[GHC] #16070: Better inferred signatures

[GHC]

#16070: Better inferred signatures
-------------------------------------+-------------------------------------
        Reporter:  simonpj           |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.6.3
      Resolution:                    |             Keywords:  TypeFamilies
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 AntC):

 Replying to [ticket:16070 simonpj]:
 > ... quite a bit shorter.

 To be clear what's intended here: the "shorter" could be achieved by
 something akin to Common Subexpression Elimination. That is:

 * Common type (sub-)expressions appearing in constaints.
 * Replace by a bare unification tyvar.
 * Add a `~` constraint between the tyvar and the type expression.

 But wait, there's more:

 > After all, it does some similar abbreviation with type classes.  Where
 we could
 > infer `(Ord a, Eq a)` we collapse out the `Eq a` and just infer the
 context `(Ord a)`.

 The `~` constraint generated for the common type expression `FieldType` is
 exactly the superclass constraint for `HasField`, just as `Eq a` is a
 superclass constraint for `(Ord a)`.

 Note also a later comment on that github discussion that the `FieldType`
 constraint holding implies there's a `HasField` instance -- because
 `FieldType` is an Associated Type. Using that implication seems harder:
 Assoc Types are trreated as sugar for top-level type families/not
 necessarily grounded in instances, which already gives trouble.

 >
 > There is nothing record-specific about this ticket. It's just about
 using inferred equality constraints to simplify the inferred signature.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/16070#comment:6>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler