Asynchronous exceptions and laziness bugs in Data.Unique.newUnique

Bas van Dijk v.dijk.bas at
Fri Mar 19 09:16:19 EDT 2010

On Thu, Mar 18, 2010 at 11:45 AM, Simon Marlow <marlowsd at> wrote:
> Thanks, I'm actually going to change this to use STM as it does a better job
> than MVars in this case.

Hi Simon, I see you actually write 'val+1' to the MVar:

newUnique :: IO Unique
newUnique = atomically $ do
  val <- readTVar uniqSource
  let next = val+1
  writeTVar uniqSource $! val + 1
  return (Unique next)

Why don't you write 'next' so that when you force the returned Unique
it doesn't need to be computed a second time?

(The attached patch fixes it)


-------------- next part --------------
A non-text attachment was scrubbed...
Name: write-next.dpatch
Type: application/octet-stream
Size: 27970 bytes
Desc: not available
Url :

More information about the Libraries mailing list