#14626: No need to enter a scrutinised value -------------------------------------+------------------------------------- Reporter: heisenbug | Owner: heisenbug Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.2.2 Resolution: | Keywords: performance Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: #13861 | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by heisenbug): Replying to [comment:26 simonpj]:
I think you believe that there is a heap-allocated `Name` with an un- evaluated `n_occ` field.
There is only one place in the code generator namely `StgCmmCon.buildDynCon'`. You could perhaps add (runtime) assertions
Yes. `n_occ` is a strict field and it appears to be non-WHNF when reading it. I was suspecting that this is because of deserialisation doing some dirty `unsafeCoerce` tricks when loading dictionaries, which contain the `Name` values, but there are other culprits too. I had to remove two further bangs on datatype fields on order to get a ''quick'' build with `make GhcStage2HcOpts="-O1 -g"` through. (https://github.com/ghc/ghc/commit/57a57f2f8cef2ea67588edd1f09f73981e86c889) So all evidence point towards a GHC defect in allocation of heap objects with strict fields. there to see if any strict fields had un-tagged pointers. I'll look into the assertion which you suggest. Btw. this might be a recent regression or something ancient. I'll find out and file another ticket if this gets confirmed. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14626#comment:27 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler