[GHC] #13674: GHC doesn't discharge heterogeneous equality constraint when it ought to

[GHC]

#13674: GHC doesn't discharge heterogeneous equality constraint when it ought to
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler (Type    |              Version:  8.0.1
  checker)                           |
      Resolution:                    |             Keywords:  TypeInType
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  GHC rejects       |  Unknown/Multiple
  valid program                      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by goldfire):

 comment:8 is essentially correct. Consider two scenarios:

 1. We have a Given proving `blah` and a Wanted `blah`. We use the Given to
 solve the Wanted. Later, we learn that `blah` is `Lcm m m ~ m`, but
 nothing above changes.

 2. We have a Given proving `blah`. We learn that `blah` is `Lcm m m ~ m`,
 so we mark it as insoluble. Then, we get a Wanted `blah`, which we see is
 `Lcm m m ~ m`. This, too, is marked as insoluble.

 In case 1, we'll succeed; in case 2, we'll fail. The problem is that the
 only difference in these cases is the order in which constraints are
 treated and/or solved, something notoriously difficult to control. The
 "fix" for this problem is not to error on occurs-checks, which would then
 allow us to succeed in case 2. That seems unsatisfactory, though, because
 even if occurs-checks don't immediately error, we'll have a hard time
 solving `m ~ Lcm m m` from `Lcm m m ~ m`.

 I'm inclined to say that we let this behavior stand. Note that when we
 accept the program, nothing goes wrong.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13674#comment:9>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler