- Haskell-Cafe - Haskell.org

Question about GUM
by Changming Ma 21 Jun '02

21 Jun '02

Hi there, I'm getting confused of the run-time system behavior of GUM. When I ran a quicksort program in GUM with profiling flag, I got different "task" distributions in the PS file generated everytime, e.g., the peak of task number may be 19 at the first time and 48 at the second. While it's quite reasonable that the clock cycles vary dramatically, it's somewhat strange to me that the task distributions are so different. Any hints? I do appreaciate it. BTW, seems the program does not stop if the input list has more than 350 elements, althought it works fine otherwise. Will it get better if I use an SMP system? Thanks in advance. PS: I'm using GUM-4.06, PVM3, RH LINUX6.2 on a single processor system. Program and RTS flags quicksort +RTS -qP -sstderr grs2gr *quicksort.???.gr > quicksort.gr gr2ps -o quicksort.ps quicksort.gr --source code module Main(main) where import System(getArgs) import Parallel forceList :: [a] -> () forceList [] = () forceList (x:xs) = x `seq` forceList xs quicksortF::[Int]->[Int] quicksortF [] = [] quicksortF [x] = [x] quicksortF (x:xs) = (forceList losort) `par` (forceList hisort) `par` losort ++ (x:hisort) where losort = quicksortF [y|y <- xs, y < x] hisort = quicksortF [y|y <- xs, y >= x] args_to_IntList :: [String] -> [Int] args_to_IntList a = if length a < 1 then error "Parallel Quick Sort: no enough args \n" else map read a main = putStr ("get " ++ (show (quicksortF l))++"\n") l = [ 658986,123141,737929,681645,329651, ... 925104]

1 0

% 6,95
by finans400 19 Jun '02

19 Jun '02

1 0

Data Tracing in GPH
by Ma Changming 17 Jun '02

17 Jun '02

Hi there, Does anyone know how to trace the GPH program to see the data processed at specific processors? I'm using gum-4.06, PVM3, RedHat Linux6.2. When I execute the following QuickSort program, I want to make sure what happens with different CPUs, i.e., how the data are partitioned between processors? Is there any debugging tools to trace the data? Thanks in advance. Changming Ma PS: QuickSort module Main(main) where import System(getArgs) import Parallel forceList :: [a] -> () forceList [] = () forceList (x:xs) = x `seq` forceList xs quicksortF::[Int]->[Int] quicksortF [] = [] quicksortF [x] = [x] quicksortF (x:xs) = (forceList losort) `par` (forceList hisort) `par` losort ++ (x:hisort) where losort = quicksortF [y|y <- xs, y < x] hisort = quicksortF [y|y <- xs, y >= x] args_to_IntList :: [String] -> [Int] args_to_IntList a = if length a < 1 then error "Parallel Quick Sort: no enough args \n" else map read a main = getArgs >>= \ a -> let l = args_to_IntList a in putStr ("get " ++ (show (quicksortF l))++"\n") __________________________________________________ Do You Yahoo!? Yahoo! - Official partner of 2002 FIFA World Cup http://fifaworldcup.yahoo.com

2 1

reversing a list
by Matthias 15 Jun '02

15 Jun '02

Hi there, I thought i can reverse a list with reverse a list easily with the reverse function in order to achieve following: I wanted to check for a certain suffix (then reversed a prefix) and substitute that. So i did -- let dstfile = case tmpfile of -- ('a':'s':'p':'.':rest) -> rest ++ ".psx" -- anything -> anything ++ ".psx" Guess what: I did not work. dstfile was reversed (tmpfile ++ ".psx"). i guess this deals with lazy evalutation, but would be glad about some definit ideas. Matthias

1 2

Say�n haskell-cafe ,
by abdyesilkart.com 13 Jun '02

13 Jun '02

1 0

typeclass versioning
by Cagdas Ozgenc 12 Jun '02

12 Jun '02

Greetings. What happens if a type is made an instance of a typeclass in two different modules with different implementations? In static linking scenarios it will be caught by the compiler. I don't know whether we can do dynamic loading in Haskell, but it seems there will be inconsistencies in that case. Even in static linking when an object is serialized over the wire where a different implementation of a typeclass exists things will go wild. What's Haskell's approach to versioning? Thanks

2 1

Generating the n! permutations in Haskell
by Shlomi Fish 10 Jun '02

10 Jun '02

Included below is a Haskell Program I wrote to generate the n! permutations of a set of n elements. There are comments that explain it throughout the program. Both of the algorithms that appear in the cut the knot site use assignment, so are not very suitable for Haskell. Regards, Shlomi Fish -- (gradual_transfer set empty_set) -- -- Gradually pops elements out of set and [1..5] pushes them into -- empty_set, -- and makes a list of both stacks in their intermediate phases. -- -- E.g: -- gradual_transfer [1 .. 5] [] = -- [([1,2,3,4,5],[]),([2,3,4,5],[1]),([3,4,5],[2,1]), -- ([4,5],[3,2,1]),([5],[4,3,2,1])] gradual_transfer :: [a] -> [a] -> [([a],[a])] -- I stop when the list contains a single element, not when it contains -- no elements at all. The reason for this is that gen_perms like it -- better -- this way, as it has no use of a zero element (a:as). gradual_transfer (a:[]) ps = [((a:[]),ps)] gradual_transfer (a:as) ps = ((a:as),ps):(gradual_transfer as (a:ps)) -- (dump ps as) is equivalent to (reverse ps) ++ as, only it should -- be much faster. dump :: [a] -> [a] -> [a] dump [] as = as dump (p:ps) as = dump ps (p:as) gen_perms :: [a] -> [[a]] gen_perms [] = [[]] gen_perms set = [ (a:rest) | (a:as,ps) <- (gradual_transfer set []), rest <- gen_perms(dump ps as) ] print_perms [] = return () print_perms (a:as) = do print a print_perms as main = print_perms (gen_perms [1 .. 8]) ---------------------------------------------------------------------- Shlomi Fish shlomif(a)vipe.technion.ac.il Home Page: http://t2.technion.ac.il/~shlomif/ Home E-mail: shlomif(a)iglu.org.il He who re-invents the wheel, understands much better how a wheel works.

2 2

Re: HTTP.hs with slow execution
by Warrick Gray 09 Jun '02

09 Jun '02

[redirecting to haskell-cafe] Hi Jens, I think I have it. >I think I found an issue with the HTTP.hs library when >making a short-lived non-persistent connection to a server >from a slower machine (i486). I can also reproduce it by >running my program with strace or ltrace on a faster machine >(PIII). Last two meaningful recv call: >recvfrom(3, 0x500a3878, 1000, 0, 0x500a3c68) = 555 >getpeername(3, 0x500a3c8c, 0x500a3ca4, 0x080dfda0, 1) = 0 >recvfrom(3, 0x500a2008, 1000, 0, 0x500a23f8) = 0 >getpeername(3, 0x500a241c, 0x500a2434, 0x080dfda0, 1) = 0 After no data is received the sending part of the TCP stream is closed: >shutdown(3, 1, 0x0805b85f, 0x080e018c, 1) = 0 Being polite, the library ensures that all incoming data has been consumed, forgetting for a moment that the incoming stream has already closed: >recvfrom(3, 0x500a2440, 1000, 0, 0x500a2830) = 0 >getpeername(3, 0x500a2854, 0x500a286c, 0x080e018c, 1) = -1 The "recvfrom" seems fine, the "getpeername" call comes from the definition of Network.Socket.recvFrom, which features: flg <- sIsConnected sock -- For at least one implementation (WinSock 2), recvfrom() ignores -- filling in the sockaddr for connected TCP sockets. Cope with -- this by using getPeerName instead. sockaddr <- if flg then getPeerName sock else peekSockAddr ptr_addr ...Leading to the error in question: >strerror(107) = "Transport endpoint is >not connec"... The answer I guess is to alter all calls to 'recvFrom' to 'recv' calls (done), or replace socket reference with ConnClosed at the first sign of a closed socket. Unfortunately the second will have to wait, but the first is implemented and available in the latest version: http://homepages.paradise.net.nz/warrickg/haskell/http Warrick. PS - nice trace. _________________________________________________________________ Get your FREE download of MSN Explorer at http://explorer.msn.com/intl.asp.

2 1

problems figuring out what the type system is telling me
by Chris Moline 08 Jun '02

08 Jun '02

hi. i am really stuck on this problem and i cant seem to figure it out and so i am posting this in the hopes someone will help me. i realize you prolly get this kind of question alot but the wiki and the papers i have been reading arent helping to clarify things. i am trying to write a little utility to make it easier to get rid of ports on my machine. the function that is giving me trouble is getDepends. it takes the name of a port and opens the appropriate +CONTENTS file and returns the dependencies it finds. here is the function and its helper. getDepends :: String -> [String] getDepends p = do handle <- openFile (portsDir ++ p) ReadMode fetchDepends handle fetchDepends :: Handle -> [String] fetchDepends handle = do l <- hGetLine handle e <- hIsEOF handle case (not e) of -- ifs keep giving indent errors so ill just use case True -> case (matchRegex (mkRegex "^@pkgdep") l) of Just [a] -> [drop 8 l] ++ (fetchDepends handle) _ -> fetchDepends handle False -> "" here is ghci's error messages. Compiling Pheobe ( Phoebe.hs, interpreted ) Phoebe.hs:19: Couldn't match `[]' against `IO' Expected type: [t] Inferred type: IO Handle In the application `openFile (portsDir ++ p) ReadMode' In a 'do' expression pattern binding: handle <- openFile (portsDir ++ p) ReadMode Phoebe.hs:24: Couldn't match `[]' against `IO' Expected type: [t] Inferred type: IO String In the application `hGetLine handle' In a 'do' expression pattern binding: l <- hGetLine handle Failed, modules loaded: none. could someone be so kind as to explain what the problem is? sincerly, chris moline

4 4

type equivalency
by Cagdas Ozgenc 06 Jun '02

06 Jun '02

Greetings. I know 2 special type constructors(there might be other that I do not know yet) -> and ( , ) where structural type equivalency is enforced and we can also create new types with an algebric type constructor notation where name equivalency is enforced. What is the rationale? I mean why 2 special type constructors, but not 5, or 10 or N? Thanks for taking time.

5 8