- Haskell-Cafe - Haskell.org

RE: [Haskell] haskell question
by Simon Peyton-Jones 19 Apr '07

19 Apr '07

[Redirecting to Haskell Cafe] Hi Greg I think it's very likely that you *can* do what you want. I have not taken long enough to understand just what you are trying to do, but I do know why your program is failing, and I think you'll agree it *should* fail. Look at these lines (slightly simplified) class CoreAgentSyntax x where offer :: (CoreProcessSyntax p) => [n] -> (p a x) -> x data MinimalAgentSyntax nn pp = Concretion [nn] pp instance CoreAgentSyntax (MinimalAgentSyntax nn pp) where offer = Concretion The class decl says that every instance of 'offer' must have type forall p, forall n, forall a. (CoreProcessSyntax p) => [n] -> p a x -> x where x is the instance type. Not the "forall n"! Even once x is fixed, we must have a function that is polymorphic in n,p,a But in your instance declaration, offer = Concretion has the type [nn] -> pp -> MinimalAgentSyntax nn pp No for-alls, because the whole instance decl binds the 'nn'. Hence it's just not polymorphic enough. I'm not sure how to fix your problem. Perhaps you need another argument for CoreAgentSyntax? class CoreAgentSyntax n x where offer :: (CoreProcessSyntax p) => [n] -> (p a x) -> x instance CoreAgentSyntax nn (MinimalAgentSyntax nn pp) where offer = Concretion now you are in better shape. There's another problem because you use 'p' with kind * in some places and (*->*->*) in others, but thats a separate issue. Oleg is good at this kind of stuff. Simon From: haskell-bounces(a)haskell.org [mailto:haskell-bounces@haskell.org] On Behalf Of Greg Meredith Sent: 18 April 2007 14:53 To: haskell(a)haskell.org Subject: [Haskell] haskell question Haskellians, In the attached file i have implemented the bare bones core of the syntax -- pre structural equivalence -- of Milner's presentation of \pi via abstractions and concretions. (There's a minor twist in that the notion of location affords more structure than in Milner's original presentation which i have added to support an ultimate probabilistic interpretation.) The file compiles with ghc as is. If you uncomment the last section, however, you see that to close the loop on the constraints for the agent syntax we hit a typing error. i think this is a hard constraint in Haskell that prevents this level of generality in the specification, but maybe you can see a way around it. i'm certainly no Haskell expert. Best wishes, --greg P.S. The reason for this level of generality is that with it you can close the loop and form a mutual recursion between the types for the syntax of processes and the type of names -- without having that choice predesigned into the structure of processes (or names). i can do this trick in OCaml with recursive modules, but it's definitely not pretty. i was hoping Haskell could make it pretty. -- L.G. Meredith Managing Partner Biosimilarity LLC 505 N 72nd St Seattle, WA 98103 +1 206.650.3740 http://biosimilarity.blogspot.com

1 0

Re: Type classes and type equality
by oleg＠pobox.com 19 Apr '07

19 Apr '07

Neil Mitchell wrote: > I'm looking for a type class which checks whether two types are the > same or not. This problem is more complex than appears. It has been solved, however. IncoherentInstances are not required, as IncoherentInstances are generally unsafe. For the full discussion of various solutions, please see Section 9 and Appendix D of the HList paper: http://homepages.cwi.nl/~ralf/HList/paper.pdf The HList code is available http://darcs.haskell.org/HList/ It includes the examples from the paper.

3 5

Fwd: [Haskell-cafe] Transpose using array comprehension
by Chris Witte 19 Apr '07

19 Apr '07

I was trying to use the array comprehension of the Data Parallel Haskell package. http://haskell.org/haskellwiki/GHC/Data_Parallel_Haskell for instance I can define {-# OPTIONS -fparr -fglasgow-exts #-} module MatTest where import GHC.PArr ident = [:[:1,0:],[:0,1:]:] aVect = [:4,5:] dotP :: Num a => [:a:] -> [:a:] -> a dotP xs ys = sumP [:x * y | x <- xs | y <- ys :] matVecMul :: Num a => [:[:a:]:] -> [:a:] -> [:a:] matVecMul xs ys = [:dotP x ys | x <- xs:] but when I tried to define a matrix matrix multiplication matMel :: Num a => [:[:a:]:] -> [:[:a:]:] -> [:a:] I was at a loss for how to do it because i couldn't define transpose. On 4/18/07, Henning Thielemann <lemming(a)henning-thielemann.de> wrote: > > On Wed, 18 Apr 2007, Chris Witte wrote: > > > I just started playing around with GHC.PArr and array comprehension > > and I was wondering if there is a way to define the transpose of a > > matrix using array comprehension? > > Why not > > let swap :: (i,j) -> (j,i) > swap (a,b) = (b,a) > in ixmap (let (lower,upper) = bounds arr in (swap lower, swap upper)) > swap arr > > ? >

1 0

Re: [Haskell-cafe] question about Data.Binary and Double instance
by Duncan Coutts 19 Apr '07

19 Apr '07

On Wed, 2007-04-18 at 08:34 -0700, Bryan O'Sullivan wrote: > Duncan Coutts wrote: > > > I'm currently exploring more design ideas for Data.Binary including how > > to deal with alignment. Eliminating unnecessary bounds checks and using > > aligned memory operations also significantly improves performance. I can > > get up to ~750Mb/s serialisation out of a peak memory bandwidth of > > ~1750Mb/s, though a Haskell word-writing loop can only get ~850Mb/s. > > What are you using to measure the peak number? It seems very low to me. > Even Opterons of a few years of vintage can manage more than 5GB/s. I was using a C word writing loop with no unrolling. On x86 that turned into a 4 asm instruction loop. Reading words is indeed a good deal faster, and on my x86-64 machine reading 8-byte words is faster still. > It's also quite normal to get about half of your peak bandwidth unless > you're going out of your way to use non-temporal loads and stores (e.g. > via SSE2), which is something that e.g. gcc is not good at at all, if > you're using -fvia-c. I'm not interested so much in the peak throughput of the machine as what could be realistically achievable for binary serialisation. So comparing to a non-unrolled C loop seems fair to me. With sufficient improvements in the GHC backend we should be able to approach that. Then any difference is overhead in the Binary library, and that's what I'm really trying to measure. Duncan

1 0

Soft references
by Benja Fallenstein 18 Apr '07

18 Apr '07

Hi all, Is there a way to achieve the effect of soft references in GHC? Or if not, is there any hope that they will be implemented in the future? (Soft references are like weak references, except that they are only reclaimed by the garbage collector if memory is short.) I'm building a memoization table of visualizations of a data set, but I can't store *all* visualizations of that set in memory at the same time. My current solution is to use a least-recently-used cache, but this is unsatisfying because I have to specify the maximum number of objects in the cache; if I make that too large, I run out of memory, so I have to make it so small that I can't possibly run out, meaning that I have to make it a lot smaller than it *could* be. With soft references, the program could memoize the "optimal" number of items, with the user being able to influence the behavior through the 'maximum heap size' RTS option. Any ideas? Thanks, - Benja

2 1

FW: [Haskell-cafe] Tutorial on Haskell
by Taillefer, Troy (EXP) 18 Apr '07

18 Apr '07

=======================LONG WINDED RANT WARNING ====================================== I am not try to hate on Haskell obviously I am a fan otherwise I wouldn't be on this list but criticizing other languages is not only a pointless waste of time it is logically flawed on many levels. A lot of the best software is written and still being written in C++ so I will have to disagree with you that C++ is unreasonable. C++ is dangerous and has its flaws especially in the hands of a noob but it is also has some merits as well. People have very logical reasons for choosing a programming language including 1. Familiarity 2. Available people to work with 3. Tools 4. Libraries 5. Legacy 6. Political 7. Personal Preference etc... In terms of reasoning it is easier to reason in a language/Paradigm that you are familiar with than one that might be better but you are totally unfamiliar with and must learn from scratch. Go on IRC #HASKELL and see how many people are struggling conceptual ideas like Monads, Arrows and Functors I don't go there a lot but every time I am there are a couple of people who are asking questions like just what is a Monad. For myself I want to learn Category Theory because it seems powerful just like Set Theory (which I have found extremely useful). I find manipulating pointers and mutable variables much easier to understand and saner. Maybe if I was a mathematician instead of an Engineer/Computer Scientist I would feel different. Saying Haskell is better and everyone should use it because it is better is about as futile as saying we should all be speaking Esperanto because it is better then English. You would be an awfully lonely person with almost nobody to talk to if all you spoke was Esperanto or Haskell for that matter. I have yet to meet face to face another person who has even written toy code in Haskell let alone any real significant app. I can't use Haskell at work because no one could maintain it. I can't use Haskell in my graduate studies since my supervisor can't read Haskell. So I am stuck writing toy code as a hobby and an educational experience. Programs are mostly for humans to read and modify and lets face it there isn't a lot of people fluent in Haskell. So if I want to work and interact with other Software professionals I better be able to speak the lingua franca which is a derivative of C aka C/C++/C#/Java ect. Obviously C++ isn't that bad because people can write good cool software in it. I like Haskell but I haven't seen a ton of Killer Haskell apps yet ( Or even one for that matter). DARCS is coming close if they fix their conflict problems then Haskell will have its first killer app that I know off. I heard they are working on it this summer looking forward to it DARCS is really cool. Sorry for the long rant but I get tired of people with their juvenile and unconstructive statements like "C++ sucks" Or Language X is better then Language Y even if it could ever be objectively true (which I am sorry it can't be) who cares. ============================END RANT=================================================== ________________________________ From: Sebastian Sylvan [mailto:sebastian.sylvan@gmail.com] Sent: Wednesday, April 18, 2007 1:50 PM To: Taillefer, Troy (EXP) Cc: haskell-cafe(a)haskell.org Subject: Re: [Haskell-cafe] Tutorial on Haskell On 4/18/07, Taillefer, Troy (EXP) <troy.taillefer(a)lmco.com> wrote: I have to strongly disagree with the statement that developers like to debug. Debugging is necessary because you can't reason about any "sizeable" piece of code just is not tractable even in Haskell. Now automated tools for reasoning about programs are very cool but lets face it no real world developer will sit down start to manually formally reason about large pieces of code. I think the emphasis when mentioning "reasoning" really shouldn't be "you can reason formally about your programs and prove that they don't go wrong", nor "when it has gone wrong, you can reason about the program to figure out why", it should be "since the language doesn't do batshit insane things behind your back, your programs will mostly work the first time". The "reasoning" isn't an active task that you schedule time for, at least for a casual user like me, it's part of the actual programming. You do "reasoning" when writing in C++ as well, but you often get it wrong (because the language is, shall we say, unreasonable?) and that causes bugs. -- Sebastian Sylvan +44(0)7857-300802 UIN: 44640862

2 1

SYB and opaque types
by Joel Reymont 18 Apr '07

18 Apr '07

Suppose I need to manually derive Data and Typeable for SourcePos from Parsec to make sure my code compiles. I won't actually be running the code I manually derive since the constructor that includes SourcePos will be skipped. With Neil Mitchell's (and #haskell) help I'm doing this to strip token locations from my AST: import Data.Generics import qualified Text.ParserCombinators.Parsec as P instance Data SourcePos where gfoldl r k x = k x typename_SourcePos = mkTyCon "SourcePos" instance Typeable SourcePos where typeOf _ = mkTyConApp typename_SourcePos ([]) strip = everywhere (mkT f) where f (TokenPos a _) = a f x = x I know that the warnings about gunfold, toConstr and dataTypeOf are harmless but how would I define them to avoid the warnings? Also, the definition of strip above requires -fno-monomorphism- restriction. Should I not worry about it? The code runs just fine, locations are being stripped and tests pass. Thanks, Joel -- http://wagerlabs.com/

3 2

Re: [Haskell] ANN: Data.Proposition 0.1
by Jan-Willem Maessen 18 Apr '07

18 Apr '07

Hmm, your BDD implementation claims (in the comment at the top) that "Equality is fast and accurate." But you don't do sharing optimizations, and use derived Eq (rather than object identity), so it's exponential in the number of nodes. Consider: A / \ B | \ / C / \ D | \ / E Here there are five nodes, but we will do two calls to see if C==C, four calls to see if E==E. If I continue the diagram downwards I add two nodes and double the number of equality calls on the leaf. The sharing optimizations are rather important to making an efficient BDD implementation. I haven't yet seen a Haskell-only BDD implementation that didn't have one or more of the following flaws: 1) IO in all the result types, even though the underlying abstraction is pure. 2) Inability to garbage collect unused nodes. 3) Loss of sharing leading to loss of fast equality. Does anyone have a Haskell-only BDD library which avoids all these problems? I wrote one with flaw #2, but was unable to make weak pointers and finalization behave in anything like a sensible fashion and gave up as it was a weekend project. The concurrency + unsafePerformIO mix was trickier than I initially expected, too. That said, this BDD implementation is pretty similar to the performance & behavior you'd get from Data.IntSet (where the bits of your int correspond to the True/False values of your variables). -Jan-Willem Maessen On Apr 16, 2007, at 9:07 AM, Neil Mitchell wrote: > Hi, > > I am now releasing Data.Proposition. This library handles > propositions, logical formulae consisting of literals without > quantification. It automatically simplifies a proposition as it is > constructed using simple rules provided by the programmer. > Implementations of propositions in terms of an abstract syntax tree > and as a Binary Decision Diagram (BDD) are provided. A standard > interface is provided for all propositions. > > Website: http://www-users.cs.york.ac.uk/~ndm/proposition/ > Darcs: darcs get --partial http://www.cs.york.ac.uk/fp/darcs/ > proposition/ > Haddock: http://www.cs.york.ac.uk/fp/haddock/proposition/Data- > Proposition.html > > The Haddock documentation also has a short example in it. > > This library is used substantially in the tool I have developed for my > PhD, and has been extensively tested. > > Thanks > > Neil > _______________________________________________ > Haskell mailing list > Haskell(a)haskell.org > http://www.haskell.org/mailman/listinfo/haskell

3 2

Re: [Haskell-cafe] Re: Translating perl -> haskell, string "fill ins" with an error on invalid inputseems awfullycomplex. Is there a way to simplify?
by Thomas Hartman 18 Apr '07

18 Apr '07

Claus and Evan ++; that was very helpful. FWIW, my gut feeling is that Claus's first version was easier to understand than the revision with printf, which seems to me to involve a lot more monadic wizardry (Functor, MonadError, fmap, mapm). The first version, which just used maybe, was clear to me within seconds. But again, I learned a lot. Thanks.

3 6

Transpose using array comprehension
by Chris Witte 18 Apr '07

18 Apr '07

I just started playing around with GHC.PArr and array comprehension and I was wondering if there is a way to define the transpose of a matrix using array comprehension? Cheers Chris

2 1