December 2005 - Haskell-Cafe

About print and side-effects
by Daniel Carrera 18 Dec '05

18 Dec '05

Hi all, The recent responses to my first question (thanks guys!) included the following bit: main = print (fact 42) Now, print is a side-effect. Shouldn't it involve a do-block or a nomad or one of those scary things you hear about when learning about side effects in functional programs? How come 'print' is allowed to exist alone if it's a side-effect? Thanks for the help. Cheers, Daniel. -- /\/`) http://oooauthors.org /\/_/ http://opendocumentfellowship.org /\/_/ \/_/ I am not over-weight, I am under-tall. /

4 4

Substring replacements
by Daniel Fischer 18 Dec '05

18 Dec '05

Okay, I have looked up KMP and implemented it. Seems to work -- my first use of QuickCheck, too. It's slower than Bulat's and Tomasz' for Branimir's test :-(, but really fast for my test. Undoubtedly, one can still tune it. Here's the code: module KMP where import Data.Array searchReplace :: String -> String -> String -> String searchReplace "" _ str = str searchReplace src@(c:cs) dst str = process 0 str "" where len = length src pat = listArray (0,len-1) src bord = array (0,len) $ (0,-1):(1,0):[(i+1,boun i (bord!i)) | i <- [1 .. len-1]] boun i j | j < 0 = 0 | pat!i == pat!j = j+1 | otherwise = boun i (bord!j) getBord s n | m < 1 = m | s == pat!m = m | otherwise = getBord s m where m = bord!n process n str _ | n >= len = dst ++ process 0 str "" process _ "" mat = reverse mat process 0 (s:st) _ | s == c = process 1 st [s] | otherwise = s:process 0 st "" process n str@(s:st) mat | s == pat!n = process (n+1) st (s:mat) | otherwise = let j = getBord s n (ret,skip) = splitAt j mat in if j < 0 then reverse mat ++ process 0 str "" else reverse skip ++ process j str ret Cheers, Daniel

5 25

RE: [Haskell-cafe] +RTS -M800M
by Simon Marlow 17 Dec '05

17 Dec '05

On 16 December 2005 10:05, Joel Reymont wrote: > I'm trying to restrict GHC to 800Mb of heap at runtime by passing in > +RTS -M800M, the machine has 1Gb of memory and top shows free > physical memory dropping below 175Mb. I suppose I'm missing something > obvious or paying attention to the wrong statistics, Unix has a good > VM manager after all. Are my runtime options correct, though? -M800m should do more or less the right thing, but it is possible for GHC to exceed this figure by a small percentage. Rather than considering the "worst case" requirements for the next GC, GHC uses a more "average case" estimate, which sometimes ends up being wrong, but in most cases results in better utilisation of the available memory. Cheers, Simon

3 2

Darcs apply error while getting the complete GHC tree
by Joel Reymont 17 Dec '05

17 Dec '05

I thought I would get the _complete_ ghc tree so that I could build ghc with patches of my own. I decided to get the libraries while I'm at it. I got two errors and don't know what to do now. Do I need to get the whole tree again? Can I somehow reapply the patches that did not get applied? Should I just get the CVS trees instead? Thanks, Joel P.S. -- darcs get http://cvs.haskell.org/darcs/ghc.ghc-6.4/ Copying patch 13136 of 13136... done!es.ghc-6.4/ Applying patch 12 of 13136... Unapplicable patch: Thu Jan 11 14:26:13 GMT 1996 partain * [project @ 1996-01-11 14:06:51 by partain] darcs failed: Error applying hunk to file ./ghc/includes/rtsTypes.lh -- and then -- darcs get http://cvs.haskell.org/darcs/libraries.ghc-6.4/ ... Copying patch 3403 of 3403... done! Applying patch 2115 of 3403... Unapplicable patch: Sat Mar 27 14:07:36 GMT 2004 panne * [project @ 2004-03-27 14:07:33 by panne] Moved around the sources and the test examples to the standard places in the hierarchical libs. The NHC98 build is currently untested, but it might work, nevertheless... :-} darcs: ./HUnit/Test: createDirectory: already exists (File exists) -- http://wagerlabs.com/

4 4

Re: [Haskell-cafe] Darcs apply error while getting the complete GHC tree
by Joel Reymont 17 Dec '05

17 Dec '05

I thought I did :-). I just didn't get to 7.1. My bad and apologies! On Dec 17, 2005, at 7:07 PM, wld wrote: > Well, RTBG (Building Guide) ! > When building from CVS (I think, from darcs repo too), > you should do: > > autoreconf > ./configure -- http://wagerlabs.com/

1 0

Haskell School of Expression?
by Creighton Hogg 17 Dec '05

17 Dec '05

Hi, Has anyone recently tried to install the graphics libraries included on http://haskell.org/soe/software.htm ? I've tried to make the libraries, but I get these errors. make: Entering directory `/home/wchogg/coding/haskell/GraphicsLib/lib/x11' ffihugs +G +L"X_stub_ffi.c" X.hs Warning: unknown toggle `G'; ignoring. Warning: unknown toggle `L'; ignoring. runhugs: Unable to initialise Hugs (Unrecognised option ) I honestly can't figure out what the incompatibility is? I'm using version 20050308 of Hugs

2 1

Barrier implementation
by Peter Eriksen 17 Dec '05

17 Dec '05

Greeting, Something is not working for me, and I could use some more eyes on this. What I'm trying to accomplish is to implement a simpel barrier for ten worker threads (id = 0..9) using STM. With or without the barrier, the program produces an unordered interleaving of the output from the workers. Here's what I get with the program below: $ ghc --make Main.lhs $ a.out 0134568027913457896012579026813423904671238455702468159367839684012570279134685049137825901642375689134057892610462578903156012389473268457910267801345923924567801304689235714013679458256702465913878... And here's what I get without the line "atomically $ barrier tv id": $ a.out 1249056782934567210845619720538461975203698469175203469850123485076912348579406123894625738942106381592740631859274163841092315768491302578416930728254169302785693024917853029640217390856490... The first run should've been something like: 012345678901234567890123456789012345... since each worker thread 0..9 should write its id out once per iteration, and the workers should iterate in sync. Here's the code: > module Main where > > import Control.Concurrent > import Control.Concurrent.STM > import System.Random > > worker :: Int -> TVar Int -> IO () > worker id tv = do > sleepingTime <- randomRIO (0, 50000) > threadDelay sleepingTime > putStr $ show id > atomically $ barrier tv id > worker id tv Each worker sleeps for some time, then outputs its id and waits at the barrier for all the other workers to finish their sleep+output. > barrier :: TVar Int -> Int -> STM () > barrier tv id = do > passed <- readTVar tv > if (passed `mod` 10 == id) > then writeTVar tv (passed+1) > else retry The barrier is simply a global variable, tv, which holds the number of times any worker passed the barrier. Now, a worker may only pass the barrier iff the worker with an id one less just passed, or else it should block. > main :: IO () > main = do > tv <- atomically $ newTVar 0 > for [0..9] $ \i -> forkIO $ worker i tv > threadDelay (10*10^6) > > for = flip mapM_ The main thread just initializes the pass counter, starts 10 worker threads, and waits for ten seconds. I'd like to hear some comments on the approach, and perhaps even some insight into why it doesn't work. Regards, Peter Eriksen

6 9

RE: Re[2]: [Haskell-cafe] Substring replacements
by Branimir Maksimovic 17 Dec '05

17 Dec '05

>From: Bulat Ziganshin <bulatz(a)HotPOP.com> >Reply-To: Bulat Ziganshin <bulatz(a)HotPOP.com> >To: "Branimir Maksimovic" <bmaxa(a)hotmail.com> >CC: daniel.is.fischer(a)web.de, Haskell-Cafe(a)haskell.org >Subject: Re[2]: [Haskell-cafe] Substring replacements >Date: Fri, 16 Dec 2005 16:51:32 +0300 > >Hello Branimir, > >Friday, December 16, 2005, 5:36:47 AM, you wrote: >BM> I've also performed tests on dual Xeon linux box and results are > >just to let you know - GHC don't uses pentium4 hyperthreading, >multiple cpus or multiple cores in these tests Oh yes it does. I clearly see multiple threads that take unequal percentage on each virtual CPU. I guess that other thread is garbage collector thread. In case of hyperthreading, speed is gained by reduced memory latency by 30-60 % > >only way to make ghc using multiple processors is to use 6.5 beta >version, compile with "-smp" and explicitly fork several threads This is not the case as I see. On windows search replace test programs spawn 3 threads and on linux I'm not sure, but I've checked program that calls Haskell from C++ and GHC spawns additional thread, which is not my thread, that also performs something constantly, and I didn't spawn any thread from Haskell. So hyperthreading helps as it helps to optimise when several threads accesses memory as is in this test case. I can't see any other explanation why KMP search is slower on AMD 20% , but faster on Intel 30%, then straightforward search with my test. Greetings, Bane. _________________________________________________________________ Express yourself instantly with MSN Messenger! Download today it's FREE! http://messenger.msn.click-url.com/go/onm00200471ave/direct/01/

1 0

Optimizing a high-traffic network architecture
by Joel Reymont 17 Dec '05

17 Dec '05

Folks, In my current architecture I launch a two threads per socket where the socket reader places results in a TMVar and the socket writer takes input from a TChan. I also have the worker thread the does the bulk of packet processing and a timer thread. The time thread sleeps for a few minutes and exits after posting a timeout event if it hasn't been killed before. My goal is to launch poker 2,000 bots that join the server "lobby" and sit there sending small keep-alive packets every few minutes. The ultimate goal is for 4,000 bots to be playing but I'm taking it one step at a time. This is Mac OSX Tiger with a couple of header files modified to allow FD_SETSIZE of 10240. This is the maximum allowed by 'ulimit -n'. I'm also running ghc 6.4.1, compiled after FD_SETSIZE has been increased. I can get to 2k bots without any trouble if I use a keep-alive timeout of 9 minutes. Memory usage with 2k bots is 161Mb of physical memory and 262Mb VM. CPU usage 20-40%. Memory usage is constant once all bots have been launched. With a 1 minute keep-alive timeout system is starting to get stressed almost right away. There's verbose logging going on and almost every event/packet sent and received is traced. The extra logging of the timeout events probably adds to the stress and so, I assume, do the extra packets. New bots are being launched very slowly even with just 200 bots already running. Based on the above, would you have any suggestions for an improved architecture? I will try 1) disabling logging alltogether and 2) increase thread stack size to 3k (+RTS -k3k) as per Simon Marlow's suggestion. As per simon if a thread stack space is between 2k and 4k then each thread gets its own memory block (right Simon?) and threads are not GCd then. I'm a bit concerned about trippling my memory use with -k3k, though. I'm not sure if switching to a continuations-based framework will help me. Has anyone tried this? Thanks, Joel -- http://wagerlabs.com/

6 27

RE: [Haskell-cafe] Battling laziness
by Simon Marlow 16 Dec '05

16 Dec '05

On 16 December 2005 15:23, Joel Reymont wrote: > Looking at http://wagerlabs.com/randomplay.hd.ps I see closures > (constructors?) in this order > > <Script.Array.sat_s46N> > W8# > I# > <Script.Array.fromIntegral_s453> > <Script.Endian.sat_s1WxM> >> > <Script.Endian.sat_s1WF2> > W16# > <Script.PicklePlus.sat_s38YS> > stg_ap_2_upd_info Ok, so your heap is mainly full of (a) thunks generated by something in Script.Array, (b) Word8s, and (c) Ints. > This tells me it's something having to do with array code. I'm > attaching the Script.Array module at the end. This report does not > tell me who is retaining the data, though. > > Looking at http://wagerlabs.com/randomplay.hy.ps I see types ordered > like this > > * > Word8 > Int > ->* > [] > Char > Word16 > TableInfo interesting... Word8 and Int correspond to the -hd output above, but '*' indicates that the type of the <Script.Array.sat_s46N> is polymorphic. Completely polymorphic closures like this are usually (error "something"), which is a silly thing to fill up your heap with :-) I'm a bit mystified though, because looking at the code for Script.Array, all your arrays are unboxed, so I don't know where all the Word8s and Ints are coming from. It might be useful to do "+RTS -hyWord8 -hc" to see who generated the Word8s. Oh, and it looks like you aren't doing -auto-all, that would probably be helpful. Cheers, Simon

2 5