[GHC] #15300: Unboxed Sums Crash

[GHC]

#15300: Unboxed Sums Crash
-------------------------------------+-------------------------------------
        Reporter:  andrewthad        |                Owner:  osa1
            Type:  bug               |               Status:  new
        Priority:  highest           |            Milestone:  8.6.1
       Component:  Compiler          |              Version:  8.5
      Resolution:                    |             Keywords:  UnboxedSums
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 osa1):

 I hacked the Id printer to print dead binders with their types. It turns
 out the dead-but-used binder is this:

 {{{
 wild2_scuI [Occ=Dead]
     :: (# GHC.Maybe.Maybe GHC.Types.Any | Packed.Bytes.Parser.Bytes# #)
 }}}

 This kind of variables (dead case binders with unboxed sum/tuple types)
 are not supposed to be used! We even have a comment about this in
 `UnariseStg`:

 {{{
 unariseExpr rho (StgCase scrut bndr alt_ty alts)
   ...
   -- general case
   | otherwise
   = do scrut' <- unariseExpr rho scrut
        alts'  <- unariseAlts rho alt_ty bndr alts
        return (StgCase scrut' bndr alt_ty alts')
                        -- bndr may have a unboxed sum/tuple type but it
 will be
                        -- dead after unarise (checked in StgLint)
 }}}

 and we actually check this in StgLint:

 {{{
 lintStgExpr (StgCase scrut bndr alts_type alts) = do
     ...
              -- Case binders of unboxed tuple or unboxed sum type always
 dead
              -- after the unariser has run.
              -- See Note [Post-unarisation invariants].
              MultiValAlt _ -> not (lf_unarised lf)
     ...
 }}}

 I don't understand yet why this program passes StgLint.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15300#comment:15>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler