November 2014 - Haskell-Cafe

Re: [Haskell-cafe] Programming videogames in haskell
by Sean Seefried 24 Nov '14

24 Nov '14

I just updated the wiki [1] with a build script I wrote in Docker [2]. Putting together the build script in Docker was a very pleasant experience. I broke the script up into many small pieces so that if one particular part failed I could simply go back to the last checkpoint and start from there. Once I had finished the script, I also had very high assurance that it would work with no problems whatsoever when I ran it again from scratch. Normally putting together build scripts is difficult because of the inherently stateful nature of a file system. If you are working on a build script that takes a long time to run the temptation is not to start again from scratch every time something goes wrong. The turn-around times are simply too long. You can't be waiting 50 minutes every time you make a small change. Therein lies madness. But because Docker uses a copy-on-write filesystem under the hood, the filesystem is really a kind of persistent data structure. When one particular part of the script failed I was able to go back to *exactly* the state it was in before it failed. The boon to my productivity and my sanity was immense. Sean [1] https://www.haskell.org/haskellwiki/Android [2] https://github.com/sseefried/docker-build-ghc-android

1 1

Munich Haskell Meeting, 2014-11-26 @ 19:30
by Christian Neukirchen 24 Nov '14

24 Nov '14

Dear all, next week, on Wednesday, 26th of November, our monthly Munich Haskell Meeting will take place again at Cafe Puck at 19:30 CET. For details see here: http://www.haskell-munich.de/dates Since Heinrich is not around much in Munich anymore, I have taken over organizing our Haskell meeting. The website will be updated soon. For the meantime, if you plan to join, please add yourself to this dudle so we can reserve enough seats: https://dudle.inf.tu-dresden.de/haskell-munich-nov-2014/ Everybody is welcome! Looking forward to see you, -- Christian Neukirchen <chneukirchen(a)gmail.com> http://chneukirchen.org

1 0

Improving *> and >> for Data.Sequence
by David Feuer 23 Nov '14

23 Nov '14

I'd like to define (*>) and (>>) for Data.Seq.Seq in a "clever" way, like replicate, but I'm a bit stuck. It kind of looks like this is the purpose behind the applicativeTree function, which bills itself as a generalization of replicateA, but something seems to have gotten stuck and the only time I see applicativeTree actually used is to define replicateA. With all the fancy nesting, I'm a bit lost as to how to go about this, and having only one example doesn't really help. Can someone help give me a clue? Thanks, David

2 16

GC times with +RTS -N2 -s
by Erik Rantapaa 23 Nov '14

23 Nov '14

When I run the sudoko2 program from the parconc-examples package, I get output like: ... SPARKS: 2 (1 converted, 0 overflowed, 0 dud, 0 GC'd, 1 fizzled) INIT time 0.00s ( 0.02s elapsed) MUT time 1.75s ( 1.44s elapsed) GC time 0.73s ( 0.14s elapsed) EXIT time 0.00s ( 0.00s elapsed) Total time 2.48s ( 1.60s elapsed) ... My full command line is: ./sudoku2 sudoku17.1000.txt +RTS -N2 -s I understand how to interpret the Total time - 2.48 seconds is the total time spent by both cores and 1.6 seconds is the running time of the program as a whole. But how should interpret the GC times? What do 0.73s and 0.14s actually represent? This is with GHC 7.8.3, OSX 10.8.5, 64-bit Haskell Platform.

1 0

Monads: external questions
by Jeffrey Brown 23 Nov '14

23 Nov '14

As opposed to the internal logic of monads, how they work, I hope to start a discussion about their external logic: how and why to use monads. design ------ How do monads change the way one * thinks about a problem? * structures data? * refactors? * tests? Should I always be giving the monads a lot of cognitive bandwidth, because they reorder the way everything should be, or is it an investment with a high initial cognitive cost but requiring little maintenance thereafter? what is their common framework? ------------------------------- Monads let data reach farther than it otherwise would. Subjectively, they feel like a controlled way of violating encapsulation. Are there other, deeper or more specific, commonalities that explain why monads are a good way to implement exceptions, output, state, and perhaps other services? varieties --------- In practice, are all monads either for exceptions, state or output? If not, what other goals are monads well suited to? How should multiple contexts coexist? What's a simple Haskell program that uses more than one kind of monad in the same scope? Should, say, output and state be nestled in parallel into a three-attribute monad (the third variable being the type variable), or would it instead be better to create separate output and state monads? If each context should have a separate monad, is it a good idea to nest them, or somehow otherwise let them overlap, within a particular function or scope? temptations to err ------------------ I have a vague sense that a common pitfall is to use monads to do things in a procedural way that eliminates the benefits of a functional language, but I'm not sure I would recognize it if I were making that error. I feel like I could learn from a comparison of code that works and uses monads badly to code that does better.

