[GHC] #14895: STG CSE makes dead binders undead

[GHC]

#14895: STG CSE makes dead binders undead
-------------------------------------+-------------------------------------
        Reporter:  hsyl20            |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.2.2
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Debugging         |  Unknown/Multiple
  information is incorrect           |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by osa1):

 Actually `isDeadBinder` check (the check that causes not printing case
 binders) is also used in code generation:

 {{{#!haskell
 cgCase (StgOpApp (StgPrimOp op) args _) bndr (AlgAlt tycon) alts
   | isEnumerationTyCon tycon -- Note [case on bool]
   = do { tag_expr <- do_enum_primop op args

        -- If the binder is not dead, convert the tag to a constructor
        -- and assign it.
        ; unless (isDeadBinder bndr) $ do
             { dflags <- getDynFlags
             ; tmp_reg <- bindArgToReg (NonVoid bndr)
             ; emitAssign (CmmLocal tmp_reg)
                          (tagToClosure dflags tycon tag_expr) }

        ; (mb_deflt, branches) <- cgAlgAltRhss
 (NoGcInAlts,AssignedDirectly)
                                               (NonVoid bndr) alts
                                  -- See Note [GC for conditionals]
        ; emitSwitch tag_expr branches mb_deflt 0 (tyConFamilySize tycon -
 1)
        ; return AssignedDirectly
        }
 }}}

 But this probably doesn't break anything because the scrutinee binder
 becomes undead only when (1) scrutinee is a data con application (2) the
 application repeats in an alternative. The code I showed above is for
 generating code when scrutinee is a primop application so it's safe.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14895#comment:3>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler