#15005: GHC fails with “StgCmmEnv: variable not found” when trying to compile order-maintenance-0.2.1.0 -------------------------------------+------------------------------------- Reporter: jeltsch | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: 8.4.3 Component: Compiler | Version: 8.4.2-rc1 Resolution: | Keywords: 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): The bug is in this line of `Exitify` {{{ -- The possible arguments of this exit join point args = filter (`elemVarSet` fvs) captured }}} You have to do at least `map zapStableUnfolding` over args. Consider {{{ joinrec j x = let {-# INLINE foo #-} foo y = blah in case x of True -> j2 foo (h x) False -> ...j.... }}} We are going to exitify the True branch of the 'case x'. But 'foo' is captured, so we get {{{ join exit foo x = jump j2 foo (h x) in joinrec j x = let {-# INLINE foo #-} foo y = blah in case x of True -> jump exit foo x False -> ...j.... }}} But we must zap `foo`'s unfolding when we lambda-abstract it! In effect, the lambda-bound `foo` is an entirely new `foo`, unrelated to the original. Actually, arguably we should give the lambda-bound foo `vanillaIdInfo`. We do this in `SetLevels.abstractVars`. I think that'd be safer. We don't seem to have a function `zapAllIdInfo` but we probably should. Over to you. Please include some version of this explanation, of course. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/15005#comment:6 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler