- Haskell-Cafe - Haskell.org

Surprising bugs
by Mark Carroll 12 May '04

12 May '04

I clearly don't understand Haskell very deeply yet because I dealt with a couple of interesting types of bug this week. One sort was where, if I have, f :: SomeType -> Stuff ... f = whatever g :: Stuff ... g = f someValue ...then I can get an error that suggests that maybe I'm violating the monomorphism restriction if I put in f's type but not g's. If I put both in, it becomes happy. Another sort was where, if I had an algebraic parametric type (is that the name? Ord a => Foo a b c, etc.) then I couldn't have, say, a Maybe (Foo a b c), where sometimes I called it with a (Just foo) and sometimes with a Nothing, because with the Nothings it would complain about, erm, some ambiguity to do with Ord a. However, it would be fixed if I made the Maybe X argument into two arguments, Bool -> X, the Bool indicating if the X was meant to have been a Nothing or a Just X. I hope both (or either!) of those made sense. I don't have the code immediately to hand to reproduce the details, but I just thought I'd try to recall what they were because I found these compiler complaints interesting. I don't fully understand what's happening, and I'm sure a newbie would be quite bewildered. (-: -- Mark

1 0

Monadic Composition
by Jorge Adriano Aires 12 May '04

12 May '04

Hello, I needed to compose monadic functions, since I couldn't find anything in Control.Monad so I defined my own. I first came up with a left associative version: compLM [f1,..., fn] a = (f1 a >>= f2) >>= ... >>= fn > compLM :: Monad m => [a->m a] -> (a->m a) > compLM mfs a = foldl (>>=) (return a) mfs Later a right associative version, which I believe to be more efficient, particularly when some computation returns fail: compRM [f1,..., fn] a = \x->f1 x >>= (\x->f2 x >>= ... >>= \x->fn x) a > compRM :: Monad m => [a->m a] -> (a->m a) > compRM = foldr (\f g-> (\x ->f x >>= g)) return This higher order function seems quite useful to me and not trivial, so I expected it to be available. When this happens I wonder if I'm missing something. Am I complicating things? Is it actually available (or some other that does the trick) ? Or is it just me who finds this usefull enough to be in the libs? J.A.

2 2

Cyclic Dependencies
by Georg Martius 12 May '04

12 May '04

Hi, I have some problems with cyclic dependencies. Module Foo: data Foo = Foo { bar :: Bar } Module Bar: class BarClass b where barfun :: b -> Foo -> Foo data Bar = forall b . (BarClass b) => Bar b Obviously I have to import both modules mutually. My workaround was to add a further type variable to BarClass. class BarClass' b f where barfun :: b -> f -> f data Bar' f = forall b . (BarClass' b f) => Bar b This seams to be fine, but in my case it is a bit more compicated. I need a Read instance of Foo. Therefor I have to implement the Read instance for the existential type Bar by my own and need to know about the instanced of BarClass. Unfortunately, the instances of BarClass need Foo and I have a cycle again. Thanks, Georg

1 0

Visualising a type class hierarchy
by Andre Pang 12 May '04

12 May '04

Hey all, Does anyone know of a good way to visualise ("graph") a type class hierarchy, including one with multiple-parameter type classes? The traditional "inherits from" arrows used in graphs of object-oriented classes doesn't deal very well with type classes. A simple case, such as: class Super a class Super a => Sub a instance Super SuperType instance Super SubType instance Sub SubType works fine, but once you get into situations such as: class SuperMulti a b class SubMulti a b c | a b -> c instance SuperMulti a Foo => SubMulti a Bar c I haven't found elegant ways of visualising this at all: how do you draw arrows from SuperMulti to SubMulti, for instance? Has anyone else had a go at this? -- % Andre Pang : trust.in.love.to.save

1 0

Re: [Haskell-cafe] Toy application advice wanted
by John Hughes 10 May '04

10 May '04

> But Scheme lets you change things if you want to! (The Haskell > afficianados gasp). Let's face it...in most long-running applications, > data changes a lot during the lifetime of a program. So somehow, you have > to manage the repercussions of state changes. Haskell does this with > monads. (I'm still wrapping my brain around them, but...) Apparently > monads plus some syntactic sugar are like functions that return a value > and the next function, which in turn returns a value and the next > function, etc. so that referential transparency is maintained. This is a > heckuva lot of work to do (set! foo 'bar). I hope I'm not far off base > with this. Gosh, this sounds complicated! I think there's a much simpler way to see it. In almost any language, some expressions have side-effects, and some don't. Use expressions with side-effects carelessly, in the wrong place, and you can screw up. In Haskell, expressions with side-effects have a type that says so (IO <type>), and that in turn lets the type-checker restrict where you use them (hence the many questions "How do I get an <alpha> from an IO <alpha>?" -- the point is, you can't). So you avoid screwing up. Not particularly complicated, not particularly costly. writeIORef foo Bar :: IO () -- empty result, with a side-effect. OK, there's more to it than this, once you start using other monads than IO or ST, and especially once you start overloading code on the underlying monad. But as an initial understanding this is enough. John Hughes

1 0

Toy application advice wanted
by mikeb＠manor.org 09 May '04

09 May '04

Hi folks, I've got an interesting task this week for my job. (Note that this will undoubtably last for longer than a week). I'm evaluating several high-level languages as development vehicles for our next suite of applications. The languages I'm currently considering are Scheme, Erlang, and Haskell. Scheme and Haskell have fairly wide acceptance in their particular roles (Scheme as the "pretty" lisp, supporting higher-order functions and mostly strict [until you roll your own laziness with macros], and Haskell the absolute purest language with no side effects and fully lazy evaluation). Erlang has a really nice distributed computing model, and has a nice union of non- and side-effect qualities (although it's completely strict). The toy application I've "designed" for myself is a simple GUI-based app that can load a Sun .au format sound file, display its waveform in a window, perform some simple filtering, play the sound for the user, and then save the changed sound file back out to disk. If I get familiar enough with the respective environments I'd like to add zooming in/out and scrolling to the waveform display. This application should allow me to see all kinds of different characteristics of the language/implementation, such as performance (how fast can the filter run), I/O (how fast can I load/save data to disk), OS interaction (how easy can I "play" the audio), user acceptance (how easy is it to do GUI programming) and of course the software engineering goals of modularity, correctness, and simplicity. The end goal is for me to have three working programs, in three different languages, so that I can present to my boss the performance/software engineering characteristics of each system to choose our next development language. So I'm writing you to solicit help. I have an amortized four days (32 hours!!!) to implement this simple application in Haskell. Can anyone point me to example code, supporting libraries, tutorials, etc. that are in this area of development? If I have to dumb down some of my application in order to make a great discovery about a language, that's fine. I'm expecting some great differences between these three programs. Any advice/pointers/flames welcome. Thanks in advance. Mike -- Mike J. Bell This is all just my opinion. "Outside of a dog, a book is man's best friend. Inside it's too dark to read." mikeb(a)manor.org

12 14

HDirect, [unique], troubles again!
by Vincenzo aka Nick Name 06 May '04

06 May '04

In my quest for a fuse binding for Haskell, which I really need at the moment, I have the following definition working: module HSFuse { interface stat{}; typedef int getattrT([string] char *,stat); typedef struct fuseOps { [ref] getattrT * getattr; } fuseOps; void fuse_main(int argc,[in,string,size_is(argc)]char** argv,[ref] fuseOps * op); }; but when I substitute "ref" with "unique", because some operations can be null, I get the errors below. The problem is that in the generated hs file there is data FuseOps = FuseOps {getattr :: GetattrT} which completely ignores my type declaration! Did I do something wrong? Commands invoked and errors follow, thanks for attention: ihc -fhs-to-c -fuse-ints-everywhere -c HSFuse.idl ghc -package hdirect -fglasgow-exts HSFuse.hs -c HSFuse.hs:136: Couldn't match `GHC.Base.String -> Stat -> GHC.IOBase.IO GHC.Base.Int' against `GHC.Ptr.Ptr (GHC.Base.String -> Stat -> GHC.IOBase.IO GHC.Base.Int)' Expected type: GHC.Ptr.Ptr (GHC.Ptr.Ptr (GHC.Base.String -> Stat -> GHC.IOBase.IO GHC.Base.Int)) -> GHC.Ptr.Ptr (GHC.Base.String -> Stat -> GHC.IOBase.IO GHC.Base.Int) -> GHC.IOBase.IO () Inferred type: GHC.Ptr.Ptr (GHC.Ptr.Ptr (GHC.Base.String -> Stat -> GHC.IOBase.IO GHC.Base.Int)) -> (GHC.Base.String -> Stat -> GHC.IOBase.IO GHC.Base.Int) -> GHC.IOBase.IO () In the second argument of `HDirect.writeunique', namely `writeGetattrT' In the definition of `writeFuseOps': writeFuseOps ptr (FuseOps getattr) = let pf0 = ptr pf1 = HDirect.addNCastPtr pf0 0 in HDirect.writeunique (HDirect.allocBytes (GHC.Real.fromIntegral sizeofGetattrT)) writeGetattrT pf1 getattr

2 1

greencard and callbacks
by Vincenzo aka Nick Name 06 May '04

06 May '04

I realize that I myself might have already asked this question to some of you, but I am really confused now. Can I have some opinion on "what is the right way to implement callbacks from C to haskell"? Will I have to learn the FFI and stop it, or is there some tool to automate the process? I need to bind a single function which takes a record of callbacks, and that's all. Can greencard support callbacks? If yes, can someone provide a simple example? I thought to do: import Foreign.GreenCard %#include<stdio.h> %fun doit :: Int -> (Int -> Int) -> IO () %code printf("%d",arg2(arg1)); which does not work, as someone might already have expected :) Thanks all Vincenzo

2 3

Re: [Haskell-cafe] Toy application advice wanted
by Jon Cast 05 May '04

05 May '04

Duncan Coutts <duncan.coutts(a)worcester.oxford.ac.uk> wrote: > On Wed, 2004-05-05 at 16:24, Andrei de A. Formiga wrote: > > I'm finding wxHaskell very nice, and a wxWidgets > > binding is something many other "advanced" languages > > don't have (even OCaml). The only downside is having a > > 'Hello World' GUI application with 7 Mb... but it runs > > quite well and smooth once it's loaded. > > We have the same problem with gtk2hs. A stripped simple(ish) prog weighs > in at about 2.5Mb. > > I'm not quite sure what's in there that makes it so big. > > Duncan ghc doesn't support dynamic linking for Haskell libraries (a combination of the amount of work involved and the platform-specific nature of the work and the size of modern hard disks and the frequency with which ghc's ABI changes and the number of implementation details which ghc sticks into the exported ABI of a library anyway (via cross-module inlining)). So any Haskell libraries you use will get statically linked. Jon Cast

1 0

Re: [Haskell-cafe] Toy application advice wanted
by mikeb＠manor.org 05 May '04

05 May '04

On Tue, 4 May 2004, Bill Walsh wrote: > I am amazed that you can even think of doing anything in a week. > I have been at it for at least 15 years: one project; one dead set > proof-of-concept . Well I'm sure you're doing something much more useful/worthwhile than I! =) I understand that the example program I initially thought of is rather, well...typically imperative. That's sort of the reason I chose it. Not only does it have real-world ramifications to my customer (the problem domain is similar) but it will hopefully show me the worst of the best. I.e., Haskell is a beautiful language [disclaimer: this comes from me not really being a Haskell programmer, but admiring it from afar if you will] and can express some unbelievably advanced concepts elegantly, but is it really nasty doing "normal" software crap like loading a bunch of data, popping up GUI windows, etc. etc.? I was hoping that I can piece together something fairly quickly within two weeks in those three languages. They don't have to be robust, or even beautiful. In fact, I can guarantee they won't be. But the real point of the exercise is for me to figure out which environment is the easiest for me to get along with. > I will follow your progress with intense interest. > Your are obviously an experienced programmer, but it is your analytical > skills which interest me most. Well, I am an experienced programmer, but certainly not in the same class as most lambda-junkies as yourselves. I have a long sordid history of programming in imperative languages and I've been bitching about switching for long enough that it's time to put my money where my mouth is, so to speak. I've dabbled in Scheme and Erlang and ML. To be frank, I've all but discarded Erlang as a viable language. Don't get me wrong...it's got some great features; I just think that I could easily implement Erlang in Scheme or Haskell and be quite happy. (At some point in the future when I'm smarter =) So to me, the real question is: Haskell, the pure, or Scheme, the dirty? Haskell is lazy all the time. That's awfully nice...I'm not sure if there's a performance penalty somewhere, but assuming there isn't, kudos to it. Scheme lets you be selectively lazy with macros. At least it's highly customizable (for instance, pattern matching can be implemented in Scheme with macros, whereas other languages that have fixed special forms are quite limited). But Scheme lets you change things if you want to! (The Haskell afficianados gasp). Let's face it...in most long-running applications, data changes a lot during the lifetime of a program. So somehow, you have to manage the repercussions of state changes. Haskell does this with monads. (I'm still wrapping my brain around them, but...) Apparently monads plus some syntactic sugar are like functions that return a value and the next function, which in turn returns a value and the next function, etc. so that referential transparency is maintained. This is a heckuva lot of work to do (set! foo 'bar). I hope I'm not far off base with this. I guess it boils down to this: Haskell ensures that your bindings are never ever ever corrupted, and thus the sematic value of your program is never compromised from the outside. The cost of this is a sharp learning curve, and quite possibly a performance hit. Scheme lets you tromp all over stuff (well, you get the picture), in effect letting you still have control over when you really need to do something imperative for performance or simplicity reasons. I'm not sure which I *like* better. So far neither is visually appealing, mostly because I'm not familiar with them. Scheme seems to have a slightly bigger population than Haskell, which of course means squat because if I really cared about how many programmers of language X there are out there I'd just code in Java (which, sadly, I'm doing now). Anyway, I'm starting to ramble, but I talked to a friend who has similar feelings but is actually pretty good at Common Lisp. He suggested I refocus my energies, and I agree: instead of biting off more than I can chew, and having to learn a whole wad of APIs that aren't really about the language (read: wxHaskell or gtk2hs/the like, or audio packages etc.), just code some really simple problems. Like the Sieve of Eratosthenes, in all three languages. Or a simple publish/subscribe framework with a "master" state holder and many slaves. Or quicksort. Etc. etc. So I'm going to head down that path right now, and try to get a feel for the languages in a slightly more pure fashion. I'll still try to get performance metrics out of them, but I'm not going to bang my head against the wall learning new languages, GUI toolkits, and FFIs all in two weeks. =) Thanks for the replies so far, I really appreciate them. -- Mike J. Bell This is all just my opinion. "Outside of a dog, a book is man's best friend. Inside it's too dark to read." mikeb(a)manor.org

7 7