August 2006 - Haskell-Cafe

Writing binary files
by Neil Mitchell 22 Aug '06

22 Aug '06

Hi, I'm trying to write out a binary file, in particular I want the following functions: hPutInt :: Handle -> Int -> IO () hGetInt :: Handle -> IO Int For the purposes of these functions, Int = 32 bits, and its got to roundtrip - Put then Get must be the same. How would I do this? I see Ptr, Storable and other things, but nothing which seems directly useable for me. Thanks Neil

5 5

Fw: [Haskell-cafe] A restricted subset of CPP included in a revisionof Haskell 98
by Brian Hulley 22 Aug '06

22 Aug '06

Hi - here is an exchange that was off-list by accident: ---- Original Message ---- From: Brian Smith To: Brian Hulley Sent: Tuesday, August 22, 2006 5:50 AM > (Brian, I see that my last reply only went to you and so I forwarded > it to the list. Since you replied to me directly, I am responding to > you directly. Feel free to forward this to the list if you want). > > > On 8/21/06, Brian Hulley <brianh(a)metamilk.com> wrote: >> The paper "Template meta-programming for Haskell" suggests >> that Template Haskell should be able to do everything that CPP >> is used for at the moment, such as conditional compilation, macros >> etc, so this might be a possible option, though I have to admit apart >> from reading the paper and doing a few experiments with it I don't >> know enough about the current state of TH to know if it would be a >> viable replacement at the moment. > > I am very interested in TH and I hope to give it a thorough workout > on an upcoming project. Until then, I don't know enough about it to > compare it to CPP. But, I think that in an Editor/IDE, a restricted > CPP would be much easier to implement than TH. > > >> I've got a feeling that even allowing #define could break the >> invariant that a module only has one meaning in a program, since >> #define changes the state of a symbol that's not yet been mentioned >> just as #undef changes the state of a defined symbol eg: >> >> >> module P where >> #define Bar >> import R >> >> module R where >> #ifdef Bar >> import S >> #else >> import T >> #endif >> >> module Q where >> import R > > I think that this program will only compile if it is split into three > files, P.hs, R.hs, and Q.hs. There would be two scopes for CPP > macros: global (each macro is bound to exactly one value for all > modules, and it cannot be undefined or redefined) and file/module > (the macro is bound exactly once in a module and that binding is not > visible to other modules). Except for the "exactly once" part, this > is how GHC and cpphs already work. > > So, in your example, let's assume there is no global binding for > "Bar"--if there was, then there would be a redefinition error in > P.hs. Then, P.hs defines Bar, but this binding is in scope only > inside of P.hs--R.hs does not "see" Bar as being defined. Further, > since R.hs does not contain a binding for "Bar," so module R will > always import module T and never module S. > > A more concrete example is this: > > module P where > #ifdef __GHC__ > import P.GHC > #else > import P.Haskell98 > #endif > > In your editor/IDE you might want to support multiple Haskell > implementations. So, sometimes __GHC__ will be defined (when in GHC > "mode") and sometimes it won't. But, I recommend to handle this in > the same way that IntelliJ handles the JDK setting. That is, if the > user wants to switch the implementation he is using (even > temporarily), then when he does so, the whole program gets > reanalyzed. If the user wants to work in Hugs mode and GHC mode in > the same time, then he should create two projects in the IDE. In one > project, the implementation will be set to GHC, and the other one > Hugs. He will have to define output paths, etc. such that the two > project's ouitput files do not overlap even though their sources > will. Then, analyze each project seperately. This technique also > allows the user to build a version of the project against, say, HaXml > 1.14, and other version against 1.15. > > >> Much as I'd like to not have to deal with CPP, I agree that on >> a practical realistic level it's not going to go away immediately, >> but at least the guidelines you posted give a good starting point >> for library authors to gradually try and reduce the dependence >> on the preprocessor, if they are motivated to (which they >> probably aren't ;-) ) > > Right. Plus, if we follow the guidelines I proposed very strictly, > then we can more easily picture what is required of a CPP > replacement. For example, if the guidelines are too prohibitive, then > we will know we need something more powerful like TH. If the > guidelines do not result in any practical loss of functionality, then > we know that we can get by with a much simpler configuaration > language. > > >> I had a wild idea of a tool to eliminate CPP from a source >> tree by first generating the set of N pre-processed source trees >> (ie now pure Haskell) then running some analysis tool over these >> trees to automatically factor out common code, but this would >> probably be a major undertaking. > > That would definitely be an interesting project, and it would be > useful for other things as well (The duplicate code detection part.) > > > >> Cabal seems to help a lot - Edison and ByteString both built on >> my machine with no trouble at all. I think the more everything can >> be done in Haskell the better it will be. > > I agree. But, I also want it to be easy to reuse C-based libraries > whenever possible. That way, we can spend more time writing code in > Haskell that does new things instead of reimplementing old ideas in > it. > > > You mentioned that library maintainers might not be motivated to make > such changes. I think that people will modify their libraries if > there are compelling tools that need those modifications. In fact, I > think people would even be willing to break previous commitments to > backwards compatibility if there was any benefit to doing so--I mean, > who really uses the latest version of HaXml with GHC 5.x or Hugs > 2002-xx-xx? > > If there was an IntelliJ-level IDE for Haskell, and people were > writing "real world" programs in Haskell, then somebody would get > around to modifying libraries to work for "real world" programs built > using such a tool. Or, somebody that had a commercial interest (a > Haskell consultant or an IDE vender) might make these modifications. > > - Brian

1 0

Description of Haskell extensions used by FPTOOLS
by Brian Smith 22 Aug '06

22 Aug '06

Is there any design document for the FPTOOLS libraries or some description of language features that are (allowed to be) used in them? I am going to be taking some significant time off from my normal jobs in the upcoming months. During part of that time, I would like to do some work to improve the Haskell toolchain. This involves creating or improving tools that parse and analyze Haskell code. My goal is to have these tools support enough of Haskell to be able to handle at least the most important libraries used by Haskell programmers. In particular, this includes all or most of the libraries in FPTOOLS. Plus, I want these tools to operate on Darcs as it is an obvious poster-child for Haskell. Thus, I need to support Haskell 98 plus all the extensions being used in Darcs and FPTOOLS as of approx. March, 2007 (as I intened to start working again at that time). It would be very nice if there was some document that described "Haskell 98 plus all the extensions being used in Darcs and FPTOOLS as of March, 2007." Besides being useful to me, it would be a useful guide for potential contributors to FPTOOLS. Regards, Brian

6 6

Fwd: [Haskell-cafe] A restricted subset of CPP included in a revisionof Haskell 98
by Brian Smith 22 Aug '06

22 Aug '06

On 8/20/06, Brian Hulley <brianh(a)metamilk.com> wrote: > Henning Thielemann wrote: > > On Thu, 17 Aug 2006, Brian Smith wrote: > > > >> I find it strange that right now almost every Haskell program > >> directly or indirectly (through FPTOOLS) depends on CPP, yet there > >> is no effort to replace CPP with something better or standardize its > >> usage in Haskell. > > > > I think there should be more effort to avoid CPP completely. > > I agree, especially as I'm trying to write an editor for Haskell which > will > certainly not cope with CPP at all! ;-) I agree with this sentiment. But, CPP also has some advantages over the other suggestions presented here. Using CPP, we can create libraries that work with GHC 5.x and old versions of Hugs. If an alternative to CPP were chosen, would we update fptools to use this alternative? If it involves new syntax, then updating FPTOOLS to use the new syntax would break backward compatibility. And, if (the vast majority of) FPTOOLS does not use the solution then it is not useful. The reason it would not cope is that CPP turns what would otherwise be one > program/module/library into several programs/modules/libraries which > simultaneously co-exist in the same text in a rather uneasy and vague > relationship, and what's even worse: the same module can have multiple > meanings in the *same* program depending on use of #ifdef #undef etc, thus > making code navigation quite impossible: the meaning of each module now > depends on how you got there and might even be different the second time > round... Notice that with my suggestions none of these problems apply, because all external "command line" parameters have a single static value over the whole program (two modules cannot depend on differing values of a single macro, #undef is not allowed). Changing the bindings for these macros would invalidate any cached information the editor has. But, the editor has to support updating its caches anyway, to deal with switching libraries in/out, etc. I won't use an IDE effectively if it can't handle the libraries in FPTOOLS, which liberally use the preprocessor. Either the IDE has to handle the preprocessor or the libraries have to stop using them, or they have to meet somewhere in the middle. I think the acid test would be to reach a point where anyone can download > the source for some large program such as GHC and just type ghc --make > Main > and expect the program to be built in one pass with no problems. I agree with this sentiment as well. Most of the little data processing programs I have written are built exactly like that already. How can we get the existing libraries to build like that though? - Brian

1 0

iterative algorithms: how to do it in Haskell?
by Tamas K Papp 21 Aug '06

21 Aug '06

Hi, I am a newbie learning Haskell. I have used languages with functional features before (R, Scheme) but not purely functional ones without side-effects. Most of the programming I do is numerical (I am an economist). I would like to know how to implement the iterative algorithm below in Haskell. f is an a->a function, and there is a stopping rule goOn(a,anext) :: a a -> Bool which determines when to stop. The algorithm looks like this (in imperative pseudocode): a = ainit while (true) { anext <- f(a) if (goOn(a,anext)) a <- anext else stop and return anext } For example, f can be a contraction mapping and goOn a test based on the metric. I don't know how to do this in a purely functional language, especially if the object a is large and I would like it to be garbage collected if the iteration goes on. Thank you, Tamas

11 18

Haddock/Cabal/base bug?
by Neil Mitchell 21 Aug '06

21 Aug '06

Hi, I want to generate documentation for the base libraries, so I darcs get the base libraries, and a very basic (the sample default) Setup.hs, and try: runhaskell Setup configure runhaskell Setup build Pretty much no luck, but I'll leave that for someone else to sort out :) runhaskell Setup haddock First task is to remove errors about -fno-implicit-bangs, which are in a few files, relatively easy to do. After that it gets quite far (invokes haddock a fair bit) before failing with: haddock.exe: Main.all_subs_of_qname: unexpected unqual'd name:IOMode This is on Windows, cabal, base and haddock all fresh from darcs, GHC 6.4.2 Can anyone point out what the problem is, whose fault it is, and what can be done to fix it? Thanks Neil

2 2

type gurus, can you please help?
by Bulat Ziganshin 21 Aug '06

21 Aug '06

Hello haskell, i'm started to write article about type classes. can you, type gurus, please check this initial text for correctness in explaining differences between classes and type classes? at this moment C++/C#/Java languages has classes and templates/generics. what is a difference? with a class, type information carried with object itself while with templates it's outside of object and is part of the whole operation for example, if == operation is defined in a class, the actual procedure called for a==b may depend on run-time type of 'a' but if it is defined in template, actual procedure depends only on template instantiated (and determined at compile time) Haskell's objects don't carry run-time type information. instead, class constraint for polymorphic operation passed in form of "dictionary" implementing all operations of the class (there are also other implementation techniques, but this don't matter). For example, eqList :: (Eq a) => [a] -> [a] -> Bool translated into: type eqDictionary a = (a->a->Bool, a->a->Bool) eqList :: eqDictionary a -> [a] -> [a] -> Bool where first parameter is "dictionary" containing implementation of "==" and "/=" operations for objects of type 'a'. If there are several class constraints, dictionary for each is passed. If class has base class(es), the dictionary tuple also includes tuples of base classes dictionaries: class Eq a => Cmp a where cmp :: a -> a -> Comparision cmpList :: (Cmp a) => [a] -> [a] -> Comparision turns into: type cmpDictionary a = (eqDictionary a, a -> a -> Comparision) cmpList :: cmpDictionary a -> [a] -> [a] -> Bool Comparing to C++, this is like the templates, not classes! As with templates, typing information is part of operation, not object! But while C++ templates are really form of macro-processing (like Template Haskell) and at last end generates non-polymorphic code, Haskell's using of dictionaries allows run-time polymorphism (explanation of run-time polymorphism?). Moreover, Haskell type classes supports inheritance. Run-time polymorphism together with inheritance are often seen as OOP distinctive points, so during long time i considered type classes as a form of OOP implementation. but that's wrong! Haskell type classes build on different basis, so they are like C++ templates with added inheritance and run-time polymorphism! And this means that usage of type classes is different from using classes, with its own strong and weak points. -- Best regards, Bulat mailto:Bulat.Ziganshin@gmail.com

5 19

Ix, Random, and overlapping instances
by Lambda The Insignificant 21 Aug '06

21 Aug '06

Greetings Haskellers, I have recently found myself in a situation where I was needing to general tuples of randomized things; in my case, two coordinates and a custom data type, Direction. As it is always better to generalize, I decided to try to write some code which would generalize randomR to any type which is Ix-able: > instance Ix i => Random i where > random = error "No supported instance for random" > randomR r g > = let > r' = range r > (i, g') = randomR (0,length r'-1) g > in > (r' !! i, g') This works splendidly; with my given example above, I could do: > data Dir = N | NE | E | SE | S | SW | W | NW deriving (Eq, Ord, Show, Ix) ghci> newStdGen >>= print . take 10 . randomRs ((1,1,N),(10,10,NW)) [(5,4,S), (3,10,SE), (6,5,N), (6,6,NW), (10,8,SE), (3,10,W), (10,3,E), (1,6,NE), (10,10,SE), (3,3,S)] As you can see, this results in random tuple(s) of these indexable types. My implementation required -fallow-overlapping-instances and - fallow-undecidable-instances for the code to compile. A few questions: 1) How does the compiler determine how to choose a particular instance for those cases where they overlap? I noticed that if I tried to generate a random :: Int, I was getting the error message defined in the above instance, indicating that the compiler was choosing my implementation over the default for Int (obviously because Int is an instance of Ix). Is there some way to exclude classes in the class qualifier, or to provide an instance as a default case without resorting to -fallow-overlapping-instances? 2) Is there a logical overlap between class Ix and class Bounded? It seems that we could use the ranges of data types of Bounded to generate a reasonably sane `random` implementation for all Ix, if we can guarantee that all Ix instances are Bounded. Would adding an additional class constraint be sufficient to catch the major cases? i.e., instance (HasBounds i, Ix i) => Random i where ... 3) Is there some reason that Ix is restricted to the Int datatype? Is there anything preventing us from moving to a genericIx,or something along those lines which support any integral type (say Integer), or is it merely for performance reasons that we use Int? Is the expectation that if you are using spaces larger than what an Int can accommodate, you should be using a different mechanism? Thanks much for any comments you all can give. Cheers, Lambda the Insignificant

2 1

cabal specify a "tested version", ghci target?
by Marc Weber 20 Aug '06

20 Aug '06

1.) I know I can use Build-Depends: lib == <version>, lib2 < version, lib3 >= version and so on. Do you think it would be useful to introducue some notation to indicate a "tested with" ? Reason, purpose: I think its sometimes the case that a author/ mantainer is quite busy with other projects and misses that some dependencies break things.. If you want to try out you're left with some compiler errors and a dependency and have to try out which version works. I would propose using this syntax: lib-1.3 >=1.1 to indicate that lib 1.1 is required at leeast and tested with up to 1.3.. Cabal might then give a warning if you try to use 1.4 or greater "using newer version than tested" or similar.. What do you think? Would this be useful? 2.) Is there a ghci target? I find it quite useful using ghci to debug test/ check types functions of single modules... My hack is using a Setup.hs file like this adding ./setup ghci which runs ghci and adds all dependency as -pakcage arguments: module Main where import Distribution.Simple import Distribution.Simple.Configure import System import System.Process import System.Exit import Monad main= do args <- getArgs -- check args to launch ghci when (length args > 0) $ do when ((args!!0) == "ghci") $ do lbi <- getPersistBuildConfig let packageArgs = (concat.concat) [ [" -package ", showPackageId pkg] | pkg <- packageDeps lbi ] system("ghci " ++ packageArgs) exitWith ExitSuccess defaultMain -- fallback to defaultMain Would this be useful for you, too? Marc W.

3 2

A backwards-compatible record proposal
by Brian Hulley 20 Aug '06

20 Aug '06

Hi - As I've been writing a Haskell program over the past few months the main problem I encounter is that record field names are not local to the record type, and any systematic way of making them local (eg by prepending "_Tycon_") results in names that are just too clunky, and I feel that identifiers *must* use systematic naming conventions to get code that will be easy to understand and maintain. Although it is relatively easy to think up a better record system, it has taken me up till now to discover how such a system could be integrated with Haskell as it is at the moment, since it's unlikely that the existing record system will ever disappear, at least not in the next few years, and in fact, there are some good points about the existing record system that I wouldn't like to lose. I'll motivate the proposal then talk about how it could be implemented. Motivation ======= Consider the following: data Vector3 a = Vector3{x :: a, y::a, z::a } data Normal3 a = Normal3{x :: a, y::a, z::a } We've got a problem because (x) has been introduced twice to the top level namespace. In the above example, it could be argued that I should have written: data Arr3 a = Arr3 {x::a, y::a, z::a} newtype Vector3 a = Vector3 (Arr3 a) newtype Normal3 a = Normal3 (Arr3 a) but for the sake of argument, let's suppose we can't do this - perhaps one of the record types has some different fields as well. A related problem is suppose I have: data Size = {width :: Int, height :: Int} data Rect = Rect{x1, y1, x2, y2::Int} width :: Rect -> Int width Rect{x1,_,x2,_} = x2 - x1 because a record field of Size has the same name as a top level function. This second conflict can be avoided by always using the rule that record fields begin with an underscore and all other variables don't. So applying this to the first problem, we have: data Vector3 a = Vector3{_x :: a, _y::a, _z::a } data Normal3 a = Normal3{_x :: a, _y::a, _z::a } Of course we haven't solved it yet! But now suppose we introduce a new piece of syntactic sugar, and write: data Vector3 a = Vector3{.x :: a, .y::a, .z::a } data Normal3 a = Normal3{.x :: a, .y::a, .z::a } ie putting a '.' before each field name. The intended meaning is that dotted field names do *not* generate top level functions. Instead they allow the compiler to generate instance decls as follows, where we've introduced a new form of identifier, the dotted id, which behaves as a postfix operator which binds more tightly than function application and can also be used as a class name (by the compiler only): class (.x) :: a b | a -> b where (.x) :: a -> b class (.y) :: a b | a -> b where (.y) :: a -> b class (.z) :: a b | a -> b where (.z) :: a -> b For each dotted id, there is a class defined as above, which is available globally to the whole program as if a module containing an infinite set of class decls as above was exported by the Prelude. In the module containing the data decl for the record, the compiler inserts the following: instance (.x) (Vector3 a) a where (.x) v = ... -- compiler generated code to access the field Then within the rest of the program we can write: magSquared :: Num a => Vector3 a -> a magSquared v = v.x*v.x + v.y*v.y + v.z*v.z -- explicit type when specific function is required vec_x = ((.x) :: (Vector3 a -> a)) The advantage of this proposal is that it is completely backwards compatible with records as they are at the moment, and we can choose which fields we want to be dotted and which we want to just keep as normal top level functions. The only extra thing we need to do is put a dot before the field names we want to access via the dotted syntax, but we in any case needed to use an underscore when we wanted a systematic way to avoid conflicts between field names and other top-level names so there is no extra effort involved. Implementation ========== The above could almost be implemented just by parsing a source file containing uses of dotted fields and using a conversion like: data Rec a = Rec { .f :: a} let rec = Rec {.f = 78} -- dot is used here too p = rec.f * rec.f ==> data Rec a = Rec a instance Dot__f (Rec a) a where __dot_f (Rec a) = a let rec = Rec 78 p = __dot_f rec * __dot_f rec So far so good, but the alert reader :-) will have noticed that we now have a *major* problem with abstraction because although we can write: module M (Rec) where ... code in another module can still say rec.f because in Haskell, all instance decls in a module are always exported [1]. This problem may disappear in Haskell' [2]. http://hackage.haskell.org/trac/haskell-prime/ticket/19 Therefore I think the desugaring would need to take place in the compiler so the compiler could avoid exporting the compiler-generated instances when the fields are not present in the module export list. Regards, Brian. [1] http://haskell.org/onlinereport/modules.html#import-instances [2] http://hackage.haskell.org/trac/haskell-prime/ticket/19 -- Logic empowers us and Love gives us purpose. Yet still phantoms restless for eras long past, congealed in the present in unthought forms, strive mightily unseen to destroy us. http://www.metamilk.com

4 9