- Haskell-Cafe - Haskell.org

RE: [Haskell-cafe] ForeignPtrs with liveness dependencies
by Simon Marlow 26 Jul '05

26 Jul '05

On 26 July 2005 14:15, Einar Karttunen wrote: > "Simon Marlow" <simonmar(a)microsoft.com> writes: >>> Now the association becomes >>> associate (Foo _ ref) bar = >>> atomicModifyIORef ref (\lst -> (touchForeignPtr bar : lst, ())) >> >> Isn't that equivalent to using addForeignPtrFinalizer? I don't think >> this fixes anything: the finalizer for bar can still run before the >> finalizer for foo. > > foo has a single finalizer which is defined like: > > fooFinalizer cfoo ref = do cdeleteFoo cfoo > vs <- readIORef ref > mapM_ (\c -> c) vs the last line is equivalent to "sequence_ vs", FWIW. > and foo is created like > > createFoo ptr = do ref <- newIORef [] > fp <- newForeigPtr ptr (fooFinalizer ptr ref) > return (Foo fp ref) > > As the finalizer of foo references the IORef which contains > the list of actions containing the "touchForeignPtr bar" > the finalizer of foo is run first. The finalizer to bar > should be able to run only when the touchForeignPtr has been > executed in the mapM_ which only happens after foo > has been cleaned up - if I understand things correctly. No, unfortunately not. You have foo's finalizer which refers to bar via a touchForeignPtr. If both foo and bar are unreachable (references from finalizers don't count), then both foo and bar's finalizers will be started together, and may run in any order. So touchForeignPtr does only one thing: it expresses the precise relationship "bar is alive if foo is alive". If both are not alive, then both finalizers can run, in any order. I realise this is very subtle. By all means suggest improvements to the docs. Cheers, Simon

2 1

Data-types in dynamically loaded hs-plugins
by Thomas Sutton 26 Jul '05

26 Jul '05

Hi all, I'm [trying to] develop a system that uses plug-ins to implement a "black-box" ADT, that is a data-type and a few typeclass instances for it. I want my system (a theorem prover) to be able to work with any of these plug-ins (which define a logic and operations on its formulae) but have been having trouble. I've had a go at using forall's, and at adding a type variable (constrained to the appropriate class) to the interface of the plug- in, but I haven't been able to get it to work and have run out of ideas. Is it possible to define data-types in a module loaded at run-time with hs-plugins and have values pass out of the plug-in (though only being used by functions within it)? Can someone point me in the right direction? Cheers, Thomas Sutton

1 0

RE: [Haskell-cafe] ForeignPtrs with liveness dependencies
by Simon Marlow 26 Jul '05

26 Jul '05

On 26 July 2005 11:26, Einar Karttunen wrote: > "Simon Marlow" <simonmar(a)microsoft.com> writes: >> You might be able to find more information on this in the mailing >> list archives. It's true that touchForeignPtr isn't enough to >> enforce an ordering on the running of finalizers, but it *can* be >> used to express a liveness relationship between one ForeignPtr and >> another (ForeignPtr A is alive if ForeignPtr B is alive). This >> should be enough if you're dealing with pointer relationships >> between memory objects, for example, where it doesn't matter which >> one gets freed first when they're both unreferenced. > > The order of the cleanup functions is significant in this case, > so that does not unfortunately help. > >> If you really do need ordering, maybe it would be possible to use >> reference counting in your case? > > I ended up using the following design, which seems to work fine: > data Foo = Foo (ForeignPtr Foo) (IORef [IO ()]) > Each ForeignPtr Foo has a single finalizer which first calls the > C-side cleanup function for Foo and then executes all the IO-actions > inside the IORef. > > Now the association becomes > associate (Foo _ ref) bar = > atomicModifyIORef ref (\lst -> (touchForeignPtr bar : lst, ())) Isn't that equivalent to using addForeignPtrFinalizer? I don't think this fixes anything: the finalizer for bar can still run before the finalizer for foo. Cheers, Simon

2 1

RE: [Haskell-cafe] FFI and callbacks
by Simon Marlow 26 Jul '05

26 Jul '05

On 26 July 2005 11:32, Sebastian Sylvan wrote: > On 7/26/05, Simon Marlow <simonmar(a)microsoft.com> wrote: >> On 25 July 2005 22:41, Sebastian Sylvan wrote: >>> Well how about not touching the "bound threads" design at all, but >>> modifying the lightweigh threads design to also include "forkIOHere" >>> (or something) which would produce a lightweight thread which is >>> always run in the "main" OS thread. >> >> This is what Duncan was angling for, I think. My point is that this >> requires the implementation to support lightweight threads, which is >> something we intentionally avoided in the bound threads design. > > No it doesn't. If they don't support lightweight threads, they won't > have the forkIOHere function available. > I.e. leave it out of the bound threads design and just implement it as > a feature in the GHC lightweight threads system. I think you misunderstand me. I'm merely pointing out that the fact that it would be a GHC-specific feature is a drawback. (not that any other compilers have bound threads either, but still). Cheers, Simon

1 0

RE: [Haskell-cafe] FFI and callbacks
by Simon Marlow 26 Jul '05

26 Jul '05

On 25 July 2005 22:41, Sebastian Sylvan wrote: > On 7/25/05, Simon Marlow <simonmar(a)microsoft.com> wrote: >> On 23 July 2005 03:38, Duncan Coutts wrote: >> >>> The problem then as John noted is that the main loop of these >>> toolkits block and so the other Haskell threads would not get a >>> chance to schedule. So the challenge is to give the Haskell threads >>> a chance to schedule. >> >> [ good description of the multi-threaded GUI problem deleted ] >> >> Thanks for describing the problem in detail, I understand it better >> now. I think it comes down to this conflict: >> >> - you want to take advantage of the fact that GHC has lightweight >> "green" threads in order to do multithreading within a single OS >> thread, but >> >> - our "bound threads" design does not require the implementation >> to support lightweight threads, and hence doesn't let the >> programmer take advantage of them. > > Well how about not touching the "bound threads" design at all, but > modifying the lightweigh threads design to also include "forkIOHere" > (or something) which would produce a lightweight thread which is > always run in the "main" OS thread. This is what Duncan was angling for, I think. My point is that this requires the implementation to support lightweight threads, which is something we intentionally avoided in the bound threads design. While I don't doubt that we could implement some way to fork another thread bound to the same OS thread as the parent, I'm worried that this would add yet more complexity to GHC's scheduler, so I think we should fully explore other possibilities first. Cheers, Simon

2 1

RE: [Haskell-cafe] ForeignPtrs with liveness dependencies
by Simon Marlow 26 Jul '05

26 Jul '05

On 25 July 2005 14:44, Einar Karttunen wrote: > What is the correct way to express liveness dependencies for > ForeignPtrs? I am wrapping a C library and need a way to keep > ForeignPtrs alive until the finalizer for an another ForeignPtr > has been executed. > > Basically I have two types, ForeignPtr A and ForeignPtr B and a > function associate :: ForeignPtr A -> ForeignPtr B -> IO (). I want > to keep all of the ForeignPtr Bs associated with a given ForeignPtr A > alive until its finalizer has been run. The relationship is M:N - > each ForeignPtr A may be associated with multiple ForeignPtr B and > each ForeignPtr B may be associated with multiple ForeignPtr A. > > GHC documentation tells that touchForeignPtr is not enough as it makes > no guarantees about when the finalizers are run. If it helps the > finalizers are C functions which neither block nor perform callbacks > into Haskell. You might be able to find more information on this in the mailing list archives. It's true that touchForeignPtr isn't enough to enforce an ordering on the running of finalizers, but it *can* be used to express a liveness relationship between one ForeignPtr and another (ForeignPtr A is alive if ForeignPtr B is alive). This should be enough if you're dealing with pointer relationships between memory objects, for example, where it doesn't matter which one gets freed first when they're both unreferenced. If you really do need ordering, maybe it would be possible to use reference counting in your case? Cheers, Simon

2 1

RE: [Haskell-cafe] building Ghc with msvc ?
by Simon Marlow 26 Jul '05

26 Jul '05

On 23 July 2005 22:52, Duncan Coutts wrote: > On Sat, 2005-07-23 at 22:49 +0200, Jan Scott wrote: >> Hi, >> >> Has anyone ever tried to build Ghc with Microsofts c compiler (msvc) >> ? > > GHC relies quite heavily on gcc. GHC's C code backend uses a number of > GNU C extensions I believe. > > I think it would be a great deal of work to make it use msvc. And > there's no need since the mingw gcc works just fine on windows and is > binary compatible with msvc (if one uses the appropriate gcc options). I'd like to make the native code generator produce code in Intel syntax so it would be possible to use the MS binutils as an alterntiave to the GNU tools (the MS tools are available for free). Playing nicely with native Windows debuggers would be useful for debugging mixed Haskell/foreign language apps, I've had no end of trouble trying to use gdb for this. I believe the C parts of the RTS has been (is?) compilable with MSVC. The non-C parts of the RTS still need ghc -fvia-C (ie. gcc), but if I ever get around to adding support for loops in the native code generator we could drop that dependency too. I seriously doubt that we'd add support to the mangler for the Intel asm syntax, but generating Intel asm with the NCG is a distinct possibility. Cheers, Simon

