[GHC] #11715: There's something fishy going on with Constraint

[GHC]

#11715: There's something fishy going on with Constraint
-------------------------------------+-------------------------------------
        Reporter:  bgamari           |                Owner:
            Type:  bug               |               Status:  new
        Priority:  high              |            Milestone:  8.0.1
       Component:  Compiler (Type    |              Version:  8.0.1-rc1
  checker)                           |
      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):

 I haven't looked at the code surrounding this, but here's a plausible
 story:

 - {{{* `eqType` Constraint}}} evaluates to `True`, in case you didn't
 know. This fact is necessary in order to simplify Core more aggressively,
 as I understand it.

 - The instance lookup mechanism looks for cached values. (See !TcInteract,
 line 1306.) An instance for `*` will serve nicely as an instance for
 `Constraint`.

 - So once an instance for `*` is found, that one is used for `Constraint`
 as well. This does not lead to any Core Lint problems (that is, it's
 perfectly type safe), because of my first bullet point.

 Of course, `*` and `Constraint` are not equal in Haskell! One long-term
 solution I have for this is to make `* ~R Constraint` but not `* ~N
 Constraint`, as we have it now. But roles in kinds are most certainly a
 story for another day.

 I do believe that a working solution is easy enough, though: just add a
 check, using `tcEqType`, in `TcSMonad.findDict`. `tcEqType` is careful
 ''not'' to relate `*` and `Constraint`. Recall that Core Lint can never
 check that we're doing this right, sadly.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11715#comment:4>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler