#10918: Float once-used let binding into a recursive function -------------------------------------+------------------------------------- Reporter: nomeata | Owner: Type: task | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.10.2 Resolution: | Keywords: 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 simonpj):
I suppose that would work, although I don’t like solving such issues by slamming just another pass to the end of the pipeline, instead of making sure the passes work well together (which is indeed tricky here).
I agree with the principle here. But it does seem hard. The float-out pass assumes, crudely, that it's good to float out a redex of any called- many lambda. But, as we see here, that's wrong for case branches that are only evaluated on one of those calls (the final one in this case). Not only is that info hard to record in the syntax tree, but it's also potentially quite fragile to program transformation, like other sorts of cardinality information. So refraining from let-floating after the final call-arity/simplifier pass does seem plausible. Annoyingly, it's just possible that inlining `(foo ys` into that `[]` branch might then put it in a context when `foo` inlines, leading to a cascade of further transformations. So it's not necessarily just a little delta. It'd be good to understand the paraffins thing. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/10918#comment:13 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler