September 2009 - Haskell-Cafe

Hopefully simple monad question
by Gregory Propf 16 Sep '09

16 Sep '09

I'm playing around with a little program that implements a simple virtual machine. I want to use a monad to represent machine state. I created a data type for the machine (VM) and a monadic type for the monadic computations using it. I declared this an instance of MonadState and Monad and created the usual operators. That stuff works. My issue is that I want to run some functions in the machine monad, call it VMS - "virtual machine w/state" and then pull the underlying VM data structure out and print it. I've read about monad transformers, lift, liftM, liftIO and all these instances in the libraries like MonadIO and am rather confused. The most sensible conclusion I can reach is that I probably need to create my own Transformer monad and define liftIO. Is this where I need to go? Also, my VMS monad doesn't really do anything different from the State monad except explicitly specify that state is a VM and not a generic type. Am I doing too much work creating my own instances here? Would a simple "type" statement work?

4 4

ANNOUNCE: levmar-0.2
by Bas van Dijk 16 Sep '09

16 Sep '09

Hello, We like to announce a new release of the high-level Levenberg-Marquardt library levmar: http://hackage.haskell.org/package/levmar-0.2 Changes: * There's one new major feature: automatic calculation of the Jacobian using automatic differentiation with Conal Elliott's vector-space library. (Conal: would you like to take a quick look at the implementation and see if I'm using vector-space the way it supposed to be used, thanks) * Furthermore we made the type Model from LevMar more type-safe by replacing: type Model m r = NFunction m r [r] with: type Model m n r = NFunction m r (SizedList n r) * Finally we did lot's of internal refactorings and updated the documentation. regards, Bas and Roel van Dijk

1 0

ANNOUNCE: bindings-levmar-0.1.1
by Roel van Dijk 16 Sep '09

16 Sep '09

I am pleased to announce the release of bindings-levmar-0.1.1: http://hackage.haskell.org/package/bindings-levmar The most important change compared to the previous version is a custom configure script (copied from hmatrix) that detects which libraries are needed. The cabal file also has some flags which allow you to choose for instance intel's mkl library instead of the normal lapack (also copied from hmatrix). Happy hacking, Roel and Bas van Dijk

1 0

Re: retrieving arguments for functions from a heterogenous list
by oleg＠okmij.org 16 Sep '09

16 Sep '09

Andrew U. Frank wrote: > I have a number of functions which have some arguments and produce a single > result. all the arguments are in a heterogenous list and the results should > update the list. If I understand the problem correctly, you have a typed-indexed collection TIP and you would like to apply a transformation function to it, whose argument types and the result type are in the TIP. The function should locate its arguments based on their types, and update the TIP with the result. The function may have any number of arguments, including zero; the order of arguments should not matter. Then the following code seems to solve the problem. It starts with the examples. {-# LANGUAGE MultiParamTypeClasses, FlexibleInstances #-} {-# LANGUAGE ScopedTypeVariables, UndecidableInstances #-} -- Transforming a TIP: applying to a TIP a (polyvariadic) function -- that takes arguments from a TIP and updates the TIP with the result module TIPTransform where import Data.HList import Data.HList.TIP import Data.HList.TypeEqGeneric1 import Data.HList.TypeCastGeneric1 import Data.HList.Label5 newtype MyVal = MyVal Int deriving Show -- A sample TIP tip1 = MyVal 20 .*. (1::Int) .*. True .*. emptyTIP -- TIP (HCons (MyVal 20) (HCons 1 (HCons True HNil))) -- Update the Int component of tip1 to 2. The Int component must -- exist. Otherwise, it is a type error tip2 = ttip (2::Int) tip1 -- TIP (HCons (MyVal 20) (HCons 2 (HCons True HNil))) -- Negate the boolean component of tip1 tip3 = ttip not tip1 -- TIP (HCons (MyVal 20) (HCons 1 (HCons False HNil))) -- Update the Int component from the values of two other components tip4 = ttip (\(MyVal x) y -> x+y) tip1 -- TIP (HCons (MyVal 20) (HCons 21 (HCons True HNil))) -- Update the MyVal component from the values of three other components tip5 = ttip (\b (MyVal x) y -> MyVal $ if b then x+y else 0) tip1 -- TIP (HCons (MyVal 21) (HCons 1 (HCons True HNil))) -- The same but with the permuted argument order. -- The order of arguments is immaterial: the values will be looked up using -- their types tip5' = ttip (\b y (MyVal x)-> MyVal $ if b then x+y else 0) tip1 -- TIP (HCons (MyVal 21) (HCons 1 (HCons True HNil))) -- The implementation class TransTIP op db where ttip :: op -> db -> db instance (HMember op db b, TransTIP' b op (TIP db)) => TransTIP op (TIP db) where ttip = ttip' (undefined::b) class TransTIP' b op db where ttip' :: b -> op -> db -> db -- If op is found in a TIP, update the TIP with op instance (HTypeIndexed db, HUpdateAtHNat n op db db, HType2HNat op db n) => TransTIP' HTrue op (TIP db) where ttip' _ = tipyUpdate -- If op is not found in a TIP, it must be a function. Look up -- its argument in a TIP and recur. instance (HOccurs arg db, TransTIP op db) => TransTIP' HFalse (arg -> op) db where ttip' _ f db = ttip (f (hOccurs db)) db

1 0

(weird stuff) Kernel Modules in Haskell ;-)
by Matthias Kilian 16 Sep '09

16 Sep '09

A fellow openbsd developer told me the URL below... I hope this hasn't been posted on this list already (at least I didn't find it in my local archives): http://tommd.wordpress.com/2009/09/13/kernel-modules-in-haskell/ Ciao, Kili -- Logging should be in debug mode only. If every network driver did a dmesg printf everytime they came up, would that be a better world? -- Theo de Raadt

2 1

adding state in GUIs (qtHaskell)
by Michael P Mossey 16 Sep '09

16 Sep '09

I'm trying to learn qtHaskell. I realize few people on this list know anything about qtHaskell, but I have a question that probably relates to all GUIs as implemented in Haskell. I just need a hint that could help me figure out the next step, which I might be able to infer from the qtHaskell API. I don't think is any tutorial-type or step-by-step type documentation for qtHaskell. I have sent some questions to the author of qtHaskell, David Harley, but he hasn't responded yet, and anyway I don't want to trouble him every time I have a question, so I'm trying to infer as much as I can. The problem relates to state. In Qt, one adds state to a widget by subclassing it and adding new member variables. For example, I want to create a widget that responds to keypresses and remembers what keypresses have taken place. I'm totally stuck on this part, because Haskell doesn't have state. There must be some kind of Haskell call that adds state to a widget, but it is hard to figure out from the qtHaskell examples David provides. Thanks, Mike

8 9

loading object code with ghci
by Ryan Wisnesky 15 Sep '09

15 Sep '09

Hello, I'm having some trouble correctly linking to externally generated object code when using ghci. I'm loading a Cabal package Foo, that has already been successfully installed: >ghci -package Foo ghc-6.8.3: unknown symbol `_iceExePath' Loading package Foo-0.1 ... linking ... ghc-6.8.3: unable to load package `Foo-0.1' I have a file Bar.o which contains the definition of _iceExePath: >nm ./Bar.o ... U _ProcessInformationCopyDictionary 000000a0 b _exepath 00000000 T _iceExePath U _kCFBundleExecutableKey ... The documentation at http://www.haskell.org/ghc/docs/6.8.3/html/users_guide/ghci-invocation.html… says that: "GHCi can also load plain object files (.o or .obj depending on your platform) from the command-line. Just add the name the object file to the command line." So, I've tried both of these command lines: > ghci ./Bar.o -package Foo > ghci -package Foo ./Bar.o but ghci still can't find _iceExePath, and the same error occurs. I'm using GHC 6.8.3. I'm not sure how to proceed and would appreciate any advice. Thanks for your help, Ryan Wisnesky

1 0

What does it mean that objects are fixpoints? (OO'Haskell)
by Manuel Simoni 15 Sep '09

15 Sep '09

Hello! I'm trying to wrap my head around OO'Haskell's notion of objects as fixpoints. Is OO'Haskell's use of mfix simply a use of something like a monadic Y-combinator to give the object access to its own "identity"? Thanks, Manuel

3 2

Announce: EnumMap-0.0.1
by John Van Enk 15 Sep '09

15 Sep '09

Hi List, I've uploaded a first version of EnumMap to hackage. EnumMap is a generalization of IntMap that constrains the key to Enum rather than forcing it to be Int. I have no idea what impact this has on performance, but it still passes all the tests that ship with IntMap. (My guess is that performance will be similar/identical unless I've missed something.) If this package competes in speed (especially with an Int key) with IntMap, I'd expect this could be merged into the containers package as an IntMap replacement. The source comments/copyright/license hasn't been changed since this is a derivative work of the original IntMap package. http://hackage.haskell.org/package/EnumMap Comments/benchmarks/criticism is appreciated. Merging this in with containers eventually would be more appreciated. :) John Van Enk

7 29

Typeclasses vs simple functions?
by Olex P 15 Sep '09

15 Sep '09

Hey guys, It's a dumb question but I'd like to know a right answer... Let's say we have some geometry data that can be Sphere, Cylinder, Circle and so on. We can implement it as new data type plus a bunch of functions that work on this data: data Geometry = Sphere Position Radius | Cylinder Position Radius Height | Circle Position Radius deriving (Show) perimeter (Sphere _ r) = 0.0 perimeter (Cylinder _ r h) = 0.0 perimeter (Circle _ r) = 2.0 * pi * r Perimeter doesn't make sense for Sphere or Cylinder. So we could define a type class for objects that have perimeter and make an instance of it only for Circle (data Circle = Circle Position Radius). Make sense. But these three functions above have desired behaviour. If user has a list of objects like [Sphere, Circle, Circle, Cylinder] he would like to calculate perimeters of each object using map perimerer list (in this case we also have to modify Geometry data type). So we could make instances of "perimeter" type class for all objects and return zero in case if perimeter doesn't make sense. Same as previous version but with typeclasses and with additional constructors (constructors for each type of object + constructors in Geometry data). Looks a bit overcomplicated. Any reasons to use type classes in this case? Maybe there is something I'm missing? Cheers, -O

3 6