- Haskell-Cafe - Haskell.org

NaN as Integer value
by Алексей Егоров 13 Apr '13

13 Apr '13

Hello haskellers, is there a reason why Integer doesn't have 'NaN' as value? I think it would be very convenient to be able to handle (1 `div` 0) as regular NaN value and not as exception. Thanks.

4 3

ANNOUNCE: monoid-subclasses-0.1.1, incremental-parser-0.2.2, and SCC-0.8
by Mario Blažević 13 Apr '13

13 Apr '13

The new package monoid-subclasses [1] exports a number of classes that sit between monoids and groups: ReductiveMonoid, CancellativeMonoid, GCDMonoid, MonoidNull, and FactorialMonoid among others. The package also comes with class instances for all applicable data types from base, vector, containers, bytestring, and text packages. These classes were initially a part of the incremental-parser package [2] but they've moved, so incremental-parser-0.2.2 now depends on monoid-subclasses which has added more class methods and instances. The parsers constructed using the incremental-parser library generally work on any ReductiveMonoid, FactorialMonoid or TextualMonoid. You can thus construct a generic parser that operates on String, ByteString, or Text input, or even on Product Integer or a Map if you are so inclined. The version 0.8 of Streaming Component Combinators (a.k.a. SCC) has also been re-designed from the ground up. In the earlier versions, as in the pipes and conduit packages, the types of streams and stream-processing components were parametrized by the type of the individual items flowing through. All sources and sinks are now parametrized instead by the type of the monoid they produce or consume. As a consequence, SCC coroutines can now communicate in chunks of ByteString or Text instead of just lists, and these chunks can be of arbitrary size. The boundaries of producer's chunks are invisible to the consumer (excepting some low-level functions), which must specify the amount of data it wants to receive using a ReadRequest. This change at the heart of SCC has had a positive impact on outer layers, and especially on the Splitter component type which has become much simpler. [1] http://hackage.haskell.org/package/monoid-subclasses-0.1.1 [2] http://hackage.haskell.org/package/incremental-parser-0.2.2 [3] http://hackage.haskell.org/package/scc-0.8

2 1

ANN: psqueue-benchmarks - benchmarks of priority queue implementations
by Niklas Hambüchen 13 Apr '13

13 Apr '13

(This is a slightly detailed email. If you are the maintainer of one of the packages benchmarked here, you might want to read it though.) Today I was looking for a Priority Queue that also allows a delete operation (some call this a "Priority Search Queue"). I found http://stackoverflow.com/questions/6976559/comparison-of-priority-queue-imp… and after looking at the 10 queue alternatives, I got to the following conclusions: * Only 3 of them allow to delete entries (are p*s*queues) * Most of them are outdated or at least unmaintained for the last 3-5 years * There was an effort to get a priority queue into containers (see the stackoverflow link), but it was not agreed on * Those efforts were driven by Louis Wasserman, who wrote one of the 3 psqueues (queuelike), but now only maintains a non-ps-queue (pqueue) that seems to be the most popular priority queue implementation PSQueue implementations ----------------------- The three packages that are psqueues are: - PSQueue (http://hackage.haskell.org/package/PSQueue-1.1) * original implementation from paper of Ralf Hinze * last upload 2008 * no test suite (but small commented out QC properties), no benchmarks * no code repository - fingertree-psqueue (http://hackage.haskell.org/package/fingertree-psqueue-0.3) * last upload 2011 * no tests, no benchmarks * no code repository - queuelike (http://hackage.haskell.org/package/queuelike-1.0.9) * last upload 2009 * no tests, no benchmarks * no code repository Benchmarks ---------- Unfortunately, none of them had tests, code in repositories or any indication about their real-world performance, so I made some criterion benchmarks. You can find them here: https://github.com/nh2/psqueue-benchmarks Graphs: http://htmlpreview.github.com/?https://raw.github.com/nh2/psqueue-benchmark… Benchmark results ----------------- * PSQueue throws a stack space overflow if you try to put in 100000 Ints * PSQueue suffers from some significant worst case in terms of queue creation, sometimes creating a queue from random numbers just takes 5 times longer (1st graph). This only happens sometimes (despite Criterion) * queuelike creation is instant - it seems to work around my benchmark somehow * converting a queuelike queue to a list surprisingly takes 10 times longer than with the other packages * in terms of average performance, the three are quite close to each other (apart from the point above). queuelike seems fastest, fingertree-psqueue is second and PSQueue slowest, with a difference of +30% to the next one My questions to the maintainers ------------------------------- @Scott: Do you have an idea why PSQueue stack overflows? @Louis: Why did you switch from queuelike to pqueue? Could you put the code up somewhere manageable (repo)? Is it possible to make pqueue a full pSqueue implementation? Niklas

7 18

Will Haskell Platform 2012.4.0.0 uninstall cleanly for the installation of 2013.2.0.0
by KC 12 Apr '13

12 Apr '13

:) -- -- Regards, KC

2 1

Warsaw Haskell Group
by Przemyslaw Kaminski 12 Apr '13

12 Apr '13

Hello! We are organizing a group of Haskell fans from Warsaw, Poland. So far we have set up a community on G+ (https://plus.google.com/u/0/communities/103183708602453146804) a mailing list is on the way. We would like to meet regularly someplace in the capital (we still decide where), and hack on Haskell together as much as we can :) Should there be more people involved, we could arrange regular lectures. Hope to see you there! Warsaw Haskell Group

1 0

unsafeInterleaveST (and IO) is really unsafe [was: meaning of "referential transparency"]
by oleg＠okmij.org 12 Apr '13

12 Apr '13

One may read this message as proving True === False without resorting to IO. In other words, referential transparency, or the substitution of equals for equals, may fail even in expressions of type Bool. This message is intended as an indirect stab at lazy IO. Unfortunately, Lazy IO and even the raw unsafeInterleaveIO, on which it is based, are sometimes recommended, without warnings that usually accompany unsafePerformIO. http://www.haskell.org/pipermail/haskell-cafe/2013-March/107027.html UnsafePerformIO is known to be unsafe, breaking equational reasoning; unsafeInterleaveIO gets a free pass because any computation with it has to be embedded in the IO context in order to be evaluated -- and we can expect anything from IO. But unsafeInterleaveIO has essentially the same code as unsafeInterleaveST: compare unsafeInterleaveST from GHC/ST.lhs with unsafeDupableInterleaveIO from GHC/IO.hs keeping in mind that IO and ST have the same representation, as described in GHC/IO.hs. And unsafeInterleaveST is really unsafe -- not just mildly or somewhat or vaguely unsafe. In breaks equational reasoning, in pure contexts. Here is the evidence. I hope most people believe that the equality on Booleans is symmetric. I mean the function (==) :: Bool -> Bool -> Bool True == True = True False == False = True _ == _ = False or its Prelude implementation. The equality x == y to y == x holds even if either x or y (or both) are undefined. And yet there exists a context that distinguishes x == y from y ==x. That is, there exists bad_ctx :: ((Bool,Bool) -> Bool) -> Bool such that *R> bad_ctx $ \(x,y) -> x == y True *R> bad_ctx $ \(x,y) -> y == x False The function unsafeInterleaveST ought to bear the same stigma as does unsafePerformIO. After all, both masquerade side-effecting computations as pure. Hopefully even lazy IO will get recommended less, and with more caveats. (Lazy IO may be superficially convenient -- so are the absence of typing discipline and the presence of unrestricted mutation, as many people in Python/Ruby/Scheme etc worlds would like to argue.) The complete code: module R where import Control.Monad.ST.Lazy (runST) import Control.Monad.ST.Lazy.Unsafe (unsafeInterleaveST) import Data.STRef.Lazy bad_ctx :: ((Bool,Bool) -> Bool) -> Bool bad_ctx body = body $ runST (do r <- newSTRef False x <- unsafeInterleaveST (writeSTRef r True >> return True) y <- readSTRef r return (x,y)) t1 = bad_ctx $ \(x,y) -> x == y t2 = bad_ctx $ \(x,y) -> y == x

6 6

Web servers: Running them multiple times in a ghci session
by Niklas Hambüchen 12 Apr '13

12 Apr '13

I'm writing a web server app, which I run in ghci: :main localhost 8000 Unfortunately, after Ctrl-C and :reload, running it again: ** Exception: bind: resource busy (Address already in use) This is pretty annoying, because quitting-and-restarting takes a lot of time since I have many dependencies. How do you deal with this? Can you propose some working code that can be wrapped around my main function to make it work? My first idea is running main in a separate process. If you have a working idea, please also post it as an answer on http://stackoverflow.com/questions/15890912/how-do-i-terminate-a-socket-ser… Thanks

3 2

cabal-install 1.16.0.2 on Mac
by Richard A. O'Keefe 12 Apr '13

12 Apr '13

Machine: an Intel Core 2 Duo desktop Mac. OS: Mac OS X 10.7.4 Xcode: 4.6.1 (including command line tools) Haskell: "Haskell Platform 2012.4.0.0 64bit.pkg" downloaded today (GHC 7.4.2) cabal update advised me to install a new cabal-install. m% cabal install cabal-install Resolving dependencies... Downloading Cabal-1.16.0.3... [ 1 of 65] Compiling Distribution.Compat.Exception ( /var/folders/_n/5bc02hb51d361crtq41g_gc4000g8p/T/Cabal-1.16.0.3-85340/Cabal-1.16.0.3/Distribution/Compat/Exception.hs, /var/folders/_n/5bc02hb51d361crtq41g_gc4000g8p/T/Cabal-1.16.0.3-85340/Cabal-1.16.0.3/dist/setup/Distribution/Compat/Exception.o ) <<< snip >>> [52 of 67] Compiling Distribution.Simple.Build.PathsModule ( Distribution/Simple/Build/PathsModule.hs, dist/build/Distribution/Simple/Build/PathsModule.o ) Distribution/Simple/Build/PathsModule.hs:210:19: Warning: Pattern match(es) are non-exhaustive In a case alternative: Patterns not matched: PPC PPC64 Sparc Arm ... [53 of 67] Compiling Distribution.Simple.GHC ( Distribution/Simple/GHC.hs, dist/build/Distribution/Simple/GHC.o ) <<< snip >>> [65 of 65] Compiling Main ( /var/folders/_n/5bc02hb51d361crtq41g_gc4000g8p/T/Cabal-1.16.0.3-85340/Cabal-1.16.0.3/Setup.hs, /var/folders/_n/5bc02hb51d361crtq41g_gc4000g8p/T/Cabal-1.16.0.3-85340/Cabal-1.16.0.3/dist/setup/Main.o ) Linking /var/folders/_n/5bc02hb51d361crtq41g_gc4000g8p/T/Cabal-1.16.0.3-85340/Cabal-1.16.0.3/dist/setup/setup ... Configuring Cabal-1.16.0.3... Building Cabal-1.16.0.3... Preprocessing library Cabal-1.16.0.3... [ 1 of 67] Compiling Paths_Cabal ( dist/build/autogen/Paths_Cabal.hs, dist/build/Paths_Cabal.o ) <<< snip >>> [56 of 65] Compiling Distribution.Client.SetupWrapper ( Distribution/Client/SetupWrapper.hs, dist/build/cabal/cabal-tmp/Distribution/Client/SetupWrapper.o ) Distribution/Client/SetupWrapper.hs:51:12: Warning: In the use of `ghcVerbosityOptions' (imported from Distribution.Simple.GHC): Deprecated: "Use the GhcOptions record instead" [57 of 65] Compiling Distribution.Client.Upload ( Distribution/Client/Upload.hs, dist/build/cabal/cabal-tmp/Distribution/Client/Upload.o ) <<< snip >>> [65 of 65] Compiling Main ( Main.hs, dist/build/cabal/cabal-tmp/Main.o ) Linking dist/build/cabal/cabal ... Installing executable(s) in /home/cshome/o/ok/.cabal/bin /Developer/usr/bin/strip: object: /home/cshome/o/ok/.cabal/bin/cabal malformed object (unknown load command 15) cabal: Error: some packages failed to install: cabal-install-1.16.0.2 failed during the final install step. The exception was: ExitFailure 1 m% file ~/.cabal/bin/cabal /home/cshome/o/ok/.cabal/bin/cabal: Mach-O 64-bit executable x86_64 The strip(1) page ends with this, which may be relevant: LIMITATIONS Not every layout of a Mach-O file can be stripped by this program. But all layouts produced by the Apple compiler system can be stripped. m% otool -l ~/.cabal/bin/cabal <<< snip >>> Load command 14 cmd LC_FUNCTION_STARTS cmdsize 16 dataoff 12743064 datasize 204136 Load command 15 cmd ?(0x00000029) Unknown load command cmdsize 16 00c58f00 00000000 So something is definitely putting something in there that the Xcode 4.6.1 tools do not like.

3 6

exp implementation
by Christopher Howard 11 Apr '13

11 Apr '13

Hi. For my own learning, I wanted to see how the exp function is implemented in GHC. I have GHC 7.4.1 source code open, but I'm having trouble figuring out which file the actual function definition is in. I see expFloat (F# x) = F# (expFloat# x) in libraries/base/GHC/Float.lhs. -- frigidcode.com

2 3

Which generic zipper?
by Luke Evans 11 Apr '13

11 Apr '13

It's sometimes difficult to figure out a good choice when you're searching for a new feature/library (when you go beyond safety of the HP). I'm looking for a generic zipper to (hopefully) navigate over an AST… so a hetrogeneous structure. It's probably sufficient to make this traversable in a fixed way, but a zipper seemed like a good idea. Google says there are lots of options (various papers and their related implementations etc). I see a generic zipper in an 'syz' package that is based on generics, which looks like a nice choice as it's now easy to add the extension for deriving generics on ADTs. However, that package looks pretty old and unmaintained. Does anyone have any suggestions/advice? Cheers -- Luke

3 5