November 2010 - Glasgow-haskell-users

Type error in GHC-7 but not in GHC-6.12.3
by Bas van Dijk 01 Nov '10

01 Nov '10

Hello, I'm updating my usb-safe package for GHC-7: darcs get http://code.haskell.org/~basvandijk/code/usb-safe It depends on the HEAD version of regions: darcs get http://code.haskell.org/~basvandijk/code/regions I think I'm suffering from the new implied MonoLocalBinds extension (I'm using GADTs) as described in: http://hackage.haskell.org/trac/ghc/blog/LetGeneralisationInGhc7 However, I'm not sure this is the problem because I'm not using local bindings and use explicit type signatures everywhere. I try to make a small isolated example when I have time but for now let's use the actual definitions: The following function type-checked fine in GHC-6.12.3 but fails in GHC-7.1.20101027: withDeviceWhich ∷ ∀ pr α . MonadCatchIO pr ⇒ USB.Ctx → (USB.DeviceDesc → Bool) → (∀ s. RegionalDeviceHandle (RegionT s pr) → RegionT s pr α) → pr α withDeviceWhich ctx p f = do devs ← liftIO $ USB.getDevices ctx useWhich devs withDevice p f The error I get is: Couldn't match expected type `forall s. RegionalDeviceHandle (RegionT s pr) -> RegionT s pr α' with actual type `RegionalDeviceHandle (RegionT s pr) -> RegionT s pr α' In the fourth argument of `useWhich', namely `f' These are the types and definitions of the other functions involved: useWhich ∷ ∀ k desc e (m ∷ * → *) α . (GetDescriptor e desc, MonadIO m) ⇒ [e] → (e → k → m α) → (desc → Bool) → k → m α useWhich ds w p f = case find (p ∘ getDesc) ds of Nothing → throw USB.NotFoundException Just d → w d f withDevice ∷ MonadCatchIO pr ⇒ USB.Device → (∀ s. RegionalDeviceHandle (RegionT s pr) → RegionT s pr α) → pr α withDevice dev f = runRegionT $ openDevice dev >>= f Note that when I inline the definition of useWhich it type-checks fine: withDeviceWhich ctx p f = do devs ← liftIO $ USB.getDevices ctx case find (p ∘ getDesc) devs of Nothing → throw USB.NotFoundException Just d → withDevice d f Since I'm not using local bindings and use explicit type signatures everywhere, I'm not sure MonoLocalBinds is the problem. Note that other applications of useWhich which also use RankNTypes type-check just fine in both GHC-6.12.3 and GHC-7.1.20101027: ------------------------------------------------------------------------------- setConfigWhich ∷ ∀ pr cr α . (pr `AncestorRegion` cr, MonadCatchIO cr) ⇒ RegionalDeviceHandle pr → (USB.ConfigDesc → Bool) → (∀ sCfg. ConfigHandle sCfg → cr α) → cr α setConfigWhich h = useWhich (getConfigs h) setConfig getConfigs ∷ RegionalDeviceHandle r → [Config r] setConfig ∷ ∀ pr cr α . (pr `AncestorRegion` cr, MonadCatchIO cr) ⇒ Config pr → (∀ sCfg. ConfigHandle sCfg → cr α) → cr α ------------------------------------------------------------------------------- withInterfaceWhich ∷ ∀ pr sCfg α . MonadCatchIO pr ⇒ ConfigHandle sCfg → (USB.Interface → Bool) → (∀ s. RegionalInterfaceHandle sCfg (RegionT s pr) → RegionT s pr α) → pr α withInterfaceWhich h = useWhich (getInterfaces h) withInterface getInterfaces ∷ ConfigHandle sCfg → [Interface sCfg] withInterface ∷ ∀ pr sCfg α . MonadCatchIO pr ⇒ Interface sCfg → (∀ s. RegionalInterfaceHandle sCfg (RegionT s pr) → RegionT s pr α) → pr α ------------------------------------------------------------------------------- setAlternateWhich ∷ ∀ pr cr sCfg α . (pr `AncestorRegion` cr, MonadCatchIO cr) ⇒ RegionalInterfaceHandle sCfg pr → (USB.InterfaceDesc → Bool) → (∀ sAlt. AlternateHandle sAlt pr → cr α) → cr α setAlternateWhich h = useWhich (getAlternates h) setAlternate getAlternates ∷ RegionalInterfaceHandle sCfg r → [Alternate sCfg r] setAlternate ∷ ∀ pr cr sCfg α . (pr `AncestorRegion` cr, MonadCatchIO cr) ⇒ Alternate sCfg pr → (∀ sAlt. AlternateHandle sAlt pr → cr α) → cr α ------------------------------------------------------------------------------- I'm happy to provide more info when needed. Regards, Bas

2 3

RE: Type error in GHC-7 but not in GHC-6.12.3
by Simon Peyton-Jones 01 Nov '10

01 Nov '10

| foo :: (forall s. ST s a) -> a | foo st = ($) runST st This is a motivating example for type inference that can deal with impredicative types. Consider the type of ($): ($) :: forall p q. (p->q) -> p -> q In the example we need to instantiate 'p' with (forall s. ST s a), and that's what impredicative polymorphism means: instantiating a type variable with a polymorphic type. Sadly, I know of no system of reasonable complexity that can typecheck 'foo' unaided. There are plenty of complicated systems, and I have been a co-author on papers on at least two, but they are all Too Jolly Complicated to live in GHC. We did have an implementation of boxy types, but I took it out when implementing the new typechecker. Nobody understood it. However, people so often write runST $ do ... that in GHC 7 I implemented a special typing rule, just for infix uses of ($). Just think of (f $ x) as a new syntactic form, with the obvious typing rule, and away you go. It's very ad hoc, because it's absolutely specific to ($), and I'll take it out if you all hate it, but it's in GHC 7 for now. Anyway, that's why you encountered the puzzling behaviour you describe below. Simon | -----Original Message----- | From: Bas van Dijk [mailto:v.dijk.bas@gmail.com] | Sent: 30 October 2010 21:14 | To: glasgow-haskell-users(a)haskell.org | Cc: Simon Peyton-Jones | Subject: Re: Type error in GHC-7 but not in GHC-6.12.3 | | On Sat, Oct 30, 2010 at 1:57 AM, Bas van Dijk <v.dijk.bas(a)gmail.com> wrote: | > I could isolate it a bit more if you want. | | And so I did. The following is another instance of the problem I'm | having but set in a more familiar setting: | | {-# LANGUAGE RankNTypes #-} | | import Control.Monad.ST | | foo :: (forall s. ST s a) -> a | foo st = ($) runST st | | Couldn't match expected type `forall s. ST s a' | with actual type `ST s a' | In the second argument of `($)', namely `st' | | Note that: 'foo st = runST st' type checks as expected. | | Surprisingly 'foo st = runST $ st' also type checks! | | I find the latter surprising because according to the report[1]: e1 op | e2 = (op) e1 e2. So either both should type check or both should fail. | | I guess that a RULE somewhere eliminates the ($) before the | type-checker kicks in. I do find that a little strange because I | thought RULES where applied after type checking. | | Regards, | | Bas | | [1] http://www.haskell.org/onlinereport/exps.html#operators

1 0

Re: [Haskell-cafe] Equivalent of withForeignPtr for System.Mem.Weak?
by Antoine Latter 01 Nov '10

01 Nov '10

On Sun, Oct 31, 2010 at 10:14 PM, Matthew Steele <mdsteele(a)alum.mit.edu> wrote: > I have an object to which I have added one or more finalizers via > addFinalizer from System.Mem.Weak. I would like to have a function that > allows me to make use of the object within a block of IO code, and guarantee > that the finalizer(s) will not be called during the code block -- sort of > like withForeignPtr, but for arbitrary objects. Does such a function exist, > and if not, how could I write one? > > I can imagine writing something like: > > \begin{code} > withObject :: a -> IO b -> IO b > withObject obj action = do > result <- action > touchObject obj -- touchObject :: a -> IO () > return result > \end{code} > CC'ing GHC-users list. The 'primitive' package has a general purpose 'touch' function: http://hackage.haskell.org/packages/archive/primitive/0.3.1/doc/html/Contro… But it doesn't state what guarantees are provided by the function. Antoine

1 0