- Haskell-Cafe - Haskell.org

unique id for data types
by Barney Hilken 18 Sep '07

18 Sep '07

In order to make my records system practically useable, I need a type family > type family NameCmp n m which totally orders datatypes. More precisely, it should return one of the following types: > data NameLT = NameLT > data NameEQ = NameEQ > data NameGT = NameGT for each pair of datatypes n & m, according to whether n < m, n = m, or n > m in some global ordering. This ordering needs to be independent of the context, so it can't be affected by whatever imports there are in the current module. What I want to know is: does GHC give datatypes any global id I could use to generate such an ordering? Would fully qualified names work? Secondly (assuming it's possible) how easy would it be for me to write a patch to add NameCmp to GHC? Where in the source should I start looking? Thanks, Barney.

2 2

Observations from ListLike
by John Goerzen 18 Sep '07

18 Sep '07

Hi folks, I just announced ListLike[1] on haskell(a)haskell.org. Rather than repeat that announcement here, I'd like to make a few observations that came out of the development of this program: * I wrote extensive QuickCheck cases for this and wrapped them in HUnit for better display and future extensibility. Two bugs in the GHC base libraries were discovered as a result (one has been fixed already in HEAD, one not) * Hugs programs that use cpphs can't use ByteString * It would be really nice if QuickCheck supported I/O and some version of HUnit's TestLabel to generate hierarchical names when failures occur. * It would also be much nicer if QuickCheck defined Arbitrary instances for Word8 and Char. On Hugs, QuickCheck also lacks an instance of Maybe, but GHC has it for some reason. * Testing functions like foldl in Hugs, where QuickCheck is supplying the list and the function, cause Hugs to crash with stack errors. Testing some functions over ByteStrings cause Hugs to crash complaining of being out of ForeignPtrs. I have yet to figure out how to fix either problem; raising the heap size doesn't help. * I had some cases in String.hs[2] where Hugs did not permit me to define the default functions within the class declaration. However, I could define the functions outside, with the exact same type signature as they'd have inside, and within the class declaration, just say something like "words = myWords" to refer to the external declaration. GHC worked fine both ways. It seems that Hugs has a number of issues with code like ListLike that uses some more advanced Haskell features, even though Hugs nominally supports them. If you look through my Mercurial changelogs, you'll see "fix blah for hugs" messages all over. Which is too bad, because Hugs seems more portable (especially to embedded devices) than GHC. [1] http://software.complete.org/listlike [2] http://software.complete.org/listlike/browser/src/Data/ListLike/String.hs

3 2

Israel Haskell Programmers
by B K 18 Sep '07

18 Sep '07

Hello, Are there any Haskell Hackers on this mailing list who live in Israel? I am interested in starting an Israel Haskell User Group.

3 2

Dynamic thread management?
by Hugh Perkins 18 Sep '07

18 Sep '07

Haskell/FP seems to have solved the hardest bit of threading, which is making it obvious which bits of a program are parallelizable, and which are not. Remains to actually parallelize out the programs. Am I being naive or is this trivial? There has been a lot of talk about parallelizing out a program statically, at compile time, but this would seem to suffer from the halting problem? The only way to know how long a function, or one of it's sub-functions, will take to run is to actually run it? Is there some reason why we cant just start a function running in a single thread, whilst running profiling, then after a while we check which bits of the function are taking time to run, and are parellizable, and we parallelize those out? This sounds reasonably trivial and workable? Basically, "parallizable" in a Haskell context means, as a first approximation, any map, foldr, foldl or derivatives, and any independent let assignments, then we can always add in extra parallizable cases later. Profiling is already built into Haskell AFAIK? so the profiling information is already available? Thoughts? Is there some reason why such an approach has been disregarded/is harder than it sounds? (By the way, can someone provide me a non-google link to the SPJ video talking about nested data parallism? google video is unavailable from my current location, and although I did watch this video once, it was on my second day of doing Haskell, a few weeks ago, so much of it was incomprehensible to me ;-) )

15 29

Re: [Haskell] Swapping parameters and type classes
by Bas van Dijk 17 Sep '07

17 Sep '07

On 9/16/07, Mads Lindstrøm <mads_lindstroem(a)yahoo.dk> wrote: > Hi all > > If I have this type: > > data Foo a b = ... > > and this class > > class Bar (x :: * -> *) where ... > > I can imagine two ways to make Foo an instance of Bar. Either I must > "apply" the 'a' or the 'b' in (Foo a b). Otherwise it will not have the > right kind. To "apply" the 'a' I can do: > > instance Bar (Foo a) where ... > > But what if I want to "apply" the 'b' ? How do I do that ? > > > Greetings, > > Mads Lindstrøm > > > _______________________________________________ > Haskell mailing list > Haskell(a)haskell.org > http://www.haskell.org/mailman/listinfo/haskell > (Redirecting to Haskell-Cafe) I think the easiest way to make it work is with a newtype wrapper around 'Foo a b' that swaps the arguments, like Brent mentioned. Without it you need some kind of type level lambda abstraction. Something like: instance Bar (\x -> Foo x b) where ... But this is not valid Haskell. cheers, Bas

1 0

Records: Examples
by Barney Hilken 17 Sep '07

17 Sep '07

> {-# LANGUAGE TypeFamilies #-} Hi Justin, thanks for your interest. Hope this helps! > module Examples where > import Records To get started, you need to define your labels. They are just singleton datatypes: > data FirstName = FirstName deriving (Show, Eq, Ord) > data Surname = Surname deriving (Show, Eq, Ord) > data Address = Address deriving (Show, Eq, Ord) > data PhoneNo = PhoneNo deriving (Show, Eq, Ord) you can define as many as you like. Next you have to define the order on fields. At the moment you have to do this by hand, but I hope to get ghc to do this automatically: > type instance NameCmp FirstName FirstName = NameEQ > type instance NameCmp FirstName Surname = NameLT > type instance NameCmp FirstName Address = NameLT > type instance NameCmp FirstName PhoneNo = NameLT > type instance NameCmp Surname FirstName = NameGT > type instance NameCmp Surname Surname = NameEQ > type instance NameCmp Surname Address = NameLT > type instance NameCmp Surname PhoneNo = NameLT > type instance NameCmp Address FirstName = NameGT > type instance NameCmp Address Surname = NameGT > type instance NameCmp Address Address = NameEQ > type instance NameCmp Address PhoneNo = NameLT > type instance NameCmp PhoneNo FirstName = NameGT > type instance NameCmp PhoneNo Surname = NameGT > type instance NameCmp PhoneNo Address = NameGT > type instance NameCmp PhoneNo PhoneNo = NameEQ Now we are ready to play! To define records, use (=:) and (+:) > barney = FirstName =: "Barney" +: Surname =: "Hilken" +: > Address =: "Horwich" +: PhoneNo =: "697223" You can use as many or as few of the fields as you like, and you can write them in any order, but trying to use a field twice in the same record will give you a (rather incomprehensible) type error. > justin = Surname =: "Bailey" +: FirstName =: "Justin" +: Address =: "Somewhere" To extract the value at a field use (.:) > myPhone = barney.:PhoneNo To delete part of a record, use (-:) > noCallers = barney -: Address To update existing fields in a record, use (|:) > barney' = barney |: Address =: ((barney .: Address) ++ ", UK") The power of the records system is that these five operators, =: +: .: -: |: are Haskell polymorphic functions. So you can define functions like > livesWith p q = p |: Address =: (q .: Address) which returns p, but with its Address field changed to that of q. Note that this function works on any records p and q with Address fields, whatever other fields they may have. You can even define functions parametrised by field names: > labelZip n m = zipWith (\x y -> n =: x +: m =: y) then 'labelZip FirstName Surname' is a function which takes two lists and returns a list of records: > names = labelZip FirstName Surname ["Barney", "Justin"] ["Hilken", "Bailey"] of course, labelZip isn't restricted to the four labels we defined earlier, it works on anything. The system is strongly typed, so record errors (such as missing or duplicated fields) are caught at compile time. There are type operators (:=:), (:+:), (:-:), (:.:) corresponding to the record operators, and classes `Contains`, `Disjoint`, `Subrecord` which allow you to express conditions on types. Unfortunately, the type system sometimes decides that a function has a different type from the one you expect, and won't accept the header you want to give it. More experience with the system is needed before we can say whether this is a problem. Barney.

1 0

Re: [Haskell] Records
by Justin Bailey 17 Sep '07

17 Sep '07

On 9/16/07, Barney Hilken <b.hilken(a)ntlworld.com> wrote: > Now that I have a version of ghc with type classes, I have had a go > at implementing records based on the ideas I mentioned on this list a > few months ago. The code of my first attempt is available at http:// > homepage.ntlworld.com/b.hilken/files/Records.hs That's pretty interesting stuff, but it's hard to see how ti works without some examples. Would you want to include some in your comments? I do appreciate the thorough documentation otherwise, though. Justin p.s. replied via haskell-cafe.

