#13429: Optimizer produces Core with an infinite <<loop>> -------------------------------------+------------------------------------- Reporter: lehins | Owner: (none) Type: bug | Status: new Priority: high | Milestone: 7.10.4 Component: Compiler | Version: 8.0.2 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: Runtime crash | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by dfeuer): I reduced the test case again. See `LoopReduced2.hs` and `mainReduced2.hs` (you'll have to rename the `LoopReduced2.hs` file to `Loop.hs` to compile). I attached `-dverbose-core2core` results before and after the early inlining patch as `dumpbefore.tgz` and `dumpafter.tgz`. I haven't figured out where things go wrong just yet, but if you look at `dumpbefore/main.dump-simpl`, you'll see {{{#!hs Rec { -- RHS size: {terms: 1, types: 0, coercions: 0, joins: 0/0} Main.$s$fArrayXe_$dNum [Occ=LoopBreaker] :: Num Word8 [GblId, Str=b] Main.$s$fArrayXe_$dNum = Main.$s$fArrayXe_$dNum end Rec } }}} which is very obviously very bad. For some reason we're "optimizing" a dictionary into a reference to itself! My vague guess is that GHC sees that it can get a `Num Word8` dictionary from an `Array` or `ColorSpace` dictionary, and that it can get `ColorSpace` and `Array` dictionaries from `Num` ones, and makes some very bad decisions. I have no idea why the early inline patch fixes this, but I'm rather nervous about it being a proper fix. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/13429#comment:13 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler