[GHC] #16310: Program fails with "Impossible case alternative" when optimized

[GHC]

#16310: Program fails with "Impossible case alternative" when optimized
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                Owner:  (none)
            Type:  bug               |               Status:  closed
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.6.3
      Resolution:  wontfix           |             Keywords:  TypeInType
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  Runtime crash     |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------
Changes (by RyanGlScott):

 * status:  new => closed
 * resolution:   => wontfix


Comment:

 After some thought, I've come to the conclusion that this isn't a bug.
 Here is what is happening. GHC scrutinizes a `$WRefl` of type `T1Sym :~:
 T2Sym`. GHC, in its infinite wisdom, recognizes that `T1Sym` and `T2Sym`
 are distinct data types, and therefore cannot possibly be equal.
 Therefore, `SimplUtils.mkCase` transforms this into `case ($WRefl :: T1Sym
 :~: T2Sym) of {}`, since the `Refl` case is "unreachable". However, due to
 our crafty use of `unsafeCoerce`, we //do// actually enter this case, and
 disaster strikes.

 I've come to the conclusion that `axiom`, much like `unsafePerformIO`,
 requires careful use of `NOINLINE` in order to use in a safe fashion. The
 fact that `apply` must be marked as `NOINLINE` reflects this reality.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/16310#comment:4>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler