[GHC] #13194: Concurrent modifications of package.cache are not safe

[GHC]

#13194: Concurrent modifications of package.cache are not safe
-------------------------------------+-------------------------------------
        Reporter:  arybczak          |                Owner:
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  ghc-pkg           |              Version:  8.0.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by duncan):

 The `fnctl` semantics is problematic in general but not for our use case
 for ghc/ghc-pkg with the package.conf file. Making a general reusable API
 is nice, but not essential. The `flock` API is probably the one to go
 with, despite NFS issues.

 As for a suitable API, either internal or for more general consumption,
 the API presented by the filelock is a useful place to start (and it has
 good code we can borrow for the FFI bits). My first guess is that a
 reasonable route to try is to go with a `Handle` based API, like

 {{{
 hLockFile :: Handle -> LockMode -> IO FileLock
 hTryLockFile :: Handle -> LockMode -> IO (Maybe FileLock)
 hUnlockFile :: Handle -> FileLock -> IO ()
 withFileLock :: Handle -> LockMode -> (FileLock -> IO a) -> IO a
 }}}

 I'm not yet totally clear what lives in the `FileLock`. Is this for the
 sake of the Win32 API because it returns a lock handle? And perhaps
 `hUnlockFile` doesn't need the `Handle` too.

 > but that requires extracting file descriptor from Handle, which as far
 as I see is problematic.

 It's actually totally possible. It simply fails for Handles that are not
 based on FDs, but that's ok.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13194#comment:2>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler