#11158: Combine exprIsTrivial and cpe_ExprIsTrivial -------------------------------------+------------------------------------- Reporter: simonpj | Owner: Type: task | Status: infoneeded Priority: highest | Milestone: 8.2.1 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): Phab:D1656 Wiki Page: | -------------------------------------+------------------------------------- Comment (by bgamari): Yesterday I start from a clean slate and started by again turning `cpe_ExprIsTrivial` calls into calls to `exprIsTrivial`. Recall that this essentially marks STG string and integer literals as non-trivial, ensuring that they are not duplicated in applications like .... After doing this I found that a `DEBUG` compiler build failed while building `GHC.Types` with an assertion failure in `CoreToStg`. Namely, the `consistentCafInfo` check failed as the `IdInfo` claimed `NoCafRefs` yet the binding itself was caffy. The binding in question is of the form, {{{#!hs $tcMode1 :: TrName [GblId, Caf=NoCafRefs, Str=m1] = \u [] case "Mode"# of sat_swgf { __DEFAULT -> TrNameS [sat_swgf]; }; }}} The reason `topStgBindHasCafRefs` considers this to be caffy is because the RHS is an updatable `StgRhsClosure`. It seems to me that the problem here is that we floated the string literal as an case analysis, meaning that we have turned what should be a `StgRhsCon` into a `StgRhsClosure`. I believe this is ultimately the same problem presented by #11312. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11158#comment:31 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler