[GHC] #12881: GHC 8.0.2 regression involving OVERLAP annotations

[GHC]

#12881: GHC 8.0.2 regression involving OVERLAP annotations
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                Owner:
            Type:  bug               |               Status:  new
        Priority:  high              |            Milestone:  8.0.2
       Component:  Compiler          |              Version:  8.0.2-rc1
      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 RyanGlScott):

 So it looks like what is going on here is that in GHC 8.0.1 and before,
 there was different code generated under the hood for default class method
 implementations. This code happened to work irrespective of whether any
 overlapping instances were marked with `OVERLAPPING` or not. That's not to
 say that leaving off `OVERLAPPING` pragmas couldn't have caused problems
 down the road in GHC 8.0.1, but in the particular case of default method
 implementations, it luckily didn't matter.

 In GHC 8.0.2, however (after
 https://ghc.haskell.org/trac/ghc/changeset/d2958bd08a049b61941f078e51809c7e63bc3354/ghc),
 GHC switched to using visible type application to implement default class
 method implementations. This now poses an issue for any overlapping
 instances with default method implementations, because in the code that
 gets generated now, e.g.,

 {{{#!hs
 instance Arbitrary Int where
   shrink = dmshrink @Int
 }}}

 `dmshrink @Int` is forced to choose a particular `Arbitrary` instance, and
 without the `OVERLAPPING` annotation, GHC can't decide between the
 `Arbitrary a` instance and the `Arbitrary Int` instance.

 So on one hand, there is a somewhat good reason why this code now fails to
 typecheck. On the other hand, it's quite annoying—a nontrivial number of
 packages in the wild now fail to build with GHC 8.0.2, and there are
 probably more examples outside of Stackage that I haven't found. I'm not
 sure how to address this.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/12881#comment:5>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler