- Haskell-Cafe - Haskell.org

unexpected behavior / bug in try . System.FilePath.Find.findWithHandler
by Thomas Hartman 18 Jan '10

18 Jan '10

say you want to execute a find function, but abort the computation if it hits any snags, such as an unreadable directory (eg chmod -R a-r dir). Currently try . System.FilePath.Find.findWithHandler will return an exception wrapped in Right, which seems Wrong. For sure it will just get ignored if wrapped in an ErrorT computation and I suspect this could lead to other glitchy/unexpected behavior when used in sysadmin scripts. You do get the expected exception wrapped in Left for certain errors (such as a missing directory), which adds to the confusion. Probable fix: you get the expected behavior if you remove unsafeInterleaveIO from findWithHandler. Does this seem like a reasonable thing to suggest to libraries@haskell? Hacky fix: I haven't actually done this, but I suppose you could trap stdout or stderr, since the exception does get printed although it's not reflected in the result type of the computation. demo of problem: import qualified System.FilePath.Find as F import System.FilePath.Find import System.IO.Error abort path err = fail $ path ++ " , " ++ (show err) prot = "/home/thartman/protected" -- a protected (unreadable) dir {- want result => Left user error (/home/thartman/protected , user error (/home/thartman/protected , /home/thartman/protected: getDirectoryContents: permission denied (Permission denied))) but get result Right *** Exception: user error (/home/thartman/protected , /home/thartman/protected: getDirectoryContents: permission denied (Permission denied)) which means, for example, you would miss the error if you wrapped this in an ErrorT there is probably a workaround from reading stderr, since it does get printed out, but this is awkward compared to just catching the error the right way and certainly feels like unexpected behavior Possible fix: you get the expected behavior if you remove unsafeInterleaveIO from System.FilePath.Find.findWithHandlers don't know why this fixes the problem, I was just messing around I don't know if this breaks anything else, but I was unable to find any breakage when I did a simple test on an unprotected directory What is the gain from unsafeInterleaveIO, and can it be nixed? -} tprot = try . findWithHandler abort (return True) (return True) $ prot tunprot = return . either (error . show) (length) =<< ( try $ findWithHandler abort (return True) (return True) unprot ) unprot = "/home/thartman/unprot" -- Need somewhere to put your code? http://patch-tag.com Want to build a webapp? http://happstack.com

2 1

ghc "make install" question
by Johannes Waldmann 18 Jan '10

18 Jan '10

Hi. How can I get "make install" from a ghc (snapshot) source dir to install binaries $prefix/bin/ghc{,i,-pkg}-$version but *not* overwrite the un-versioned symlinks? Thanks - J.W.

1 0

ANN: atom-0.1.3
by Tom Hawkins 18 Jan '10

18 Jan '10

This release of Atom slightly changes the semantics of assertions and coverage. Assertion and coverage are now checked between the execution of every rule, instead of only when the rules containing assertions are fired. They are still subject to parental guard conditions, but not period or phase constraints. This means... period 20 $ atom "checkSomeStuff" $ do cond ok assert "A" a assert "B" b cover "C" c ... A, B, and C are checked all the time "ok" is true, not just every 20th cycle. Checking between every rule execution obviously impacts simulation time, but the increased testing rigor is worth it. I also added 'linear' to Common, which does linear interpolation and extrapolation on a line given two points. (I found I was replicating this function everywhere for sensor calibrations, control limits, etc.) http://hackage.haskell.org/package/atom

2 5

A numpy equivalent for Haskell
by yairchu＠gmail.com 18 Jan '10

18 Jan '10

Hi Cafe, I've created a numpy equivalent for Haskell. (Numpy is a python library for multi-dimensional arrays and operations on them) Code at http://github.com/yairchu/numkell (not yet on hackage because it needs better names) A numkell array is a pair of a function from integer inputs and a range for its inputs (size). This allows for easy memoizing into in-memory arrays, and additionally, numkell arrays also support useful operations like numpy's newaxis and folding axes away. As the "Array" name was already taken, numkell's array is currently called "Funk" (name suggestions very appreciated). An example: Given an bunch of vectors as a 2d array, compute the distance between each pair of vectors {-# LANGUAGE DeriveDataTypeable, GeneralizedNewtypeDeriving, TypeOperators #-} import Data.HList import Data.NumKell import Data.Typeable newtype PersonIdx = PersonIdx Int deriving (Enum, Eq, Integral, Num, Ord, Real, Show, Typeable) newtype FeatureIdx = FeatureIdx Int deriving (Enum, Eq, Integral, Num, Ord, Real, Show, Typeable) let personProps = fFromList [[5,3,2],[4,8,1],[2,6,9],[5,3,0]] :: Funk (HJust PersonIdx :*: HJust FeatureIdx :*: HNil) Double > personProps FeatureIdx 0 1 2 PersonIdx + - - - 0 | 5.0 3.0 2.0 1 | 4.0 8.0 1.0 2 | 2.0 6.0 9.0 3 | 5.0 3.0 0.0 > sumAxes (fmap (** 2) (liftF2 (-) (personProps !/ (SNewAxis .*. HNil)) (personProps !/ (SAll .*. SNewAxis .*. HNil)))) (TFalse .*. TFalse .*. TTrue .*. HNil) PersonIdx 0 1 2 3 PersonIdx + - - - - 0 | 0.0 27.0 67.0 4.0 1 | 27.0 0.0 72.0 27.0 2 | 67.0 72.0 0.0 99.0 3 | 4.0 27.0 99.0 0.0 In Python the last line looks shorter: >>> ((personProps[newaxis] - personProps[:,newAxis]) ** 2).sum(2) Mostly due to Python's slicing syntax sugar. Still, numkell has one large benefit over numpy (apart from being for Haskell): With numpy this example creates a temporary 3d array in memory. In numkell the array is not allocated in memory unless "fMemo" is called. If anyone has comments, suggestions, naming suggestions, complaints, etc, I would very much like to hear. cheers, Yair

3 2

Takusen - reading from one db inserting into another
by Günther Schmidt 18 Jan '10

18 Jan '10

Hi, I know how to use Takusen with a *single* connection, but I cannot figure out how to read from one database while inserting to another. Could someone please explain to me how I can do that? Günther

3 3

Visualizing function application
by Martijn van Steenbergen 18 Jan '10

18 Jan '10

Dear café, I am deeply impressed with Vacuum[1][2], Ubigraph[3] and especially their combination[4]. I can trivially and beautifully visualize the ASTs that my parser produces. I can visualize zippers of the ASTs and confirm that sharing is optimal. Ubigraph is also able to animate graph *mutations*, as shown by the various demos on Ubigraph's website. How cool would it be if we could tell vacuum-ubigraph: here, show this tree, and then show how the tree changes when we apply this function on it. We could see how [1,2,3] is transformed into a ring when cycle is applied on it, or we could see how a list is consumed node by node when a foldr is applied. I have no idea how difficult this is or how to begin, so I thought I'd throw the idea out here. Perhaps it is appealing enough that someone picks it up and implements it. :-) Martijn. [1] http://hackage.haskell.org/package/vacuum [2] http://www.youtube.com/watch?v=oujaqo9GAmA [3] http://ubietylab.net/ubigraph/content/Demos/index.html [4] http://hackage.haskell.org/package/vacuum-ubigraph

2 1

Parse error
by Andrew Coppin 18 Jan '10

18 Jan '10

Is there a specific reason why GHC consistently refuses to accept the following perfectly reasonable code snippet? main = do putStrLn "Line 1" putStrLn "Line 2" let xs = do x <- [1..10] y <- [1..10] return (x+y) print xs No matter which way I rearrange this, it *insists* that there's a parse error. This is very frustrating, given that it's utterly clear what I want...

9 12

xmonad cannot find its library after upgrading GHC
by zaxis 18 Jan '10

18 Jan '10

After `pacman -Syu`, the ghc is upgraded to 6.12.1 %ghc --version The Glorious Glasgow Haskell Compilation System, version 6.12.1 However, the xmonad still stay as it is: %pacman -Q|grep -i xmonad xmonad 0.9-2.1 xmonad-contrib 0.9-1.2 %pacman -Ql xmonad|more ...... xmonad /usr/lib/ghc-6.10.4/ xmonad /usr/lib/ghc-6.10.4/site-local/ xmonad /usr/lib/ghc-6.10.4/site-local/xmonad-0.9/ xmonad /usr/lib/ghc-6.10.4/site-local/xmonad-0.9/HSxmonad-0.9.o xmonad /usr/lib/ghc-6.10.4/site-local/xmonad-0.9/XMonad.hi xmonad /usr/lib/ghc-6.10.4/site-local/xmonad-0.9/XMonad/ ..... So it will fail when running `xmonad --recompile` as xmonad cannot find its library in /usr/lib/ghc-6.12.1. Do i need to re-install xmonad ? Sincerely! ----- fac n = foldr (*) 1 [1..n] -- View this message in context: http://old.nabble.com/xmonad-cannot-find-its-library-after-upgrading-GHC-tp… Sent from the Haskell - Haskell-Cafe mailing list archive at Nabble.com.

2 1

Re: [Haskell-cafe] Re: Arrow instance of Transducer (Was: [Haskell] ANN: Streaming Component Combinators 0.4)
by Mario Blažević 17 Jan '10

17 Jan '10

On Sun 17/01/10 4:47 PM , Twan van Laarhoven twanvl(a)gmail.com sent: > The Arrow class is too big, it includes too many things and it should be > split up. This is yet another example. You can get some of the benefits of > standard classes by making Transducer an instance of Control.Category.Category, for > which you only need to implement (.) and id. Yes, perhaps I'll do that. Is there any relatively accepted package that defines PointedCategory?

1 0

Re: Arrow instance of Transducer (Was: [Haskell] ANN: Streaming Component Combinators 0.4)
by Mario Blažević 17 Jan '10

17 Jan '10

>>> Stupid question: Is it related to Arrows? >> >> Not really. You might say it's more general than arrows, because >> the streaming components are not restricted to a single input and >> single output type. On the other hand, they are all specific to stream >> processing, much like Fudgets and Yampa which are arrow-based. > > Arrows use tuple values for multiple inputs and outputs. Actually I'm > using arrows this way for signal processing. > >> I suppose the Transducer data type [1] could be made an Arrow >> instance, which would let me rename the 'connect' combinator [2] to >> (>>>). I'll have a look at Yampa to see if I can harvest more >> combinator names, thank you! After some investigation, I've concluded that Transducer cannot be made an instance of Arrow in any way that's both practical and general. The reason is that a Transducer converts a finite stream to another finite stream of arbitrary length. If we ignore effects, it's like a function of type [a] -> [b]. When trying to define an instance of Arrow for > newtype Transducer a b = Transducer ([a] -> [b]) the arr and (>>>) methods are trivial: > instance Arrow Transducer where > arr f = Transducer (map f) > Transducer t1 >>> Transducer t2 = Transducer (t2 . t1) but there is no satisfactory definition for the method first. A sensible definition, IMO, would have to give rise to a pairwise (***), so if we have two transducers t1 and t2 which respectively convert [a,b] to [c,d] and [1,2] to [3,4], t1 *** t2 would have to map [(a,1), (b,2)] to [(c,3), (d,4)]. There is no general way to define such a method first for the Transducer data type above. One can try appending the right-hand sides of the input to a queue and adding them to the outputs of the argument transducer: > first (Transducer t) = Transducer (uncurry (zip . t) . unzip) but this works only for injective transducers that map each new input item to exactly one output item. If the input stream has finite length, the input and output stream must be of equal length. In his paper "Generalising Monads to Arrows" from 1998, John Hughes defines a similar Arrow instance for stream processors operating on infinite streams which theoretically works for non-injective transforms, but leaks infinite space and can't be adopted to finite stream processors. HXT defines a different kind of an Arrow instance similar to the list monad instance. This approach can combine non-injective transforms, but in it the t1 *** t2 example above would map the input [(a,1), (b,2)] to [(c,3), (c,4), (d,3), (d,4)]. While this is a simple and lawful Arrow instance, I don't find it useful for stream processing. Yampa and other Functional Reactive Programming libraries associate a unique timestamp to each stream item. Method first can use this timestamp to associate input and output items even if the transducer leaves out some output items, and if the time stamps are ordered in each stream I guess it could also handle extra output items unrelated to any input. I'm not certain to what extent the FRP transducers (i.e., signal processors) are allowed to do this. To summarize: while I'd love the benefits of satisifying an Arrow instance, the loss of generality seems too high for me. SCC transducers work on finite streams of any values, and their output stream need not have any resemblance to the input stream. I may define an Arrow-conformant subclass of Transducer in a future version of SCC, perhaps operating on streams with timestamped values. >> >> [1] >> http://hackage.haskell.org/packages/archive/scc/0.4/doc/html/Control-Concur… >> >> [2] >> http://hackage.haskell.org/packages/archive/scc/0.4/doc/html/Control-Concur… > > In whatever way it is related to arrows, maybe you can mention it on the > project page. I'll copy this message there.

2 1