March 2007 - Haskell-Cafe

Re: Performance Help
by Dominic Steinitz 19 Mar '07

19 Mar '07

Thanks. > Fusing the ws is trickier. Directly appealing to the fibonacci-number > example is not recommended because this would mean to keep the last 16 > ws in memory and shifting them right to left by hand. But as the Are you saying this because we don't want a 16-tuple? > "Alternate method of computation" on the website suggests, you can > delegate the shifting to an index that shifts around mod 16. Having a > mutable array is helpful, then. Are you saying that because haskell doesn't really have mutable arrays, this is ruled out? > Of course, you can also fill a large static (boxed!) array of 80 Word8s via > > ws :: Data.Array.IArray.Array Int Word8 > ws = accumArray 0 (0,80) (curry snd) $ > zip [0..15] xs ++ [(i, xxor i) | i<-[16..80]] > where > xxor i = ws ! (i-16) `xor` > ws ! (i-3) `xor` ws ! (i-8) `xor` ws ! (i-14) > > or something like that (I didn't care about correct indices and bounds). > GHC can fuse such array accumulations. Why is an array better than a list? Is it because (!) is O(1) and (!!) is O(n)? > > In general, keeping stuff in lists is not wrong, but ByteStrings are > more adapted to current CPU and RAM architecture. I'm not clear how ByteStrings help here. We need to put bits into 32 words and operate on these. Dominic.

2 1

Re: Re: HS-Plugins 1.0 chokes on simple test, WinXP GHC-6.6 (Conal Elliott)
by Vivian McPhail 19 Mar '07

19 Mar '07

Hi, I just setup and installed hs-plugins from darcs on WinXP using ghc-6.6 and MSYS. The hs-plugin test suite all passes. Can you send me something that generates your error and I'll have a look at it. Vivian > Message: 2 > Date: Sat, 17 Mar 2007 18:06:12 -0700 > From: "Conal Elliott" <conal(a)conal.net> > Subject: Re: [Haskell-cafe] Re: HS-Plugins 1.0 chokes on simple test, > WinXP GHC-6.6 > To: haskell-cafe(a)haskell.org, "Donald Stewart" <dons(a)cse.unsw.edu.au> > Message-ID: > <ea8ae9fb0703171806h641da4a8q28fa19b86bd2ad85(a)mail.gmail.com> > Content-Type: text/plain; charset="iso-8859-1" > > Glad to know, Ian. Thanks. > > I'd sure love to have a working hs-plugins again, especially > in time to help with my ICFP submission. I wouldn't begin to > know how to crack something like this bug. Does anyone have > any ideas? > > Cheers, - Conal > > > On 3/17/07, Ian Lynagh <igloo(a)earth.li> wrote: > > > >c:/ghc/ghc-6.6/HSbase.o: unknown symbol `_free' > > > > > > > > >Don mentioned this is a known problem. Is it on > anyone's todo list? > > > > Not mine. > > > > > > > On 3/17/07, Ian Lynagh <igloo(a)earth.li> wrote: > > > > On Fri, Mar 16, 2007 at 09:52:10PM -0700, Conal Elliott wrote: > > > BTW, to get hs-plugins to build, I changed two lines in > > > hs-plugins/configure. > > > > As it happens, I sent Don a similar patch last night, so hopefully > > it'll be fixed in darcs soon. > > > > > First I tried tr -d '\n', but it didn't work, and I don't > know why. > > > > Because it's a \r you're trying to delete. > > > > > On 3/16/07, Conal Elliott <conal(a)conal.net> wrote: > > > > > > > >I got hs-plugins to compile fine on winxp, but now when > I run it I > > > >get > > a > > > >crash with this message: > > > > > > > >c:/ghc/ghc-6.6/HSbase.o: unknown symbol `_free' > > > > > > > >Don mentioned this is a known problem. Is it on > anyone's todo list? > > > > Not mine. > > > > > > Thanks > > Ian > > > > >

2 1

Visibility of definitions in Hugs Prelude
by Fritz Ruehr 19 Mar '07

19 Mar '07

I am wondering about the visibility of definitions in the Hugs Prelude. More specifically, when I ask for info about the Eq class, I see a lot of instances, including: > instance Eq Key On the other hand, Key is not in scope: > Hugs> :info Key > Unknown reference `Key' I imagine this has to do with importing/exporting, or maybe some modern Hugs empty-Prelude funkiness. Or perhaps it is because Key is a "non-standard export" (but that is just a comment in the Prelude). In any case, is it intended that ":info" will tell me about things that I (and it) can't otherwise see? Please forgive my naivete re module rules in advance ... . -- Fritz

1 0

Using external programs with Haskell in windows.
by Christian Lean 18 Mar '07

18 Mar '07

I'm looking for a way to run an external program and get the results in haskell. Something similar to HSH but that will work in windows. I don't need anything too complex, just to provide the command to be run as a string and get the result as a string. Is this possible? One of the HSH examples is all I need: runS $ "echo /etc/pass*" -> "/etc/passwd /etc/passwd-"

2 4

Stack Space Overflow.
by John T Davis 18 Mar '07

18 Mar '07

oHello, I'm writing a simulation of a game of Nim with genetic programming. I've run into a stack space overflow problem with one of my functions (part of a class instance), which I've included below. crossover_mutate enp = enp { progPop = ansPop, numProgs = (numProgs enp) + (crossNum enp)} where crossResult = crossNimPlayerPop (progPop enp) (seed enp) (theBoard enp) (numProgs enp) (crossNum enp) mutateResult = mutateNimPlayerPop crossResult (mutateRatio enp) popParts = crossResult ++ mutateResult newPop = Set.toAscList (Set.fromList popParts) ansPop = tournament (newPop, newPop) (theBoard enp) Now, here's where it gets weird. I've got test data that I've run all the constituent functions (crossNimPlayerPop, mutateNimPlayerPop, tournament) on, and they all work as they should. Now, when I run the same data on crossover_mutate, I get the aforementioned stack space overflow. I'm totally stumped on this... I've even run the equivalent of ansPop on the command line, and despite how truly unreadable the code is in that format, it runs fine. Any suggestions or thoughts would be greatly appreciated. Have a nice day. :-) Thank you, John T Davis

3 2

Does inlinePerformIO = unsafePerformIO . unsafeInterleaveIO semantically?
by Isaac Dupree 18 Mar '07

18 Mar '07

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Compare the guarantees (in practice and/or theoretically) of inlinePerformIO and unsafePerformIO . unsafeInterleaveIO (i.e. function composition) (aside from inlinePerformIO being better optimized). Are they the same, or if not, what do they mean - what are the differences? Not yet understanding any of these operations fully, Isaac -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.3 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFF/QsBHgcxvIWYTTURAtC6AJ4rKqhfmVi5cLC2MhXF9J4Pxk30KQCdGnQ7 Wd0BrIreZhIc0Gzm/tQLa78= =poy9 -----END PGP SIGNATURE-----

3 2

Haskell and SSL
by Thomas David Baker 18 Mar '07

18 Mar '07

I'm writing an application that scrapes some data from some websites. I was using Network.Browser from http://haskell.org/http/. So far, so good. However, now I need to scrape some websites that use SSL. Network.Browserdoesn't seem to support this (although for some reason it does work on one of the sites so maybe I'm getting this wrong). Someone on #haskell pointed me at hsgnutls which is an wrapper for parts of GNU TLS. But I actually had more success writing a wrapper around wget. I know that darcs uses curl in a similar way for some stuff but it still feels like I'm doing the Wrong Thing. Does anyone know of a Haskell library that does support scraping SSL enabled web pages? Or would adding it to Network.Browser be terribly difficult? Looking at GNU TLS it looks more complicated than I imagined. Thanks in advance for your help, Tom

2 1

HS-Plugins 1.0 chokes on simple test, WinXP GHC-6.6
by Alistair Bayley 17 Mar '07

17 Mar '07

Does anyone have hs-plugins working on WinXP with ghc-6.6? When I run the simple test below I get this error: Main: c:/ghc/ghc-6.6/HSbase.o: unknown symbol `_free' Main: user error (Dynamic loader returned: user error (resolvedObjs failed.)) Am I doing something obviously dumb? Alistair module Test1 where test1 = putStrLn "test1" module Main where import Prelude hiding (catch) import Control.Exception import Data.List import System.Environment import System.Plugins instance Show (LoadStatus a) where show (LoadFailure errors) = "LoadFailure - " ++ (concat (intersperse "\n" errors)) show (LoadSuccess m p) = "LoadSuccess" main = do a <- getArgs let modName = case a of (n:_) -> n _ -> "Test1" let modPath = "./" ++ modName ++ ".o" let method = "test1" fc <- catch (load modPath [""] [] method) (\e -> return (LoadFailure ["Dynamic loader returned: " ++ show e])) case fc of LoadFailure errors -> do fail (concat (intersperse "\n" errors)) LoadSuccess modul proc -> do let p :: IO (); p = proc proc

7 11

How do I avoid stack overflows?
by oleg＠pobox.com 17 Mar '07

17 Mar '07

DavidA wrote: > I'm trying to write some code which involves lots of matrix multiplications, > but whenever I do too many, I get stack overflows (in both GHCi 6.4.2, and > Hugs May 2006). By placing a couple of strictness annotations, your test' gives the expected answer (given some time) on Hugs. GHCi unfortunately runs into some kind of bug (it says so itself), an unimplemented opcode. The test'' below gives that bug instantly... import List (transpose) -- not needed here -- foldl' f z [] = z -- foldl' f z (h:t) = (foldl' f $! f h z) t -- sum' l = foldl' (+) 0 l map' f [] = [] map' f (h:t) = scons (f h) (map f t) -- strict cons. Could be associated with an infix op, e.g., :$ scons :: a -> [a] -> [a] scons x l | x `seq` l `seq` False = undefined scons x l = x:l u <.> v = sum $ zipWith (*) u v a <<*>> b = multMx a (transpose b) where multMx [] _ = [] multMx (u:us) bT = scons (map' (u <.>) bT) (multMx us bT) id3 = [[1,0,0],[0,1,0],[0,0,1]] -- test = iterate (<<*>> id3) id3 !! 1000000 iterate' f x = x : seq x' (iterate' f x') where x' = f x test' = iterate' (<<*>> id3) id3 !! 1000000 test'' = head $ drop 1000000 $ iterate' (<<*>> id3) id3

3 2

Creating static libraries using GHC
by Tom Hawkins 16 Mar '07

16 Mar '07

I have a chunk of Haskell code I would like wrap up and distribute as a library. Is there a way to build a static library (*.a) that includes my code plus the Haskell runtime, which C programs can easily link against? Here is what I have tried so far... ghc --make -fffi MyLib # Builds MyLib.o and MyLib_stub.o. gcc -c -I/usr/local/lib/ghc-6.4.1/include MyLibWrapper.c ar -r libMyLib.a MyLib.o MyLib_stub.o MyLibWrapper.o This works fine when I use ghc to compile and link a C program... ghc main.c libMyLib.a But if I use gcc, it throws a lot of unresolved references. I have no problem compiling with ghc, but the folks using my library probably won't have it installed. I added -v to ghc to see how it calls gcc; it seems to link in different libraries based on what Haskell libraries are being used, and it undefines a bunch is symbol references. Are there any switches to have ghc return a single archive with everything included? Please excuse my ignorance of the linking process. Thanks for any help! -Tom

3 2