#15038: Memory Corruption (strange closure type) -------------------------------------+------------------------------------- Reporter: andrewthad | Owner: (none) Type: bug | Status: new Priority: highest | Milestone: 8.6.1 Component: Compiler | Version: 8.4.1 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: | -------------------------------------+------------------------------------- Changes (by osa1): * cc: simonpj, bgamari (added) * priority: normal => highest Comment: Finally found the problem. In summary; in unarise we introduce `absentError` calls for unboxed sum slots that are not supposed to be used and `absentError` id (`aBSENT_ERROR_ID`) is marked as CAFFY. Now suppose we had a top-level definition that is not CAFFY but uses unboxed sums. `TidyPgm` thinks that it's not CAFFY. But in unarise we make it CAFFY. This causes mismatch between the CAF info recorded in the binder id for this top-level definition and the actual RHS. The code that generates closure layout doesn't add a static_link field, but the code that generates the info table generates a SRT and SRT bitmap. Some potential solutions: - Why is `absentError` marked as CAF? I think error ids are conservatively marked as `MayHaveCafRefs` but perhaps it's fine to not mark `absentError` as `MayHaveCafRefs`. (I implemented this and it fixed the bug) - Update CAFFY-ness of ids in unarise if we introduce `absentError` (haven't tried this yet) - Something else? Any ideas? Ping @simonpj @bgamari -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/15038#comment:15 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler