December 2008 - Haskell-Cafe

Strange error with HDBC-odbc and MySQL
by Justin Bailey 30 Dec '08

30 Dec '08

I have run into a weird bug with HDBC-odbc that only occurs on my unix system. When I execute a select which returns more than 435 rows twice in the same process, the second execution fails with this error message: SQL error: SqlError {seState = "[\"HYT00\"]", seNativeError = -1, seErrorMsg = "connectODBC/sqlDriverConnect: [\"1045: [unixODBC][MySQL][ODBC 3.51 Driver]Access denied for user 'dummy'@'localhost' (using password: NO)\"]"} I think HDBC-odbc is at fault because the same query works fine from ruby. In both cases the unixODBC library is used. isql (the unixODBC command line utility) also has no problem. I have these software packages installed: ghc 6.8.3 HDBC 1.1.5 HDBC-odbc 1.1.4.4 unixODBC - 2.2.11 MySQL - 5.0.38 Linux - 2.6.16-xenU The program below demonstrates the problem. It does require a table with 500+ rows. On most production servers, information_schema.columns will usually do the trick. \begin{code} import System.Environment import Database.HDBC import Database.HDBC.ODBC import System.IO (hPutStrLn, stderr) import qualified Data.ByteString.Lazy.Char8 as L -- Execute with DNS name as only argument. E.g.: -- ./TestMain localdb main = do [dsn] <- getArgs testMatches dsn testMatches dsn testMatches dsn = do let handler e = do hPutStrLn stderr ("SQL error: " ++ show e) -- Execute a statement and return all the rows getAllRows conn sql = do stmt <- prepare conn sql _ <- execute stmt [] fetchAllRows' stmt handleSql handler $ do conn <- connectODBC $ "dsn=" ++ dsn -- 435 rows works fine, but not 445. rows <- getAllRows conn "select 1 from information_schema.columns limit 445" let evaledRows = filter (\[id] -> (fromSql id) /= (-1 :: Int)) rows putStrLn $ "Got " ++ show (length evaledRows) \end{code}

2 3

Re: [Haskell] hsc2hs can't find gcc under windows
by Duncan Coutts 30 Dec '08

30 Dec '08

On Thu, 2008-12-25 at 15:17 +0900, Curt Sampson wrote: > I'm trying to upgrade from ghc 6.8.3 to 6.10.1. Unfortunately, there > seem to be problems with the Windows version. (I've installed it on > two different machines, both Windows XP, and they've both had the same > problem. Both worked fine with 6.8.3.) > > When I run hsc2hs, on an hsc file, I get the error, "Can't find gcc". > > Just as an experiment, I tried adding c:\ghc\ghc-6.10.1 to my path. That > lets it find gcc, but then it can't find cc1. If I go even further down > this road and add c:\ghc\ghc-6.10.1\gcc-lib to my path, it can find cc1, > but then I get a stream of errors about not being able to find header > files, such as HsFFI.h. > > Does anybody have any clues here to help me? I can provide code, or even > access to the machine, if needed. So here's the story: Up until ghc-6.10, ghc shipped a fork of hsc2hs. The forked version used ghc as the C compiler rather than gcc. This had the advantage that it would know about ghc packages and the include files shipped with packages. The disadvantage was that it was a fork and behaved differently than using hsc2hs with hugs. It also meant that real C compiler options had to be double-escaped to get hsc2hs to pass them to ghc and get ghc to pass them to gcc. In ghc-6.10, ghc ships the normal hsc2hs. What we did not notice before 6.10.1 was released was that calling hsc2hs on its own now does not work without passing additional flags. We never noticed because few people actually do call hsc2hs directly, it is typically used via Cabal. Cabal tells it where gcc lives and what include dirs to use. So that's why at the moment in the Windows installer, hsc2hs cannot even find gcc, because gcc is not on the %PATH% and hsc2hs does not know about ghc's special bundled version of gcc. It also does not know about ghc's include directories. It's clear we'll need to think a bit more about how to get hsc2hs to be both a compiler neutral tool where we can share a single instance between multiple compilers and also have it work "out of the box". > We're also happy to pay for support, if someone out there is offering > commercial support for ghc. We offer commercial support through Well-Typed: http://www.well-typed.com/ Duncan

2 2

IORef vs TVar performance: 6 seconds versus 4 minutes
by Jim Snow 30 Dec '08

30 Dec '08

I decided to try to implement a graph algorithm using STM. Each node in the graph has a set of TVar-protected lists of the nodes it links to and the nodes that link to it. Also, there is a global TVar-protected Data.Map that contains all the nodes in the graph, indexed by name (which is polymorphic): data Node k r = Node { fwdPos :: TVar [Node k r], -- forward links (nodes we like) fwdNeg :: TVar [Node k r], -- we allow "negative" links, too (nodes we don't like) revPos :: TVar [Node k r], -- backlinks (nodes that like us) revNeg :: TVar [Node k r], -- negative back links (nodes that don't like us) currRep :: r, -- extra user-defined data name :: k -- node's unique identifier } deriving Show data Network k r = Network { node :: TVar (M.Map k (Node k r)), -- map of nodes by name trusted :: TVar [Node k r] -- a list of nodes we need to iterate over occasionally } deriving Show I tried loading a datafile of about 20,000 nodes into the graph in one big transaction, and found that it takes about 4 minutes. This seemed rather slow, so I replaced all the TVars with IORefs (and substituted STM with IO in the type signatures), and the same operation with the new version took about 6 seconds! This is all with one thread, so there should be no contention for the TVars. Is there something about STM that makes it scale worse than linearly wrt the number of mutations in a transaction? Above performance numbers are for ghc-6.10.1. With ghc-6.8.3, the STM version takes more than 9 minutes. According to profiling, one of my trouble spots is this function, which just adds an entry onto a TVar [a]: stmcons :: k -> TVar [k] -> STM () stmcons x tv = do xs <- readTVar tv writeTVar tv (x:xs) This seems like it ought to be pretty innocuous, unless the whole list is getting evaluated each time I cons a new entry, or if readTVar or writeTVar are much more expensive than they appear. -jim

6 10

Timing difference
by Aai 30 Dec '08

30 Dec '08

Programming for Rosetta Codes task http://www.rosettacode.org/wiki/Probabilistic_Choice I observed a remarkable (that's to say: for me) difference in timing using lambda function in the one case and point free in the other. Timing was measured in GHCi. Compiled there's no difference! Using lambda ============ ... labels = ["aleph", "beth", "gimel", "daleth", "he", "waw", "zayin", "heth" ] piv n = take n . flip (++) (repeat ' ') main = do g <- newStdGen let rs :: [Float] rs = take 1000000 $ randomRs(0,1) g ps, sps :: [Float] ps = ap (++) (return. (-) 1 .sum) $ map recip [5.0..11.0] sps = scanl1 (+) ps ix = map ((/1000000.0).fromIntegral.length). group.sort $ map (\x -> fromJust $ findIndex not $ map (x>) sps) rs mapM_ (\(l,s,c)-> putStrLn $ (piv 6 l) ++ " " ++ (piv 12 $ show $ s) ++ " " ++ ((piv 12 $ show $ c) )) $ zip3 labels ps ix Using point free ================ replace $ map (\x -> fromJust $ findIndex not $ map (x>) sps) rs by $ map (fromJust. findIndex not. flip map sps. (>)) rs lambda: ======= *Main> main ... (18.24 secs, 2524489600 bytes) point free: =========== *Main> main ... (12.15 secs, 2470893260 bytes) Should I program using point free as much as possible? :-) Thanks =@@i

4 5

ANN: Control.Monad.IfElse 0.85
by Jeff Heard 30 Dec '08

30 Dec '08

Updated Control.Monad.IfElse to include Wren Thornton's Control.Monad.Extras, unlessM, ncond (unless-else chaining), ncondM, and more binary conditional operators.

3 2

represent data sturcture using function
by Raeck chiu 30 Dec '08

30 Dec '08

Merry Christmas! Hope everybody is enjoying the Christmas! I am doing some readings and I found something seems to be interesting. People sometime will try to represent a quantity-regard-only data structure (such a order-regadless List) using functions instead of ta concrete data structure (like the haskell build-in []), any particular reason for this? For example, > data Sex = Male | Female > classA :: Sex -> Int > classA Male = 100 > classA Female = 200 when I should prefer the above solution instead of using a concrete data structure such as [] to represent the classA members? It seems to be very difficult to change the number of Male or Female if a concrete data structure is not used. Is it possible change the number of Male in classA when represent classA using function? Thank you very much! Best wishes, Raeck _________________________________________________________________ It’s the same Hotmail®. If by “same” you mean up to 70% faster. http://windowslive.com/online/hotmail?ocid=TXT_TAGLM_WL_hotmail_acq_broad1_…

5 6

Maintaining laziness
by Henning Thielemann 30 Dec '08

30 Dec '08

In case someone cares - after some battles with functions that are less lazy than expected, I have written a tutorial on how to get functions lazy and how to test, whether they are actually lazy: http://www.haskell.org/haskellwiki/Maintaining_laziness I found it especially enlightening, that one can avoid a 'force' function in a lazy parser by making the possibility of a parser failure explicit in its type. I.e. a parser that cannot fail, need no 'force'. (I learnt that in polyparse the 'force' function is hidden in 'apply': http://hackage.haskell.org/packages/archive/polyparse/1.1/doc/html/Text-Par… )

4 5

Data.List.sort, not so good?
by Adrian Neumann 29 Dec '08

29 Dec '08

Hello Haskell-Cafe, lately I've been playing around with sorting. I discovered that Data.List.sort is not as optimized I thought. At least not for random lists. I think we should consider adding a Data.List.Sort package to hackage (Similar to Data.List.Split) that offers a wide variety of sorting algorithms. I don't consider myself to be a very advanced Haskell programmer, but I could come up with a Mergesort that beats List.sort, time- and spacewise. > http://hpaste.org/13403 On my machine List.sort takes ~10 sec, mergeSort 7, qs 4 (compiled with -O2). List.sort eats too much ram to sort 20.000.000 ints, my algorithms don't. QuickCheck says my implementations are correct. I suppose List.sort does clever things for nonrandom lists, that I didn't consider? I had a look at the source, but I couldn't find anything obvious. Adrian

3 3

Trouble with the ST monad
by Andre Nathan 29 Dec '08

29 Dec '08

Hello, I'm trying to write a function that would take an STArray and and shuffle its elements. I'm having trouble with the ST monad, though, and couldn't find out how fix this issue. The problem happens when I use runST to extract the shuffled array from the ST monad. I'm getting the following error: Inferred type is less polymorphic than expected Quantified type variable `s' is mentioned in the environment: a :: STArray s Int a The full code is at http://hpaste.org/13240#a1 Any help would be appreciated. Thanks, Andre

4 7

Using classes for a heterogeneous graph structure
by Jeremy Bowers 29 Dec '08

29 Dec '08

I'm trying to use a class to create a heterogeneous graph structure. Part of this involves holding things in a Data.Map identified only as "instances of some class", but then I can't figure out how to actually use these things. What is mystifying me is exemplified in the following code: ----------------- module Sample where import qualified Data.Map as Map class Thing thing where set_int :: thing -> Integer -> thing -- for wrapper wrapper :: thing -> ThingWrapper wrapper thing = ThingWrapper thing instance Thing Integer where set_int me i = i -- in the generic case, this actually does something type ThingsByString = (Thing thing) => Map.Map Integer thing update :: Integer -> Integer -> ThingsByString -> ThingsByString update key value map = let (Just thing) = Map.lookup key map in Map.insert key (set_int thing value) map test = let my_map = Map.empty::ThingsByString map1 = Map.insert 0 1 my_map map2 = update 0 8 map1 in map1 data ThingWrapper = forall t. (Thing t) => ThingWrapper t instance Thing ThingWrapper where set_int (ThingWrapper thing) i = wrapper $ set_int thing i wrapper thing_wrapper = thing_wrapper ----------------- This gives an error message (in GHC 6.8 w/ -fglasgow-exts , and yeah, that's deprecated, I'm still in dev): Sample.hs:27:16: Inferred type is less polymorphic than expected Quantified type variable `thing' is mentioned in the environment: map1 :: Map.Map Integer thing (bound at Sample.hs:26:9) my_map :: Map.Map Integer thing (bound at Sample.hs:25:9) In the third argument of `update', namely `map1' In the expression: update 0 8 map1 In the definition of `map2': map2 = update 0 8 map1 Now, I sort of understand this message. What I don't understand is how to do what I want to do; I've rearranged things in a number of ways but this comes up every which way. It seems like once I have one of these "things" there's absolutely nothing I can do to it (or absolutely no way to do what update does, which is mutate it using only the defined class interface without ever knowing what's in there). What's annoying is that at least as a human, what I want seems well-defined to me. Moreover, as shown in the code above, I can create a "wrapper" structure, and if I do that and manipulate only the wrapper (not shown in this example), everything works as I'd like. (If that's the answer, that's OK with me, but it seems klunky, wasteful, and amenable to being automatically done by the compiler as I understand it, and I expect there's a reason it isn't.) My question is: How can I have this heterogeneous graph structure and actually be able to manipulate it solely through a class interface? Or, is it impossible? (I'm open to "you're doing it entirely wrong and you should do X", but in that case I'd like to know what using classes to specify the interface a node should conform to is the wrong thing, because it sure *seems* right, based on what the interface is and how classes are documented. And it would need to work for a heterogeneous graph structure, not my very-simplified example above.) ---------------------------------- Barracuda Networks makes the best spam firewalls and web filters. www.barracudanetworks.com

2 1