[GHC] #10447: DeriveFoldable rejects instances with constraints in last argument of data type

[GHC]

#10447: DeriveFoldable rejects instances with constraints in last argument of data
type
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                   Owner:
            Type:  bug               |                  Status:  new
        Priority:  normal            |               Milestone:
       Component:  Compiler          |                 Version:  7.10.1
      Resolution:                    |                Keywords:
Operating System:  Unknown/Multiple  |            Architecture:
 Type of failure:  GHC rejects       |  Unknown/Multiple
  valid program                      |               Test Case:
      Blocked By:                    |                Blocking:
 Related Tickets:  #8678             |  Differential Revisions:
-------------------------------------+-------------------------------------

Comment (by RyanGlScott):

 > You expect that the {{{Foldable}}} instance for {{{data T2 a = (a ~ Int)
 => MkT2 Int}}} give you the {{{Int}}} because there happens to be an
 {{{Int}}} equality constraint?

 That is what I am expecting, yes, especially since {{{T1}}}, {{{T2}}}, and
 {{{T3}}} are equivalent representations of the same thing.

 > That seems pretty unintuitive to me. What about e.g. {{{data E = E
 Int}}}, and {{{data A a = A E}}} vs. {{{data A a = (a ~ Int) => A E}}},
 and then inlining E?

 I'm not sure how inlining would cause a derived {{{Foldable}}} instance
 for {{{data A a = (a ~ Int) => A E}}} to fail. Does inlining occur before
 {{{DeriveFoldable}}} generates its code? If so, I don't see how GHC could
 confuse {{{E}}} for {{{Int}}}.

 > derived {{{Foldable}}} instances in the presence of equality constraints
 need some more justification.

 I find myself wanting to derive {{{Foldable}}} instances for deeply
 embedded DSLs that are represented as GADTs. Here's a simple example from
 [https://github.com/ku-fpg/hermit-
 ghci/blob/8a7556bc53a2e79fb46a807875e913d753b5da37/src/HERMIT/API.hs#L57-58
 hermit-ghci]:

 {{{#!hs
 data ShellEffect :: * -> * where
   ShellEffect :: Value -> ShellEffect ()
 }}}

 (which could equivalently be represented as {{{ShellEffect :: a ~ () =>
 Value -> ShellEffect a}}})

 This isn't a {{{Functor}}} or {{{Traversable}}}, but it readly admits a
 {{{Foldable}}} instance which {{{DeriveFoldable}}} would be able to infer:

 {{{#!hs
 instance Foldable ShellEffect where
     foldr f z (ShellEffect a) = z
     foldMap f (ShellEffect a) = mempty
 }}}

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10447#comment:8>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler