------+-------------------------------------
        Reporter:  crockeea          |                Owner:
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  7.10.2
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by goldfire):

 This is by design. GHC refuses to infer so-called exotic contexts,
 believing that sufficiently exotic contexts should be written by the user.
 This behavior is independent of whether GND or other `deriving` features
 are used. Here is an extreme example:

 {{{
 data X a b = MkX (a -> b) deriving Eq
 }}}

 This fails, complaining about a missing `Eq (a -> b)` instance. (My choice
 of `data` vs. `newtype` is utterly irrelevant here.) But I can write this:

 {{{
 deriving instance Eq (a -> b) => Eq (X a b)
 }}}

 Should GHC infer the context automatically here? I think not -- it masks a
 deeper problem.

 Now, returning to the original post: should GHC infer that context? It
 looks just like the example I just gave, where the context requires an
 `Eq` constraint on some other type constructor.

 It all boils down to yet another knob we can turn in the internals of GHC.
 It's a very easy knob to turn, with no interactions throughout the code.
 See the relevant function
 [https://github.com/ghc/ghc/blob/8f5ad1a009eddd05447ff8057792b4d03983cd35/compiler/typecheck/TcValidity.hs#L949
 here], with a Note directly above.

 If you can suggest a better setting for the knob, go right ahead. :)

 Regardless, we should document this in the manual somewhere, as it is all
 user-facing.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11008#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler