[GHC] #12660: singletons doesn't compile with GHC 8.0.2 snapshot

[GHC]

#12660: singletons doesn't compile with GHC 8.0.2 snapshot
-------------------------------------+-------------------------------------
        Reporter:  bgamari           |                Owner:
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.0.2
       Component:  Compiler          |              Version:
      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 see what's going on, but I don't know why things are the way they are
 nor how to fix.

 The problem is, essentially, that we unflatten and then continue solving.
 Then, we have a fuv (flattening unification variable) that's filled in
 with the function it equals. So the flattener ends up bouncing between the
 function application and its fuv (as found in the flat-cache).

 The unflattening is happening in `TcInteract.solve_simple_wanteds`, as
 called from `solveSimpleWanteds`. But then the `go` loop in the latter
 goes around again. It does so in the same instance of the `TcS` monad, so
 that flat-cache is retained, even though unflattening has happened. And
 then we're in trouble.

 Questions:
 1. Should the flat-cache be wiped every time we unflatten?

 2. Should we even allow unflattening in the `TcS` monad? I always thought
 unflattening was about restoring nice-looking types for error messages and
 something that should happen just as we leave the `TcS` monad. But we
 shouldn't resume solving after unflattening, I think.

 I'm afraid I've reached the limit of what I can do with this ticket.
 Simon? :)

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/12660#comment:4>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler