[Haskell-cafe] Re: Timeouts that don't cause data growth.

[Bas van Dijk]

On Tue, Mar 23, 2010 at 10:20 PM, Simon Marlow <marlowsd at gmail.com> wrote:
> The leak is caused by the Data.Unique library, and coincidentally it was
> fixed recently.  6.12.2 will have the fix.

Oh yes of course, I've reported that bug myself but didn't realize it
was the problem here :-)

David, to clarify the problem: newUnqiue is currently implemented as:

newUnique :: IO Unique
newUnique = do
   val <- takeMVar uniqSource
   let next = val+1
   putMVar uniqSource next
   return (Unique next)

You can see that the 'next' value is lazily written to the uniqSource
MVar. When you repeatedly call newUnique (like in your example) a big
thunk is build up: 1+1+1+...+1 which causes the space-leak. In the
recent fix, 'next' is strictly evaluated before it is written to the
MVar which prevents a big thunk to build up.

regards,

Bas