[GHC] #14765: Levity polymorphism panic

[GHC]

#14765: Levity polymorphism panic
-------------------------------------+-------------------------------------
        Reporter:  dfeuer            |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.6.1
       Component:  Compiler          |              Version:  8.5
      Resolution:                    |             Keywords:
                                     |  LevityPolymorphism
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  crash or panic                     |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by simonpj):

 Richard, this is a continuation of #12709.  During desugaring we have
 {{{
 f a b
 }}}
 Turns out that there's a levity-polymophism error in `(f a)`, so
 (bizarrely)
 `dsWhneNoErrs` returns `()`.  Then the desugarer tries to apply that to
 `b`,
 and that fails because `()` isn't a function type.

 Hhe cure (for #12709) was ugly enough; now we find that it doesn't
 acutally work! Replacing
 an expression by `()` doesn't make sense -- the types are wrong.  I can
 see two
 solutions (there may be others):

 * Instead of returning `()` return `undefined @ty` where `ty` is the
 needed type.

 But the whole `dsWhenNoErrs` machinery is very heavy; it calls `tryM`
 (which in
 turn allcoates fresh IO refes for errors etc) on every single application
 node.
 The cure is worse than the disease.
 So instead

 * We could define desugarer-only version of `mkCoreApp`, which doesn't
 blow up
   on levity-polymorphic args.  (Maybe it builds an ordinary App, say.)
 It's
   a bit of code to duplicate but not much, and it's easy to understand.

 (Alternatively we could make `isUnliftedType` return False for a levity-
 poly type.  That would be even simpler.)

 What do you think?

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14765#comment:2>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler