December 2010 - Beginners

Linking modules
by Russ Abbott 11 Dec '10

11 Dec '10

I have two modules Structure and Component. Module Structure defines data Structure = Component ... Module Component defines data Component = ... I would like Structure to import Component. But Component includes functions that take a Structure as an argument. So I have (or would like) something like this organization. Module Structure where import Component data Structure = Component ... Module Component where data Component = ... f :: Structure -> Component ->Structure The functions in Component are really very Component related and should not be moved to Structure. So how can I set up this circular relationship? I thought that one approach would be to do something like the following. But when I try I get all fouled up. Help would be appreciated. Module Structure where class Comp data Structure = Comp ... Module Component where import Structure data Component = ... instance Comp Component If I do this I get a complaint that class Comp needs a type parameter. If I give it an artificial one, that leads to all sorts of other problems. Is there a better way? Thanks. * -- Russ *

3 2

Using derived Show most of the time
by Russ Abbott 11 Dec '10

11 Dec '10

I have a data type for which I want to generate my own show string in a few cases, e.g., *data *T = T1 ... | T2 ... | T3 ... ... I'd like to do something like this, *instance *Show T *where* show T5 ... = <something special> show t = showTheDerived t Is there a way to do something like that simply? * -- Russ *

2 1

Fwd: Functional Dependencies
by dan portin 11 Dec '10

11 Dec '10

Hi, *unflt* was a poor choice of name. *flt2*, which flattens a datatype containing a flattenable datatype, would have been a better name. > > If I was making a flatten class, I wouldn't want Monad as a constraint > on the result container. Monad doesn't really say anything about > "containers" - there are some containers that are Monads, but there > are monads that aren't containers and there are containers that aren't > monads. > You're right. My initial class was class Monoid m => t -> m where flt :: t -> m instance (Tree a) [a] where flt = fltTree instance (Maybe a) [a] where flt = maybeToList for some tree type like `data Tree a = Nil | Node a (Tree a) (Tree a)` (for simplicity). The problem is that I couldn't figure out how to type the "flatten twice" function, which required a declaration like flt2 :: (F t' m) => t' t -> m which is clearly incorrect (since it gives rise to a kind mismatch). "Monad" and "Functor" were introduced just to get the definition of *unflt* to work, and was unsatisfactory. I need to be able to assert that *flt2* is a nested flattenable datatype, and that the return type of the nested and containing datatype is the same monoid. I can't figure out how to type this, however.

3 2

About 'Curry'
by 吴兴博 11 Dec '10

11 Dec '10

pl a b = a ++ b pl a = a ++ The second definition cannot be accept by neigher ghci nor ghc. What's the problem? (ghc 6.12.3) -- ---------------- WU XingBo

3 2

state and randomness, Rand, RandT
by Amy de Buitléir 10 Dec '10

10 Dec '10

I'm experimenting with different ways of working with both state and randomness at the same time. I put together this simple little example as a starting point, where the state in question is a list of Ints, and I want to be able to push a random number onto the stack. {-# LANGUAGE PackageImports #-} import "mtl" Control.Monad.State import Control.Monad.Random type Stack = [Int] -- push a random number onto the stack pushRandom:: State (Stack, StdGen) () pushRandom = do (xs,gen) <- get (r,gen') <- return $ randomR (1,100) gen put(r:xs,gen') That implementation works just fine. GHCi> execState pushRandom ([5,4,3,2,1], mkStdGen 0) ([46,5,4,3,2,1],40014 40692) But... I think it would be nicer to use Rand. Unfortunately, I can't figure out how to write the correct type signature. (Compiler message is: `Rand g' is not applied to enough type arguments Expected kind `*', but `Rand g' has kind `* -> *'). pushRandom2 :: (RandomGen g) => State (Stack, Rand g) () pushRandom2 = do (xs,gen) <- get (x,gen') <- getRandomR (0,100) gen put (1:xs,gen) And I'd really like to try using RandT, because this seems like the right situation for it. This compiles just fine. pushRandom3 :: (RandomGen g) => RandT g (State Stack) () pushRandom3 = do xs <- get r <- getRandomR (1,100) put (r:xs) But I don't know if it works, because I can't figure out the magic incantation to get it to run! GHCi> evalRandT $ pushRandom3 (mkStdGen 0) [5,4,3,2,1] <interactive>:1:12: Couldn't match expected type `StdGen -> [t] -> RandT g m a' against inferred type `RandT g1 (State Stack) ()' In the second argument of `($)', namely `pushRandom3 (mkStdGen 0) [5, 4, 3, 2, ....]' In the expression: evalRandT $ pushRandom3 (mkStdGen 0) [5, 4, 3, 2, ....] In the definition of `it': it = evalRandT $ pushRandom3 (mkStdGen 0) [5, 4, 3, ....] To summarise: Q1: How can I fix pushRandom2? Q2: How can I run pushRandom3? Thank you in advance for anyone who can help. Amy

2 2

Relying on memoization
by Russ Abbott 10 Dec '10

10 Dec '10

(Is that even the right term?) Let's suppose I have a fairly complex computation to perform on a data structure. Normally I would cache the computed value in a field of the structure. The next time I need it I don't have to compute it again. (Of course if the structure changes, I have to recompute the value and store it again.) In Haskell is it the case that caching of this sort is redundant since Haskell does it for me? That is, if I call f someStructure multiple times at different places in my code and if both "f" and "someStructure" refer to the same things each time, can I rely on Haskell not to perform the computation multiple times but to look up the result it previously computed? Is there some limit to that? If every computation is stored, doesn't that create a very large collection of stored results? * -- Russ *

2 3

Functional Dependencies
by dan portin 10 Dec '10

10 Dec '10

I was trying to make a class of "flattenable" data structures in Haskell. My first thought was to use multi-parameter type classes to assert a relation between two data types, one which is flattened and another which holds flattenable elements. Such a class is the class *F* below without the functional dependency. When given an instance declaration like *F Tree []*, applying *flt* to a structure fails with messages like *no instance for `F Tree m`*. I ended up writing the following (where *foldT* is fold over trees, etc.), in order to avoid said errors: class (Functor m, Monad m) => F t m | t -> m where flt :: t a -> m a unflt :: F t' m => t' (t a) -> m a unflt = join . fmap flt . flt instance F Tree [] where flt = foldT (\a b c -> a : b ++ c) [] instance F Maybe [] where flt = maybeToList Now, I would like to have *F* be a relation between a flattenable data type and, for instance, a monoid in which I can accumulate the flattened data type. Two things: I can't get a monoid constraint to operate properly over the variable *m* in *unflt*. This fails with messages like *no instance for `Monoid Tree [Integer]`*, and so on. I'm not sure why. Secondly, I would like to be able to have *flt* be able to return different "container structures", e.g., a custom sequence or a list. But the functional dependency I introduced prohibits this. How would you handle this problem? Thanks.

2 1

Disjunction
by Russ Abbott 09 Dec '10

09 Dec '10

I recently was writing code long these lines. f x = p1 x || p2 x || ... || pn x I would have liked to write that in pointfree form but couldn't find a library function to do it. I created this. disj :: [a -> Bool] -> a -> Bool disj ps = (\x -> or $ map ($x) ps) disj2 p1 p2 = disj [p1, p2] disj3 p1 p2 p3 = disj [p1, p2] disj4 p1 p2 p3 p4 = disj [p1, p2] ... With that I can write my function as follows. f = disj pi p2 p3 ... pn Is there a standard way to do this? * -- Russ*

4 4

DPH help
by Luca Ciciriello 09 Dec '10

09 Dec '10

Hi All. I'm trying to load in GHCi the following code: {-# LANGUAGE Arrows #-} {-# LANGUAGE PArr, ParallelListComp #-} module Test where import Control.Arrow import Control.Parallel dotp :: Num a => [:a:] -> [:a:] -> a dotp xs ys = sumP [:x * y | x <- xs | y <- ys:] ...... Then I get the error: Not in scope 'sumP'. What I have to import to avoid this error? Where is defined sumP? Thanks in advance for any answer. Luca

2 1

A newbie question
by Mianlai Zhou 08 Dec '10

08 Dec '10

Hello All, Today I wrote my first Haskell program, which is module Main where f 1 = 1 f n = n*f(n-1) main = putStrLn (show (f 5)) After compilation, this program runs with no problem and gives 120 (which is correct). I have two questions though: 1. What does "where" in line 2 mean? Why I should put it here in this place? If there is no "where" here, the program cannot run. But what is the syntactical meaning of "where"? 2. Why this module can end without an "end" sentence? And what are the blank lines treated as? Thank you for your answer, well, in advance. These are newbie questions indeed. Best wishes, Mr. Zhou

2 2