January 2006 - Haskell-Cafe

Re: [Haskell-cafe] x86 code generation going wrong?
by Chris Kuklewicz 09 Jan '06

09 Jan '06

Brian Sniffen wrote: > The first couldn't even complete on my 2.26 GHz Celeron! It's only got > 512 MB of RAM, which may be part of the problem. I should not leak memory but it may be an optimization problem. Try explicitly using "ghc -O2 -funbox-strict-fields". > > Stack space overflow: current size 1048576 bytes. > Use `+RTS -Ksize' to increase it. > ./test1 < sum-file-test-input 25.33s user 2.89s system 18% cpu 2:32.02 total > > ./test2 < sum-file-test-input 4.79s user 0.20s system 94% cpu 5.276 total > > ./test3 < sum-file-test-input 4.46s user 0.14s system 99% cpu 4.623 total > > -- > Brian T. Sniffen > bts(a)alum.mit.edu or brian.sniffen(a)gmail.com > http://www.evenmere.org/~bts >

3 3

fastest Binary library!
by Bulat Ziganshin 08 Jan '06

08 Jan '06

Hello yes, i did it! today i spend time to optimize my own Binary library and got the (de)serialization speed about 50 mb/s with my 1 ghz cpu. it is a peek speed for unboxed arrays, in real life GC times and other overhead expenses will need much more time than (de)serialization itself. but at least noone can now say that serialization is a bottleneck of Haskell program :) i uploaded current version to http://freearc.narod.ru/Binary.tar.gz Main.hs in package is a testing program which builds, puts to memory buffer and then gets back a 40mb array and prints runtime of each operation Joel, if you are interested in switchinh to my library - write me. i have ideas about supporting your 150 records using your own DDL. smth like this: data Command = Cmd1 field1 Word32 333 field2 Word16 0 | Cmd2 field3 String "" field4 [SubRecord] [] you will describe all your records just one time and then these definitions will be compiled for the language/library you are using -- Best regards, Bulat mailto:bulatz@HotPOP.com

3 3

Project postmortem II /Haskell vs. Erlang/
by Joel Reymont 08 Jan '06

08 Jan '06

Simon, Please see this post for an extended reply: http://wagerlabs.com/articles/2006/01/01/haskell-vs-erlang-reloaded Thanks, Joel On Dec 29, 2005, at 8:22 AM, Simon Peyton-Jones wrote: > | Using Haskell for this networking app forced me to focus on all the > | issues _but_ the business logic. Type constraints, binary IO and > | serialization, minimizing memory use and fighting laziness, timers, > | tweaking concurrency and setting up message channels, you name it. > > That's a disappointing result. Mostly I think Haskell lets you > precisely focus on the logic of your program, because lots else is > taken > care of behind the scenes. You found precisely the reverse. > > It'd be interesting to understand which of these issues are > - language issues > - library issues > - compiler/run-time issues > > My (ill-informed) hypothesis is that better libraries would have > solved > much of your problems. A good example is a fast, generic > serialisation > library. > > If you felt able (sometime) to distil your experience under headings > like the above, only more concretely and precisely, I think it might > help to motivate Haskellers to start solving them. -- http://wagerlabs.com/

4 10

do {x<-[1,2,3]; True <- return (odd x); return x}.. why? (do notation, monads, guards)
by Marc Weber 08 Jan '06

08 Jan '06

Here is a simple program implementing the above function in 4 different ways.. See my comments to get to know where I have problems: ---------- begin test.hs ---------- module Main where import IO import Control.Monad.List {- list1,2 are both implementations of the same function f=[1,3] ;-) I've both rewritten with the translation rules for do notation to better understand what's going on and where the differences are -} list1=do { x <- [1,2,3]; True <- return (odd x); return x} list2=do { x <- [1,2,3]; guard (odd x); return x} -- <- provided by xerox list1rewritten :: [Int] list1rewritten=let ok x = let ok2 True = do return x --1r1 ok2 _ = fail "ok2" --1r2 in return (odd x) >>= ok2 --1r3 ok _ = fail "outer" --1r4 in [1,2,3] >>= ok {- The outer let .. in >>= is used to "call" the inner >>= for each element of [1,2,3] (the list Monad causes this) True <- return (odd x): really nice trick...! if x is odd then line --1r1 is matched the values is returned otherwise line --1r2 is matched calling fail which is implemented as = [] ignoring the message hence no element is added but I'm not sure which implementation of >>= is used in --lr3: It should satisfy (Monad m) => m Bool -> (Bool -> m Int), right ? Looking at the definition taken from GHC/Base.lhs: class Monad m where (>>=) :: forall a b. m a -> (a -> m b) -> m b and a sample implementation: instance Monad [] where m >>= k = foldr ((++) . k) [] m I wonder how a, b (from m a and m b) and m (from class Monad m) are renated? Can you tell me how the implementation declaration of m a -> (...) -> m b differs in these cases: eg: 1. a = Int, b=String 2. the other way round: a=String b=Int? -} list2rewritten :: [Int] list2rewritten = let ok x = guard (odd x) >> return x ok _ = fail "I think never used?" in [1,2,3] >>= ok {- Here ok is feeded with 1,2 and three due to the list Monad again? So fail will never be called, right? I also know that guard returns either the monad data type constructor mzero or return () But how is this used in combintation with >> return x::Int to return either [] or [x] ? -} main=do -- print result of all implementations to show that they are equal sequence [ print x| x <- [[1,3], -- [1,3] should be the result list1, list1rewritten, list2, list2rewritten ] ] -------------- end ------------------------- I hope there will be some time when I can say: Monads.. I don't bother anymore I'm practicing every night while dreaming.... ;-) Marc

3 3

Re: Shootout favoring imperative code
by Isaac Gouy 07 Jan '06

07 Jan '06

I sent a private email and the response to it has appeared on this mailing-list, so let me just correct some of the interpretations that have been made. > > You can say that again! > Ah..sarcasm, I know that one. No, it was emphatic agreement (the ordinary usage of that phrase). > > Is a persecution complex required for Haskell > > programming :-) This one is sarcasm. Hi Sebastian :-) > > I'm waiting for the complaints that binary-trees was designed > > to favour functional programming languages ;-) > ;-) == sarcasm Actually no - binary-trees was written with functional programming languages very much in mind. best wishes, Isaac __________________________________________ Yahoo! DSL Something to write home about. Just $16.99/mo. or less. dsl.yahoo.com

1 0

Re: Shootout favoring imperative code
by Chris Kuklewicz 06 Jan '06

06 Jan '06

igouy(a)yahoo.com wrote: >>There is no need to beat a dead horse, though. This benchmark sets out >>to test fgets / atoi, and that is all. There are better benchmarks to >>spend time on. > > > You can say that again! > Ah..sarcasm, I know that one. Actually, I submitted a slightly faster sum-file entry for Haskell tonight. So I did kick the horse around, and I learned how to use -funbox-strict-fields. > Is a persecution complex required for Haskell programming :-) The sum-file benchmark is not about my cleverness. It is designed to test the language's library routines. I quote Brent: > Yes -- it was designed as a test of the standard I/O > system. > > -Brent See...It is like the "startup" benchmark -- which just tests how long it takes to start the program (and print "Hello World."). > > ackermann, sum-file, random, startup (aka hello world) are all > left-over from the old Doug Bagley Great Computer Language Shootout - > they are just little snippets of nothing that provide an easy starting > point - takfp and harmonic are much the same. takfp and ackerman and harmonic can be good tests of how well the langauage handles recursion. > > I'm waiting for the complaints that binary-trees was designed to favour > functional programming languages ;-) ;-) == sarcasm > > best wishes, Isaac > Cheers, Chris

1 0

HDBC Roadmap
by John Goerzen 06 Jan '06

06 Jan '06

I've had some questions about my intentions for HDBC. The Current Plan is: Release 1.0.0 next week. Before that happens, I want to have a stable and useful API. I have one feature request to implement yet (examinine properties of arbitrary tables in the database). I also want to finish the ODBC driver first, which should happen yet this week. After that, I will probably also release a MySQL backend. I hope that HDBC can be a community effort. Towards that end, I hope to be able to get some trac pages going on somebody's server and perhaps a mailing list where ideas and implementations for future enhancements can be discussed. I'd also like to be able to help people that would like to write backends for additional databases. (I also should announce hsql-testsuite, the common HUnit suite that is used to test each database backend.) The development time for a HDBC backend looks like about 1 day per database, including the amount of time it takes to learn that database's C API. (A bit more for ODBC because its API is weird and the documentation is unhelpful) Post 1.0.0, I will spend some time optimizing performance in the existing database drivers. There are several opportunities to do that, and they should yield significant results for certain use cases (such as inserting thousands or millions of rows of data at once.) Comments and suggestions are quite welcome. Patches too ;-) -- John

3 3

Fannkuch Entry
by David F. Place 06 Jan '06

06 Jan '06

Hi All, I've been enjoying following the thread about the "shootout" programs and trying to understand the performance of the entries, especially the Fannkuch benchmark. I've been comparing Sebastion's and Bertram's proposals found on the wiki. On my machine (PowerBook g4 (512M), GHC 6.4.1), Bertam's is about 3x faster. I think the main reason for the difference is that Bertram's conses much less and keeps much less data live on the heap. I don't find Bertram's version especially hard to read. I tripped on '>>=', but it was easy to look it up. You can substitute "concatMap" without penalty if you hate monads. The technique used in Bertram's `flop' is known to anyone who has been introduced to streams. I think Bertram solution is an idiomatic "engineered" Haskell program. Cheers, David ------------------------------------------------- [Marcel:~/pfann] david% time ./Sebastion 9 123456789 ... Pfannkuchen(9) = 30 8.132u 0.250s 0:08.89 94.2% 0+0k 0+2io 0pf+0w ------------------------------------------------- [Marcel:~/pfann] david% time ./Bertram 9 123456789 ... Pfannkuchen(9) = 30 2.629u 0.100s 0:03.10 87.7% 0+0k 0+4io 0pf+0w ------------------------------------------------- ------------------------------------------------- ./Sebastion 9 +RTS -K256m -c -sstderr 123456789 ... Pfannkuchen(9) = 30 911,697,700 bytes allocated in the heap 3,138,348 bytes copied during GC 5,172 bytes maximum residency (2 sample(s)) 3478 collections in generation 0 ( 0.28s) 2 collections in generation 1 ( 0.01s) 2 Mb total memory in use INIT time 0.00s ( 0.00s elapsed) MUT time 7.85s ( 8.52s elapsed) GC time 0.29s ( 0.44s elapsed) EXIT time 0.00s ( 0.00s elapsed) Total time 8.14s ( 8.96s elapsed) %GC time 3.6% (4.9% elapsed) Alloc rate 116,139,834 bytes per MUT second Productivity 96.4% of total user, 87.6% of total elapsed ------------------------------------------------- ./Bertram 9 +RTS -K256m -c -sstderr 123456789 ... Pfannkuchen(9) = 30 271,671,720 bytes allocated in the heap 818,676 bytes copied during GC 5,220 bytes maximum residency (1 sample(s)) 1036 collections in generation 0 ( 0.08s) 1 collections in generation 1 ( 0.00s) 1 Mb total memory in use INIT time 0.00s ( 0.01s elapsed) MUT time 2.56s ( 2.90s elapsed) GC time 0.08s ( 0.09s elapsed) EXIT time 0.00s ( 0.00s elapsed) Total time 2.64s ( 3.00s elapsed) %GC time 3.0% (3.0% elapsed) Alloc rate 106,121,765 bytes per MUT second Productivity 97.0% of total user, 85.3% of total elapsed ------------------------------------------------------ -------------------------------- David F. Place mailto:d@vidplace.com

3 2

bug in WinHugs Version 20051031?
by Bulat Ziganshin 06 Jan '06

06 Jan '06

Hello the following program fails to load in Hugs2005 with all Haskell extensions enabled: {-# OPTIONS_GHC -fglasgow-exts -fallow-undecidable-instances #-} class (Monad m) => Stream m h where class (Stream m h) => CharStream m h where vGetChar :: h -> m Char vGetContents :: h -> m String vGetContents h = func where func = do vGetChar h return "" class (Stream m h) => ByteStream m h where vGetByte :: h -> m () instance (CharStream m h) => ByteStream m h where vGetByte h = undefined instance (ByteStream m h) => CharStream m h where vGetChar = undefined message printed is: ERROR file:.\SystemStreamClasses.hs - *** The type checker has reached the cutoff limit while trying to *** determine whether: *** CharStream a b *** can be deduced from: *** () *** This may indicate that the problem is undecidable. However, *** you may still try to increase the cutoff limit using the -c *** option and then try again. (The current setting is -c40) this program compiles ok in GHC 6.4.1. if call to `func` is replaced with its body, then program loads ok: vGetContents h = do vGetChar h return "" can problem with loading my code in Hugs be resolved by giving to `func` some type signmature? i can't replace call with its body because code in real program is much more complicated: vGetContents h = let loop = let func = do c <- vGetChar h next <- loop c `seq` return (c : next) handler e = if isEOFError e then return [] else mError e in mCatch func handler in do firstchar <- vGetChar h rest <- loop return (firstchar : rest) class (Monad m) => MonadHelper m where mError :: IOError -> m a mCatch :: m a -> (IOError -> m a) -> m a class (Show h, MonadHelper m) => Stream m h | h -> m where .... -- Best regards, Bulat mailto:bulatz@HotPOP.com

2 2

Re[2]: Joels Time Leak
by Bulat Ziganshin 05 Jan '06

05 Jan '06

Hello Simon, Wednesday, January 04, 2006, 7:33:22 PM, you wrote: >>>>The minimum time between context switches is 20 milliseconds. >> SM> Sure, there's no reason why we couldn't do this. Of course, even >> SM> idle Haskell processes will be ticking away in the background, so >> SM> there's a reason not to make the interval too short. What do >> SM> you think is reasonable? >> >> Simon, the talk is about changing GHC _tick_, which is a _minimal_ >> possible context switch interval SM> Yes, I know. in this case, why you say that idle processes will be ticking in background? it's entirely up to program writers/users if they wants to decrease this interval. by default switches will occur in the same 20 ms, lesser tick will have no impact on ordinary programs as long as default switching time will be not changed -- Best regards, Bulat mailto:bulatz@HotPOP.com

2 2