ANNOUNCE: Data.IVar 0.1
Several very elegant FRP approaches are emerging, most visibly FRP.Reactive, which rely on blocking on multiple variables at once, continuing when the *first* of them is available. . . inside an unsafePerformIO, so the beautiful STM "orElse" solution is not available. The current solution is to race threads against each other, and have the one that finishes first kill the other one. This is implemented, for example, in Data.Unamb. However, our empirical tests have shown that the GHC scheduler is not *quite* good enough to handle this efficiently, and ends up introducing too much latency and nondeterminacy. The Data.IVarhttp://hackage.haskell.org/cgi-bin/hackage-scripts/package/data-ivar-0.1module, just uploaded to hackage, provides an alternative to thread racing as a solution to this problem. It provides *write-once* variables which can be blocked on in parallel, without forking any threads or using STM (so it is safe to use in unsafePerformIO). Example usage from the documentation: import qualified Data.IVar as IVar import Control.Concurrent main = do iv <- IVar.new iv' <- IVar.new forkIO $ threadDelay 10000000 >> writeIVar iv' "my spoon is too big" let merger = IVar.read iv `mplus` IVar.read iv' print =<< IVar.nonblocking merger -- most likely "Nothing" print =<< IVar.blocking merger -- waits a while, then prints writeIVar iv' "i am a banana" -- throws error "IVar written twice" Enjoy! Luke
lrpalmer:
Several very elegant FRP approaches are emerging, most visibly FRP.Reactive, which rely on blocking on multiple variables at once, continuing when the *first* of them is available. . . inside an unsafePerformIO, so the beautiful STM "orElse" solution is not available. The current solution is to race threads against each other, and have the one that finishes first kill the other one. This is implemented, for example, in Data.Unamb. However, our empirical tests have shown that the GHC scheduler is not *quite* good enough to handle this efficiently, and ends up introducing too much latency and nondeterminacy.
Cool! Does you IVar implementation have anything in common with previous proposals for things called IVar (or say, 'ports', http://www.cse.unsw.edu.au/~chak/haskell/ports/)? What's the background for this abstraction? -- Don
On Wed, Oct 8, 2008 at 12:20 AM, Don Stewart
Several very elegant FRP approaches are emerging, most visibly FRP.Reactive, which rely on blocking on multiple variables at once, continuing when the *first* of them is available. . . inside an unsafePerformIO, so the beautiful STM "orElse" solution is not available. The current solution is to race threads against each other, and have
one that finishes first kill the other one. This is implemented, for example, in Data.Unamb. However, our empirical tests have shown that
lrpalmer: the the
GHC scheduler is not *quite* good enough to handle this efficiently, and ends up introducing too much latency and nondeterminacy.
Cool! Does you IVar implementation have anything in common with previous proposals for things called IVar (or say, 'ports', http://www.cse.unsw.edu.au/~chak/haskell/ports/http://www.cse.unsw.edu.au/%7Echak/haskell/ports/ )?
Yes, I picked up the name from haskell-cafe discussions a while back. Various forms have been popping in and out of the reactive libraries. I hadn't seen ports before. What's the background for this abstraction?
So like I said, I'm not too sure, I just stole the name and vague idea from discussions about it. As is easily noticed from the jive above, it's motivated by the continuous barrage of suboptimal FRP libraries. While I'm working on reactive from the top down, this is the beginning of one I have planned, taking baby steps from the bottom up (and it'll likely get incorporated into reactive too). Luke
On Oct 8, 2008, at 1:36 AM, Luke Palmer wrote:
What's the background for this abstraction?
So like I said, I'm not too sure, I just stole the name and vague idea from discussions about it.
I believe IVar or something similar used to be in the standard GHC libraries a long time ago. Conal had asked at some point why it was removed and if it could be added back, but ultimately gave up on that path and found another. Upon discovering this as I was working on my own from the bottom up, I found a couple ways to implement it and made it publicly known, but I never released a library besides some source code pastes here and there. So I didn't come up with the idea myself by any stretch, but I think I helped repopularize it. I'm attaching one of the more well-known variants in case anybody is interested, although to be honest I can't remember which one was actually the best as I have moved on from this approach. - Jake
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Oct 8, 2008, at 7:28 AM, Jake Mcarthur wrote:
I'm attaching one of the more well-known variants in case anybody is interested, although to be honest I can't remember which one was actually the best as I have moved on from this approach.
I just looked back over it. I had forgotten how far from the original idea I had gone. Perhaps this would be better with a different name now. - - Jake -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.8 (Darwin) iEYEARECAAYFAkjsqI4ACgkQye5hVyvIUKm/dQCfbzNfeTZTzmPQa6a2ZwGkqlbp ZZUAn2V5NrX1zR549EsEm8Oj5psU4fl6 =8vsl -----END PGP SIGNATURE-----
forkIO $ threadDelay 10000000 >> writeIVar iv' "my spoon is too big"
writeIVar iv' "i am a banana" -- throws error "IVar written twice"
Nice Don Hertzfeldt reference. ;) If this means nothing to you, here's the animated film from whence this came: http://www.youtube.com/watch?v=MuOvqeABHvQ Sean
participants (4)
-
Don Stewart -
Jake Mcarthur -
Luke Palmer -
Sean Leather