5 6

Re: [Haskell-cafe] Generating valid html
by Tobias Dammers 21 Nov '14

21 Nov '14

On Wed, Nov 19, 2014 at 03:45:37PM +0100, Wojtek Narczyński wrote: > On 19.11.2014 13:17, Tobias Dammers wrote: > >That's pretty much what my tamper library > >(http://hackage.haskell.org/package/tamper) is supposed to be. It's > >pretty rough around the edges still, and nowhere near as complete as > >Blaze itself, but it works well enough for basic HTML templating, and > >unlike Blaze, it is implemented as a monad transformer, which means you > >can integrate it with whatever monad stack you like. I couldn't find an > >existing solution to this myself, which is why I rolled my own. > Tobias, > > > I'm puzzled how your library can ensure that, for example <head> has only > one <title>, how / where are such rules enforced? Do you have a sample or > testcase? Ah, sorry for getting your hopes up. No, it doesn't ensure valid parent/child combinations or anything, it just does what basic Blaze-HTML does, but generalized to a monad transformer. > > > On the practical side, wouldn't you prefer to write, for example? > > {-#LANGUAGE Rank2Types #-} > > type ElementType = (Monad m, Ord t, IsString t) => TamperT t m () -> TamperT > t m () > > a, abbr, address, area, article, aside, audio :: ElementType > a = tag "a" > abbr = tag "abbr" > address = tag "address" > area = tagS "area" > article = tag "article" > aside = tag "aside" > audio = tag "audio" Good point; patches welcome in case I forget doing it myself ;)

1 0

Generating valid html
by Wojtek Narczyński 21 Nov '14

21 Nov '14

Hello list, I have a question. How to create a DSL for HTML generation, that would (statically) allow only valid HTML to be generated? Parent - child relations, valid attributes only. Or does such a DSL already exist, perhaps? -- Kind regards, Wojtek Narczynski

6 11

NYC Haskell Developer Position
by Christopher Lewis 21 Nov '14

21 Nov '14

Karamaan Group, a principal investment firm based in Manhattan, is looking for an outstanding software developer to develop tools for financial analysis and knowledge management. We are a growth oriented firm that values people who take a craftsman's pride in their work. Our ideal candidate is an experienced software developer with strong analytical, organizational, and communication skills. A candidate who demonstrates an intense focus on quality, but has the ability to recognize and make the tradeoffs that are a necessary part of day-to-day software development. Candidates should have at least a degree in a quantitative field and a keen interest in building robust and elegant computer programs. This is a high-impact, high-visibility position where successful candidates will be entrusted with a lot of responsibility for products that have a direct effect on the P&L of the firm and influences our workflow. The ideal candidate will have experience with Haskell, relational database technologies, and Java. All new development is performed in Haskell, but our legacy Java code base occasionally requires maintenance. Unlike most finance companies, our atmosphere is informal and intellectual. We don't require previous experience in finance or business, but knowledge in those areas is a plus. Karamaan Group is situated in Manhattan, we are an investment adviser to hedge funds and family offices. We value innovative thinking, encourage an entrepreneurial atmosphere, and enjoy spirited conversations. We also offer lunch, and every once in a while we go bowling. Please send your CV and cover letter to recruitment at karamaan dot com.

1 0

Space leak in WAI 3.0 application
by Thomas Koster 20 Nov '14

20 Nov '14

Hello list, I am quite new to Haskell and I love the language and community, but I am frustrated by a space leak in a WAI 3.0 Application that for now just echoes the request entity back in the response. Specifically, I am having trouble understanding *why* I have the space leak. I intend to pass some or all of the request entity on to another web service whose response will influence the HTTP status code and headers of my service's response. At the moment, I am preparing the request entity as a lazy bytestring using lazy I/O just like Data.ByteString.Lazy.hGetContents does, at least until I can get around to learning pipes or conduit. When I use this technique to echo the request entity back in the response, it looks like two copies of the entire request entity are being accumulated in memory, presumably the original from the request and a copy for the response. The heap profile says it is all in "PINNED", which I am assuming are the bytestring buffers. However, the efficacy of this technique turns out to be irrelevant as I have been able to distill the problem down to a much simpler example: a WAI Application that responds with 100 MB of zeros read from /dev/zero using Data.ByteString.Lazy.hGetContents. I have prepared two variations that differ only in the composition of the operations. Both applications create identical responses, but version A accumulates the entire 100 MB entity in memory (heap profile shows a huge "PINNED" cost just over 100 MB in size) whereas version B streams the entity in constant space. Source code and a cabal file follow. I am using GHC 7.8.3. It is not necessary to make heap profiles - the symptoms are evident with "+RTS -s". ==== BEGIN Zeros.hs ==== {-# LANGUAGE OverloadedStrings #-} import Blaze.ByteString.Builder import Control.Concurrent import qualified Data.ByteString.Lazy as LBS import Data.Int import Network.HTTP.Types import Network.Wai import Network.Wai.Handler.Warp import System.IO -- | This version binds the large LBS of zeros /outside/ of the -- 'responseStream' body lambda. -- -- This version has the space leak. -- -- @ -- curl -v -o \/dev\/null localhost:3000\/zeros\/a -- @ zerosAppA :: Application zerosAppA _req respond = withZeros 100000000 $ \ largeLBS -> respond $ responseStream status200 [] $ \ write _flush -> write $ fromLazyByteString largeLBS -- | This version binds the large LBS of zeros /inside/ of the -- 'responseStream' body lambda. -- -- This version streams the response entity in constant space. -- -- @ -- curl -v -o \/dev\/null localhost:3000\/zeros\/b -- @ zerosAppB :: Application zerosAppB _req respond = respond $ responseStream status200 [] $ \ write _flush -> withZeros 100000000 $ \ largeLBS -> write $ fromLazyByteString largeLBS -- | Do something with /n/ bytes read lazily from @\/dev\/zero@. -- -- This part is common to both 'zerosAppA' and 'zerosAppB'. withZeros :: Int64 -> (LBS.ByteString -> IO a) -> IO a withZeros n f = withBinaryFile "/dev/zero" ReadMode $ \ h -> do zeros <- LBS.hGetContents h let largeLBS = LBS.take n zeros f largeLBS main :: IO () main = do _ <- forkIO $ run 3000 app putStrLn "Using port 3000. Press ENTER to exit..." _ <- getLine putStrLn "Exit." app :: Application app req respond = case pathInfo req of ["zeros", "a"] -> zerosAppA req respond ["zeros", "b"] -> zerosAppB req respond _ -> respond $ responseLBS status404 [] "Not found." ==== END Zeros.hs ==== ==== BEGIN zeros.cabal ==== name: zeros version: 0.1.0.0 build-type: Simple cabal-version: >=1.10 executable zeros main-is: Zeros.hs build-depends: base >=4.7 && <4.8, blaze-builder ==0.3.3.4, bytestring ==0.10.4.0, http-types ==0.8.5, wai ==3.0.2, warp ==3.0.2.3 default-language: Haskell2010 ghc-options: -Wall -rtsopts ==== END zeros.cabal ==== Why does version A not process the LBS in constant space? What in version A is preventing the GC from collecting the LBS chunks after they have been fed to Warp? What is it about version B that permits the LBS chunks to be collected? Although I believe the issue is not actually specific to WAI or Warp, I am unable to reproduce the space leak without them. But because I am new to Haskell, I suspect I have missed something obvious about lambda bindings, laziness (or strictness) of IO, and GC. Thanks. -- Thomas Koster

3 5

Forcing recomputation
by Todd Wilson 20 Nov '14

20 Nov '14

I'm curious: is there a reliable way to force Haskell to recompute a function application instead of sharing the result of the forced thunk? For example, suppose I have a function f and two arguments e1 and e2, and I want to compare the time it takes to compute f e1 vs. f e2, but ghci reports that each takes 0.00 sec (i.e., less than 0.005 sec) to compute. The traditional approach would be to compute each value a large number of times so that the total time becomes significant, except that all of the ways I can think of to do that in Haskell don't actually recompute the function application, but rather compute it once and share, or otherwise optimize away the redundant computation. Todd Wilson, PhD Department of Computer Science California State University, Fresno

4 5