July 2009 - Haskell-Cafe

Cont, ContT and IO()
by Günther Schmidt 04 Jul '09

04 Jul '09

Hi, I've got an IO action, some file system IO, traversing one level only and iterating over files found. I wish to build in an "early" exit, ie. if an IO action in the loop encounters a particular value I want it to abort the loop. Now so far, pls don't shoot, I have done this by throwing IO Exceptions and catching them. I'm trying to rewrite this using Continuatios / callCC but can't figure out where to place what. I certainly don't have the intuition yet and funny enough not even in RWH I could find some Cont/ContT examples. Would someone please draw me an example? Günther

7 12

Half-integer
by Andrew Coppin 04 Jul '09

04 Jul '09

I just wrote a small module for dealing with half-integers. (That is, any number I/2 where I is an integer. Note that the set of integers is a subset of this; Wikipedia seems to reserve "half-integer" for such numbers that are *not* integers.) module HalfInteger where data HalfInteger i instance (Eq i) => Eq (HalfInteger i) instance (Ord i) => Ord (HalfInteger i) instance (Integral i) => Show (HalfInteger i) instance (Integral i) => Num (HalfInteger i) half :: (Num i) => HalfInteger i fromNum :: (Integral i, RealFrac x) => x -> HalfInteger i toNum :: (Integral i, Fractional x) => HalfInteger i -> x isInteger :: (Integral i) => HalfInteger i -> Bool Note carefully that the set of half-integers is *not* closed under multiplication! This means that for certain arguments, there are two reasonable products that could be returned. (E.g., 1/2 * 1/2 = 1/4, so 0 or 1/2 would be a reasonable rounding.) I haven't put a lot of effort into the rounding details of (*) or fromNum; which answer you get is kind of arbitrary. (However, addition and subtraction are exact, and for multiplications where an exact result is possible, you will get that result.) The Show instance outputs strings such as fromInteger 5 fromInteger 5 + half fromInteger (-5) - half depending on the isInteger predicate. Now, the question is... Is this useful enough to be worth putting on Hackage?

10 29

ORM for haskell?
by Marc Weber 04 Jul '09

04 Jul '09

Some time ago I stumbled upon SQLAlchemy which is a great ORM wrapper library for python. It has a nice syntax I'd like to see in a haskell library as well. SQLAlchemy already provides some lazy features such as loading subitems on access etc. All haskell SQL libraries I know only let you run SQL statements but not much more. To start real business you no longer want to write many SQL commands. Example why it matters: schools - 1:n - teachers - 1:n - pupils If you want to list all schools which have a pupil with age > 3 you'd write an sql query like this: SELECT dictinct * FROM schools as s JOIN teachers t ON (t.school_id = s.id) JOIN pupils as p ON (p.teacher_id = t.id) WHERE p.age > 3 in SQLAlchemy it looks like this: session.query(School).join(School.teachers).join(Teacher.pupils).filter(Pupil.age > 3).all() difference? Because SQLAlchemy knows about the relations you don't have to remember alias names. So there is no chance to get that wrong. Another example: Updating the age of a pupil: row = SELECT * FROM pupils where age = 13; UPDATE pupils SET age = 14 WHERE id = <the id you got above> p = session.query(Pupil).filter(Pupil.age==13).one().age=14 session.commit() difference? You don't have to care about ids. you just assign a new value and tell the engine that it should commit. So again less chances to get something wrong. What about trees (eg web site navigation) id | title | parent_id 1 | top | null 2 | submenu | 1 3 | submenu2 | 1 should result in top - submenu - submenu2 using SQLAlchemy you can just do parent = session.query('nodes').filter(Node.id = 1) def print(node): print node.title print node.subnodes # this will run a subquery automatically for you returning submenu{,2} Again no sql. No chance to get something wrong? You can skim the manual to get a better idea how SQLAlchemy works http://www.sqlalchemy.org/docs/05/sqlalchemy_0_5_5.pdf I have to admit that I haven't used SQLAlchemy in a real project yet. However I can imagine doing so. Comparing this to what we have on hackage I'd say some work has to be done to get close to SQLAlchemy. The backend doesn't have to be a relational database. However I'd like to use this kind of abstraction in haskell. Is there anyone interested in helping building a library which a) let's you define kind of model of you data b) let's you store you model in any backend (maybe a relational database) c) does static checking of your queries at compilation time? Right now I'd say the best way to go is define the model in the application and not get the scheme from an existing database because there is not way to store all scheme details within a relational model. I think SQLAlchemy does it right by providing a way to define the model in python. Of course haskell doesn't have "objects" to store. But GADTs could be stored (data Foo = ...) So are there any volunteers who are interested in helping writing this kind of storage solution for haskell which could be used in real world business apps? Maybe this does already exist and I've missed it? Marc Weber

9 13

space leak hints?
by Uwe Hollerbach 04 Jul '09

04 Jul '09

Good evening, all, I wonder if I could tap your collective wisdom regarding space leaks? I've been messing about with haskeem, my little scheme interpreter, and I decided to see if I could make it run reasonably space-efficiently. So far... no. Here's what I tried: I wrote a tiny scheme program to compute Collatz sequences for successive numbers, starting from 1 and incrementing forever (well, in principle). Because real scheme implementations are fully tail-call-optimized, this'll run in constant memory; I checked that with mzscheme, and it does indeed work. With my little interpreter, that's not the case: memory usage grows continually, although apparently less-than-linearly. I've built the interpreter with the profiling stuff described on the wiki and in RWH Ch 25 turned on and have made a few runs with that; I stuck the postscript plot that's the result of one of those runs onto my web site at http://www.korgwal.com/haskeem/run2.ps. The full source to the interpreter is a little large to paste into this message; it's available on my web site, and also on hackage. But according to the plot, there appear to be three main memory allocation issues, and they seem to all be related, if I'm reading stuff correctly. The core of the interpreter is a function, evalLisp, which evaluates scheme forms. There are of course a fair number of different forms, but the largest generic usage is "evaluate each of a list of forms, returning the value of the last of them as the overall result". In order to express that in a couple of different places, and to accomodate the possibility of an empty list, I have a really tiny function lastOrNil which just calls "last" on a (haskell) list, checking for the possibility of an empty list, and returning a haskeem LispVal object: > lastOrNil = liftM lON > where lON [] = List [] > lON l = last l (sorry, proportional fonting may be throwing this off). It's this function which is directly implicated in two of the top three memory pigs, and nearly directly in the third one as well. If I could eliminate the memory growth in these three cost centers, I would already capture over 90% of the growth in this benchmark, which would make me very happy indeed. But I don't really understand what is going on here. It seems entirely plausible, indeed likely, that the list which I'm traversing there is not fully evaluated. So I've tried adding 'seq' to this function. Uhhh... from memory, I dumped it after it didn't work, I had > lastOrNil = liftM lON > where lON [] = List [] > lON (l:[]) = l > lON (l:ls) = seq l (lON ls) (again, proportional fonting might mess me up here.) As I said, I dumped this after it made no difference whatsoever. I also tried bang-patterns in a couple of places, strictness annotation of my basic LispVal types... nothing. It all made exactly no difference, as far as I could tell. I've tried a couple of google searches, but haven't come up with anything better than what I've already described. So I'm a stumped chump! I'd be grateful for any suggestions... thanks in advance! Uwe

1 0

How to present the commonness of some objects?
by Magicloud Magiclouds 03 Jul '09

03 Jul '09

Hi, I thought "class" was for this purpose. But it turns out not. Code as following could not compiled. 1 main = do 2 mapM_ (\(x, y, widget) -> do 3 a <- widgetRun widget 4 putStrLn $ show a 5 ) widgetList 6 7 widgetList :: (Widget w) => [(Integer, Integer, w)] 8 widgetList = [] 9 10 class Widget w where 11 widgetRun :: w -> IO () --- % ghc --make tmp/test.hs [1 of 1] Compiling Main ( tmp/test.hs, /tmp/Main.o ) tmp/test.hs:3:16: Ambiguous type variable `t' in the constraint: `Widget t' arising from a use of `widgetRun' at tmp/test.hs:3:16-31 Probable fix: add a type signature that fixes these type variable(s) -- 竹密岂妨流水过山高哪阻野云飞

6 8

Help needed to pick a better name then "hack"
by Jinjing Wang 03 Jul '09

03 Jul '09

Hack is such an inconvenient name for a package, may I get some inspiration from renaming it? Also, is there an idiom to use when upgrading package name on hackage? Best, -- jinjing

3 2

On the parallel between java annotations and monads
by Rafael Almeida 03 Jul '09

03 Jul '09

Hello, After an extensive search (5 minutes googling) I could not find any comparison between Java annotations (or Python annotations for that matter) and Monads. I think they are similar in various aspects and I want to discuss them here. I'm sure several of you have experience with Java programming and its annotation system. At first they seem like an extension to the typing system, by using annotations a single variable or method may have more than one type: the real type and all the annotations types. The way you declare them is not important for this discussion, but assume that you've created the annotations GreaterThanFive and LessThanTwenty, you could annotate a method like this (let's forget about all the class verbosity for a while): @GreaterThanFive @LessThanTwenty int doYourThing(); You could than use reflection to read those annotations and you could check if the returned integer is in the correct range. You could even make the compiler generate code that will do that checking automatically for you. Even though that was my first impression of them, that's not really how they are used in practice (at least the practice that I have experimented). People use them to implement Monad-like features to the language. Actually, you can forget monad and think of computation containers. On an EJB environment there's something called a container and it will do all sorts of computations when you call an annotated method or when you create a new annotated class. For instance, if, inside of an entity bean, you annotate a variable like this: @Column(name = "NAME") private String name; then, when you create a method like so: public String getName() { return name; } that name is actually something retrieved from the database, even though, it looks just like something in memory. As far as I know the interaction with monads for handling databases is not very different. There are other annotations which allows the container to "inject" some object on a variable, such as: @PersistenceContext private EntityManager entityManager; now you won't even create an EntityManager instance, but will use the object living inside the container. I compare that to getting a value from an State Monad. Like there, you'll do some computations using entityManager, but it will ultimately be returned to the container. While each computational container in Haskell is very rigid and well defined, in java there's nothing rigorous about the annotation. However, quite often the annotations in Java are used in a manner similar to the implementation of a Monad in Java. []'s Rafael

3 2

hiding instance definitions in imported modules or redefining them, can it be done?
by John Lask 03 Jul '09

03 Jul '09

Is there a way to hide instance definitions when importing a module for instance I want to import Data.Monoid but wish to redefine the instance of Monoid for ( a->b) I don't suppose this is possible? Specifically, I wish to define a specialised instance of Monoid for String->Int, but it seems overlapping instances will not give me what I want, as this option seems to need to be applied at the definition site (aka Data.Monoid module, looking for confirmation of this?). The easy fix is just to do newtype M a b = M (a->b) etc..., but I was hopping to avoid this. I've encountered this issue several times. Why can't we redefine instances? I can't of thought that there are any seriouse technical issues in doing so? If so what are they? jvl

2 1

Monad Input/Output and Monad Transformers
by Maciej Piechotka 03 Jul '09

03 Jul '09

1. Learning haskell I discovered that I/O should be avoided nearly 'at all costs'. The problem is that the IO monad is the only one which have more interactive work flow. There is Reader/Writer monad but in fact AFAIU first one is about the environment and second one is about logging. Implementation of Writer can be defined in terms of handles but it is rather limited (either non-full implementation which may be confusing or need of caching the result for pass etc.). I searched the hackage but I didn't find package with pure I/O. Such package may look like: class (Monad m, Monoid v) => MonadInput v m where -- | Gets an element from input (line of text [with \n], 4096 bytes, -- or something like that). mzero on end getChunk :: m v class (Monad m, Monoid v) => MonadOutput v m where -- | Puts an element putChunk :: v -> m () In similar way filters (for example buffered input) can be defined: class (MonadInput v m) => MonadBufferedInput m where -- | If not whole chunk has been consumed at once (for example only -- first 3 elements from list) rest can be returned. It will be -- returned as part of the input on next getChunk call. returnChunk :: v -> m () data (MonadInput v m, Monoid v) => BufferedInputT v m a = BufferedInputT a v Also pipes may be defined (as far as I understand but I'm not 100% sure) which probably will simplify the writing of network tests: -- | Evaluates the first argument. If the getChunk is called the -- evaluation is passed to second argument until the putChunk is -- called, which argument is returned in the first argument callPipeT :: (Monad m, Monoid v) => PipeInputT v m a -> PipeOutputT v m b -> m (a, b) I've started some tests but I'd be grateful for comments (well - probably I'm not the first who come to this idea so a) there is such package or b) my level of Haskell does not allow me to see the problems). 2. I find writing monad transformers annoying. Additionally if package defines transformer A and another transformer B they need to be connected 'by hand'. I find a simple solution which probably is incorrect as it hasn't been used: instance (MonadState s n, Monad (m n), MonadTrans m) => MonadState s (m n) where get = lift get put = lift . put (requires FlexibleInstances MultiParamTypeClasses FlexibleContexts UndecidableInstances - two last are not extensions used by mtl) Regards

6 8

How to declare a Typeless Function
by Fernan Bolando 03 Jul '09

03 Jul '09

Hi I have a function that swaps rows of an array of double swap :: Array (Int,Int) Double -> [Int] -> Array (Int,Int) Double I then create a function that swaps rows of arrays of Complex Double swap :: Array (Int, Int) (Complex Double) -> [Int] -> Array (Int, Int) (Complex Double) In reality the function swap does not care whether its working on a double or a complex number. how do I declare swap so that it will work whether it's a complex or a double array. I tried googling but I wasn't sure what to google. fernan -- http://www.fernski.com

4 3