December 2008 - Haskell-Cafe

RE: Haskell Performance
by Simon Peyton-Jones 29 Dec '08

29 Dec '08

Dear Dave | My question to you is, if you were asked to make a recommendation on | which programming language to use for a program which would run on a | multiple core processor, but also required adequate memory and running | time performance, would you lean towards or away from Haskell? What | other factors would you consider? In other words, was I correct in my | initial opinion that Haskell was more concerned about code expression | rather than performance? Thanks for your email. I'm ccing the Haskell Cafe, which is full of people with practical experience (many more than I) of using Haskell for real applications. I don't think it's possible to give a categorical yes/no answer to your question. Yes, Haskell is designed primarily for high-level expressiveness rather than performance. This is good for programmers, but it cuts both ways for implementation. On the one hand, the programmer is taking less control, and the implementation may not be as clever as a programmer could be (the quicksort example you cite is a good example). On the other hand, the high-level-ness gives the opportunity for the implementation to do stuff that the programmer could not feasibly do by hand; the vectorisation transformation in nested data parallelism is a good example (http://research.microsoft.com/~simonpj/papers/ndp/index.htm) Long term, my bet is that the second trend will win. If you want to write parallel programs, it makes sense to start with a language that is by-default parallel, rather than by-default sequential. But that does not mean that you get parallel performance without tears. Far from it: parallelism is a complex, multi-faceted beast (concurrency, communication, locality, granularity...). So the short-term answer might be different to the long term one. That said, GHC comes out very competitively in the (flawed, but available) language shootout http://shootout.alioth.debian.org/u64q/benchmark.php?test=all&lang=all. So performance is not at all bad. Simon | -----Original Message----- | From: Dave Roberts [mailto:dave.a.roberts@gmail.com] | Sent: 15 December 2008 14:42 | To: Simon Peyton-Jones | Subject: Haskell Performance | | Dear Mr. Peyton Jones, | | I recently watched a video of you giving a talk about type driven | testing in Haskell. | | http://video.google.com/videoplay?docid=- | 4991530385753299192&ei=PWFGSaazDJPuqAL779n1DA&q=simon+peyton+jones+testing&emb=1 | | I found it to be a very informative and entertaining presentation | about Haskell. About halfway through the video you examine the | performance of the n-shell problem in Haskell versus C++. You say the | functional solution to the problem is 100x faster than the C++ | implementation. You continue to examine reasons why, citing that set | operations in a purely functional language can make assumptions that | side-effect laden languages cannot. For example creating a union of | two sets does not need to allocate a new set in memory because it can | reuse large portions of the original sets. You continue to show that | purely functional languages lend themselves to parallelism better than | imperative languages because state is not being changed by default. | | Taking this into consideration, the Haskell website has a section | titled, When C is better. | | http://www.haskell.org/haskellwiki/Introduction#When_C_is_better | | (I suspect you have already read this if not wrote it yourself) This | section looks at the quick sort example presented in the wiki and | claims that if written in an imperative language, the quick sort can | have time and memory requirement advantages over the Haskell version. | I read the haskellwiki long before viewing the video and at the time I | branded Haskell as a language which stresses expression over | performance. | | My question to you is, if you were asked to make a recommendation on | which programming language to use for a program which would run on a | multiple core processor, but also required adequate memory and running | time performance, would you lean towards or away from Haskell? What | other factors would you consider? In other words, was I correct in my | initial opinion that Haskell was more concerned about code expression | rather than performance? What you suggest in the video seems to be to | the contrary. | |

1 0

Incremental trasnformations (not Haskell topic)
by frantisek kocun 29 Dec '08

29 Dec '08

Hi, this is not a Haskell question but I'm not sure if anyone in java community can answer this. And I'm sure every Haskell programmer know OO programing. I'm programming one big boring db application and making a MDD tool to model and generate such applications. Many times object have computed attributes: E.g. value attribute of class Invoice is sum of values of all associated classes Item (Invoice 1 <>---> * Item) So value attribute of invoice is computed every time it is demanded. So far easy.Sometimes it is necessary not to compute it when it is needed but to have it precomputed (computed it every time, item is added/removed to invoice or item value is changed). Contract of attribute value is the same but it is not obvious from code if you program it in event-based as if you program it with expression which is computed every time. Event-based=no big picture. Expression=big picture but little performance. Question: Is there any way how to transform the OO transformation (invoice value is a sum of values of items) to incremental transformation (update invoice value, when item is created/deleted/changed)? Is there any product doing it. I am not interesting in product itself I want only see how it works? Is there any science paper on this? I call this not a function but a OO transformation because it can be much more complicated. I know it is not possible for all the functions (transformations), you have to do undo step, but it is possible for that what we use. One more complicated example: Provider provides money to receiver. It provides money with special code. Receiver sees all the money from all the providers grouped by the code and summed. Every summed group on the code has a subgroup summed on the provider. Provider A provides to R 10 on code "ABC" Provider A provides to Q 10 on code "ABC" Provider A provides to R 5 on code "ABC" Provider A provides to R 20 on code "DEF" Provider B provides to R 30 on code "ABC" Reciever R - recieves 45 on code "ABC", 15 from A, 30 from B - recieves 20 on code "DEF", 20 from A Reciever Q - recieves 10 on code "ABC", 10 from A This example can be easily programmed declarative way in Scala the way that everytime everything is computed. But have no clue if it can be transformed automatically to incremental way. Hope you understand what I mean. Thanks for your help and opinions Fero

5 7

Cross compiling from Linux to windows?
by Erik de Castro Lopo 29 Dec '08

29 Dec '08

Hi all, I've googled for this but haven't really found an answer yet. Is it possible to build (or better yet download) a GHC compiler that runs on Linux and generates windows binaries? I already use the MinGW C compiler to generate windows binaries from C code and I run my test suite under WINE. This has been working very well for me for some time and now I'd like to do the same for Haskell code. Cheers, Erik -- ----------------------------------------------------------------- Erik de Castro Lopo ----------------------------------------------------------------- Those who do not understand Unix are condemned to reinvent it, poorly. -- Henry Spencer

3 3

Data.ByteString join
by Galchin, Vasili 28 Dec '08

28 Dec '08

Hello, http://cvs.haskell.org/Hugs/pages/libraries/base/Data-ByteString.html but vigalchin@ubuntu:~$ ghci GHCi, version 6.8.2: http://www.haskell.org/ghc/ :? for help Loading package base ... linking ... done. Prelude> :m +Data.ByteString Prelude Data.ByteString> :t join <interactive>:1:0: Not in scope: `join' Prelude Data.ByteString> Why no "join" function? Regards, Vasili

4 5

Transactional container for storing anonymous deletable objects
by John Ky 28 Dec '08

28 Dec '08

Hi, I need a container data structure for storing anonymous objects - most likely things that have not value such as STM (), but probably other things as well. This will allow me to later on, iterate over the container and process those objects. Additionally I have the requirement that I need to be able to remove individual objects from this container. It needs to be linear time. main = do container <- nil key1 <- cons 1 container key2 <- cons 2 container key3 <- cons 3 container key4 <- cons 4 container key5 <- cons 5 container unlink key3 container unlink key2 container unlink key4 container list <- toList container putStrLn (show list) The above should give: [1, 5] What should I use? Thanks -John

4 3

Request for feedback: Understanding Haskell Monads
by Ertugrul Soeylemez 28 Dec '08

28 Dec '08

Hello fellow Haskellers, In the last few weeks I have written a comprehensive tutorial about Haskell monads [1], and I was hoping to get some constructive feedback. I'd appreciate any well meant criticism. [1] http://ertes.de/articles/monads.html Greets, Ertugrul. -- nightmare = unsafePerformIO (getWrongWife >>= sex) http://blog.ertes.de/

5 6

generating random lists until some predicate holds
by Nicholas O. Andrews 28 Dec '08

28 Dec '08

Hi all, What's the best way to implement the following Python code in Haskell? It is purposefully written in a functional style (and as a result will kill your recursion stack every other run). # begin Python from random import * def genList (): return [randint(0,9) for x in range(10)] def randWhile (predicate): result = genList () if predicate(result): return result else: return randWhile (predicate) def allEven (list): return reduce(lambda x,y: x and y, [x%2 == 0 for x in list]) print randWhile (allEven) # End Python Thanks!

3 2

Fram
by Haakon Meland Eriksen 28 Dec '08

28 Dec '08

Dear Haskell community, It was kindly suggested to me by John Peterson that I might reach you here and tell you a little bit about Fram, a small Mediawiki extension. Fram lets you intersperse your explanation with actual code, and then Fram extracts and concatenates the code, giving you a file with code for download. This means anyone with access to Fram can improve your code, either by explaining it or by enhancing the code itself. It is released under the GNU GPL version 3 at no cost. Fram is now at version 0.2, and equals the Mediawiki extension RawFile 0.2, written by Belgian developer Philippe Teuwen. Phil and I are collaborating to improve Fram into a fully fledged literate programming tool like "noweb" written by Norman Ramsey. It was originally Ramsey who suggested to me that Fram may be interesting to the Haskell community wiki, so here we are. If this sounds useful, please have a look at http://far.no/fram/ or go below the main-deck into the cabins to explore the articles http://far.no/fram/index.php?title=Fram http://far.no/fram/index.php?title=Test http://far.no/fram/index.php?title=Usbpix and if these don't make sense, have a look at our help page http://far.no/fram/index.php?title=Help:Contents If you are curious about the name, "Fram" is also the name of a famous ship used by Norwegian polar explorers in the last century. It means "Forward". Yours sincerely, Haakon Meland Eriksen

2 3

Missing Network Functions
by John Van Enk 28 Dec '08

28 Dec '08

While working on a project, I discovered that hton[sl] and ntoh[sl] are missing from the networking libraries. It appears there was some justification for this, but I've fallen upon a few instances where things were *not* working as expected. (Specifically when newtype'ing around HostAddress.) As such, I'm putting together a few extra functions that I think should exist in Network somewhere. I'm calling the package Network.Util (but am open to changes). Currently this only has: - htons - htonl - ntohs - ntohl Before I drop this on hackage: 1. Is there a reason I shouldn't do this? 2. Are there other suggestions for missing functions? (preferably cross-platform suggestions) Thanks all. /jve

5 6

chroot a single thread
by Jeremy Shaw 28 Dec '08

28 Dec '08

Hello, I am working on an application where I would like to chroot a thread, but I am not seeing a way to do it. I already have code which can run an IO action inside a chroot. The type signature is: fchroot :: FilePath -> IO a -> IO a The first argument is the new root, the second argument is the action to run with that root. You can see the implementation here if you are curious: http://src.seereason.com/build-env/Chroot.hs The problem with that function is that chroot affects the root of the whole process. In a single-threaded program this is (possibly) ok, because the original root is restored after the IO action completes (though with unsafeInterleaveIO, perhaps bad stuff will happen). In a multithreaded program it can be disasterous. As far as I know, the only solution would be to implement a function like fchroot which looks more like: pchroot :: FilePath -> IO () -> IO ExitCode Here the IO action would be forked off and run in a whole new process so that changing its root does not affect other 'threads'. Of course, you also can't use any of the nifty Haskell intra-thread communication stuff. Basically you can pass some values in and get back an exit code. However, I am not even sure how to implement a useful version of pchroot. In theory, I can just use forkProcess, but, the Giant Warning for forkProcess indicates that this will not be very useful in practice: forkProcess comes with a giant warning: since any other running threads are not copied into the child process, it's easy to go wrong: e.g. by accessing some shared resource that was held by another thread in the parent. Another example is the I/O manager thread: since the I/O manager isn't running in the child, attempting to do any Handle-based I/O will deadlock. Anyone have an ideas ? Thanks! - jeremy

4 4