1 0

Control.Monad.Cont fun
by oleg＠pobox.com 26 Jul '05

26 Jul '05

On Thu, Jul 07, 2005 at 07:08:23PM +0200, Tomasz Zielonka wrote: > Some time ago I wanted to return the escape continuation out of the > callCC block, like this: > > getCC = callCC (\c -> return c) It seems using shift/reset is better not only in principle but in practice as well. > module Foo where > > import Control.Monad.Cont > > -- From http://www.haskell.org/hawiki/MonadCont > reset :: (Monad m) => ContT a m a -> ContT r m a > reset e = ContT $ \k -> runContT e return >>= k > > shift :: (Monad m) => > ((a -> ContT r m b) -> ContT b m b) > -> ContT b m a > shift e = ContT $ \k -> > runContT (e $ \v -> ContT $ \c -> k v >>= c) return We can define a more general getCC' form simply as follows: > getCC' init = shift (\f -> f (init,jump f)) > where jump f x = f (x,jump f) The compiler figures out the types. Here's the usage example from the original Tomasz's message > test1' :: IO () > test1' = (`runContT` return) $ > do > reset (do > (x, jump) <- getCC' 0 > lift (print x) > when (x < 10) $ jump (x + 1)) > lift (putStrLn "finish") That was using the ContT monad exactly as it comes with GHC 6.4. With the CC_CPST monad transformer (with multiple polymorphic prompts) we can do better: > getCCP' p init = shiftP p (\f -> f (return (init,jump f))) > where jump f x = f (return (x,jump f)) The following code (written a couple of weeks) prints 1 through 10, then again 4 through 10 and then again 5 through 10, and then "finish". So we can not only note a label to jump to: we can return the label so to jump to it from different `functions'. In a sense, we emulate multiple entry points to a block and cross-block jumps. > test1'' :: IO () > test1'' = runCC ( > do > p <- newPrompt > p1 <- newPrompt > j <- pushPrompt p1 ( pushPrompt p ( > do > (x, jump) <- getCCP' p 0 > lift (print x) > when (x < 10) $ jump (x+1) > shiftP p1 (const (return jump))) >> undefined) > lift (putStrLn "again") > pushPrompt p1 (j 4 >> undefined) > lift (putStrLn "and again") > pushPrompt p1 (j 5 >> undefined) > lift (putStrLn "finish")) In this terminating code, undefined plays a great role.

1 0

Control.Monad.Cont fun
by Tomasz Zielonka 26 Jul '05

26 Jul '05

Hello! Some time ago I wanted to return the escape continuation out of the callCC block, like this: getCC = callCC (\c -> return c) But of course this wouldn't compile. I thought that it would be useful to be able to keep the current continuation and resume it later, like you can in do in scheme. Well, it was easy to do this in Haskell in the (ContT r IO) monad, by using an IORef. But I wasn't able to solve this for all MonadCont monads. After more then year of on-and-off trials, I've finally managed to do this! ;-) import Control.Monad.Cont getCC :: MonadCont m => m (m a) getCC = callCC (\c -> let x = c x in return x) getCC' :: MonadCont m => a -> m (a, a -> m b) getCC' x0 = callCC (\c -> let f x = c (x, f) in return (x0, f)) getCC allows to set a backward "goto label". Here's an example: -- prints "hello!" in an endless loop test :: IO () test = (`runContT` return) $ do jump <- getCC lift (putStrLn "hello!") jump getCC' allows to jump back with arguments: -- prints integers from 0 to 10, then prints finish and ends test' :: IO () test' = (`runContT` return) $ do (x, jumpWith) <- getCC' 0 lift (print x) when (x < 10) (jumpWith (x + 1)) lift (putStrLn "finish") I think this is rather cool! :-) Besides sharing my happiness, I want to ask some questions: - what would be the best names for these functions - is it possible to define a MonadFix instance for Cont / ContT? - do you think it would be a good idea to add them to Control.Monad.Cont? Best regards Tomasz

5 10

RE: [Haskell-cafe] FFI and callbacks
by Simon Marlow 25 Jul '05

25 Jul '05

On 23 July 2005 03:38, Duncan Coutts wrote: > The problem then as John noted is that the main loop of these toolkits > block and so the other Haskell threads would not get a chance to > schedule. So the challenge is to give the Haskell threads a chance to > schedule. [ good description of the multi-threaded GUI problem deleted ] Thanks for describing the problem in detail, I understand it better now. I think it comes down to this conflict: - you want to take advantage of the fact that GHC has lightweight "green" threads in order to do multithreading within a single OS thread, but - our "bound threads" design does not require the implementation to support lightweight threads, and hence doesn't let the programmer take advantage of them. When we were thinking about bound threads, the idea was to accommodate multiple implementations including the one-to-one and giant lock approaches. We consider(ed) GHC's mixture of lightweight and heavyweight threads to be an optimisation. However, the GUI library example is interesting, because if we could take advantage of lightweight threads it seems we could have multithreaded access to the GUI with better performance and no requirement to make all GUI calls from a single thread. But I'm not convinced. There are advantages to using the single GUI thread approach, and we don't have any measurements for whether the overhead is too much. Suppose you set up a channel containing requests of type (IO a, MVar a), where the MVar takes the result. You can build a combinator withGUI :: IO a -> IO a to perform a GUI operation (and this allows you to try multiple approaches without changing the code). Using withGUI you can send a whole batch of GUI operations in one go, it doesn't have to be one request per GUI call. You could also build asyncGUI :: IO a -> IO () which fires off a GUI request that doesn't require a result, and combines it with existing requests in a single batch if possible. Ordering would be retained, of course. The GUI thread will run multiple requests sequentially, it doesn't have to context switch between each request. And you get more parallelism, because the other threads continue to run while the GUI thread is in the main loop. Cheers, Simon

4 5