[GHC] #11715: Constraint vs *

[GHC]

#11715: Constraint vs *
-------------------------------------+-------------------------------------
        Reporter:  bgamari           |                Owner:
            Type:  bug               |               Status:  new
        Priority:  high              |            Milestone:  8.2.1
       Component:  Compiler (Type    |              Version:  8.0.1-rc1
  checker)                           |
      Resolution:                    |             Keywords:  Typeable
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 int-index):

 Couldn't we theoretically have constraints passed as unboxed data? In
 https://ghc.haskell.org/trac/ghc/wiki/UnliftedDataTypes I see that Edward
 Kmett suggests a separate flag to `TYPE`, `Constraintiness`, which would
 be more disciplined than misusing `RuntimeRep` and would solve the same
 problems.

 Except I think that it should be called `Coherency`, not
 `Constraintiness`, because the main feature of things that go to the left
 of `=>` is that they should have coherency properties (as was already
 discussed).

 {{{
 data Coherency = Coherent | Incoherent
 }}}

 This way we can even make `->` coherency-polymorphic (and eliminate
 `Dict`, that would be safe). `=>` could then require something `Coherent`,
 and `-XIncoherentContexts` (a companion to `-XIncoherentInstances`) could
 lift this restriction.

 Then `Int =>` would be disallowed unless the pragma is specified, but none
 of my objections would apply.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11715#comment:52>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler