#4012: Compilation results are not deterministic -------------------------------------+------------------------------------- Reporter: kili | Owner: Type: bug | Status: new Priority: high | Milestone: 7.12.1 Component: Compiler | Version: 6.12.2 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Type of failure: Other | Unknown/Multiple Blocked By: | Test Case: Related Tickets: | Blocking: | Differential Revisions: -------------------------------------+------------------------------------- Comment (by thoughtpolice): Yes. I removed `shortOutIndirections` in `SimplCore`: {{{ diff --git a/compiler/simplCore/SimplCore.hs b/compiler/simplCore/SimplCore.hs index 0a2f8e4..95c5cce 100644 --- a/compiler/simplCore/SimplCore.hs +++ b/compiler/simplCore/SimplCore.hs @@ -667,7 +667,7 @@ simplifyPgmIO pass@(CoreDoSimplify max_iterations mode) -- -- ToDo: alas, this means that indirection-shorting does not happen at all -- if the simplifier does nothing (not common, I know, but unsavoury) - let { binds2 = {-# SCC "ZapInd" #-} shortOutIndirections binds1 } ; + let { binds2 = {-# SCC "ZapInd" #-} if True then binds1 else shortOutIndirections binds1 } ; -- Dump the result of this iteration dump_end_iteration dflags print_unqual iteration_no counts1 binds2 rules1 ; }}} Then I compiled the test in comment:76 with `-fno-cse` and observed the ABI difference went away. So killing CSE may get us a long way for NixOS I admit I tested this on GHC 7.10, and not the HEAD branch. Brian, you couldn't reproduce this on HEAD at all? I can double check this, but that somewhat surprises (and worries!) me. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/4012#comment:98 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler