dgit.raspbian.org Git - git-annex.git/blob

   1 [[!comment format=mdwn
   2  username="joey"
   3  subject="""comment 20"""
   4  date="2018-11-19T17:04:18Z"
   5  content="""
   6 I don't see anything in ghc 8.0.2's release notes that points to a breaking
   7 change to STM/MVar handling. There were a couple of changes in that version
   8 that involve concurrency and MVars though, that could somehow have led to
   9 the problem.
  10
  11 Parallel programs should be significantly more reliable on platforms with
  12 weak memory consistency guarantees
  13 <https://ghc.haskell.org/trac/ghc/ticket/12469>
  14
  15 Scheduling bug with forkOS + MVar
  16 <https://ghc.haskell.org/trac/ghc/ticket/12419>
  17
  18 One thing that seems worth trying re the second of those
  19 is to edit CmdLine/Action.hs and delete the setNumCapabilities line:
  20
  21         --- a/CmdLine/Action.hs
  22         +++ b/CmdLine/Action.hs
  23         @@ -178,8 +178,6 @@ allowConcurrentOutput a = do
  24            where
  25                 goconcurrent n = do
  26                         c <- liftIO getNumCapabilities
  27         -               when (n > c) $
  28         -                       liftIO $ setNumCapabilities n
  29                         withMessageState $ \s -> case outputType s of
  30                                 NormalOutput -> ifM (liftIO concurrentOutputSupported)
  31                                         ( Regions.displayConsoleRegions $
  32
  33 A compiler bug is looking not entirely unlikely..
  34 """]]