1 0

Cross-compiling of GHC
by L.Guo 17 Sep '07

17 Sep '07

Hi all: How to build a GHC which can run in a embed linux system ? I have toolchain for the targer system. Using which, I can compile on PC and run the program on the embed system. Is this means, if I change CC env-var to the toolchain compiler and compile GHC manually, I can get a program for the embed system ? I have read Building/Porting for GHC[1], that does not mention anything about how to build a GHC on a target machine using a cross-compiler like xxx-linux-gcc. *or* What do I need to do besides the stages that page mentioned ? Regards [1] http://hackage.haskell.org/trac/ghc/wiki/Building/Porting -------------- L.Guo 2007-09-17

2 1

Re: [Haskell] Linear Equation Solver Using Arrays
by Andrew Wagner 17 Sep '07

17 Sep '07

(Replying on haskell-cafe) Let me start with a disclaimer: I haven't looked at your code extensively. That said, the feeling that I get from it is one of listening to a non-native speaker. The things in your program are obviously not completely inaccurate, or they wouldn't have type-checked. And, presumably, you've already done some testing, so it probably produces the right answers. Similarly, there are many things in English that you _could_ say, which would be entirely grammatically correct, but which native speakers would never say. For example, in your code, I noticed a pattern which I've just recently learned: <code> do ... tmp <- readArray scaleInfo j writeArray scaleInfo imax tmp </code> Now, if you were to de-sugar the last 2 lines, it would look something like this: <code> readArray scaleInfo j >>= \tmp -> writeArray scaleInfo imax tmp </code> ...but as soon as you see that lambda expression, you should (hopefully) say "oh, that can be written more easily: <code> readArray scaleInfo j >>= writeArray scaleInfo imax </code> Of course, a native speaker will probably not even think about the intermediate lambda expression, but the point is, this is an idiom. And, really, this is a very tiny example. Your entire program is written in a very imperative style. "Do this, then do that, then do the other thing." And this is to be expected. After all, you said that you were directly porting the algorithm from an imperative language. However, if you're going to do this idiomatically, you really need to think very differently. I suggest reading and pondering this well-known blog entry for more ideas about the differences between the approaches of functional and imperative programming: http://blogs.nubgames.com/code/?p=22 Sorry, I don't have more specific suggestions, but I suspect this is what most people thought when they read your code: it's hard to make specific suggestions, when you're really approaching the problem in a non-idiomatic way. On 9/16/07, Xiao-Yong Jin <xj2106(a)columbia.edu> wrote: > Dear Haskellers, > > My thanks to people on the irc channel, especially `int-e'. > With their help, I managed to write a linear equation solver > using STUArrays. > > > It is basically a direct rewrite of a C version, which > adopts Crout's LU decomposition method with implicit > pivoting, in chapter 2.3 of the book Numerical Recipes. > > I must admit that the code is in a really bad style, > somewhat ugly, because of the limit of my Haskell knowledge. > The code is attached in this email for your amusement. I > would be really thankful if you can give me any kind of > comments, idea, improvement, criticism, because I sincerely > want to make the code better. > > Best regards, > Xiao-Yong > -- > c/* __o/* > <\ * (__ > */\ < > > _______________________________________________ > Haskell mailing list > Haskell(a)haskell.org > http://www.haskell.org/mailman/listinfo/haskell > > >

2 1

Parsec Scanner Question
by Jude 16 Sep '07

16 Sep '07

Haskell Folks, I have an existing Parsec CharParser parser that I would like to extend to include line continuation support. When there's a backslash-newline combination anywhere in the token stream, I'd like to remove it so it's not read by the rest of the parser, even if its in the middle of a keyword. Currently I'm doing this with a separate "preprocess" function that weeds out all the '\\\n' strings from the input before passing it to the parser. The problem with this is it messes up the line position when there's an error below the '\\\n' point, since the newline is never seen by the parser. So I've been trying to figure out how to handle this better with Parsec. I think I need something that acts just like CharParser, except it skips over the backslash-newline tokens, It looks like section 2.11 in the Parsec manual (Advanced: separate scanners") covers this, but I'm not having much luck turning it into working code. Could someone who's written their own scanner spell out how it's done in a little more detail ? Thanks ! - jude

1 0