#13422: INLINE CONLIKE sometimes fails to inline -------------------------------------+------------------------------------- Reporter: dfeuer | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: 8.4.1 Component: Compiler | Version: 8.1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Type of failure: Runtime | Unknown/Multiple performance bug | Test Case: Blocked By: | Blocking: Related Tickets: #7206 | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by bgamari): I've quickly looked at this and it seems that one crucial difference is that when `n` is banged the first phase simplifier considers it to be a `ValueArg` as it has a conlike unfolding, whereas without it is merely ` TrivialArg`. At first simplification proceeds similarly in both cases, * `$fEnumInt_$cenumFromTo` * `sum` is inlined (the unfolding of which contains a `foldl`). It is interesting to note that the list argument is considered to be a `ValueArg` with the bang, but a `TrivialArg` without. * `foldl` is inlined * `$fNumInt_$cfromInteger` is inlined This is where the two programs diverge. While the `fold/cheapBuild` rule fires in the banged case, it fails to fire on the unbanged program. After that point, things obviously proceed much differently. While `cheapBuild` itself is eventually inlined in the unbanged version, it the fusion rule never fires. This is rather curious since, * the desugared core is identical between the banged and unbanged case * the performed inlinings are identical -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/13422#comment:3 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler