- Glasgow-haskell-users

RE: Profiling and Data.HashTable
by Simon Marlow 17 Oct '05

17 Oct '05

On 14 October 2005 20:31, Jan-Willem Maessen wrote: > That "5K" number made me immediately suspicious, so I took a look at > the source code to Data.HashTable. Sure enough, it's allocating a > number of large IOArrays, which are filled with pointers. The > practical upshot is that, for a hash table with (say) 24 entries, the > GC must scan an additional 1000 pointers and discover that each one > is []. > > I've seen other implementations of this two-level technique which use > a smallish sEGMENT_SIZE in order to avoid excessive GC overhead for > less-than-gigantic hash tables. This might be worth doing in the > Data.HashTable implementation. > > [Curious: what (if anything) is being used to test Data.HashTable? > I'd be willing to undertake very small amounts of fiddling if I could > be sure I wasn't slowing down something which mattered.] The few benchmarks I've tried seem to indicate that Data.HashTable performance isn't overwhelming, and can often be bettered by Data.Map. I've never investigated too deeply. GC overhead seems a very plausible cause, though. Please by all means tweak the implementation and try some benchmarks, I'll incorporate any performance fixes that arise. Cheers, Simon

1 0

Dynamic Map
by Lajos Nagy 17 Oct '05

17 Oct '05

Hi, Would it be possible to implement a Map in Haskell that, when asked for a key it doesn't have, would return a 'fresh' (meaning: not in the Map already) value, and afterwards it would consistently return the same value for the given key. In other words, it would behave like a dynamic map which silently extends itself when someone asks for a non-existent key. If the user of this map only depends on the returned value being different from the rest, then it's easy to see that this dynamic map cannot break equational reasoning so one would expect to be able to assign it a non-monadic type: lookup :: k -> DMap a -> a insert :: k -> a -> DMap a -> DMap a Of course, DMap cannot support certain operations because that would break equational reasoning. For example: size :: DMap a -> Int would depend on the order of lookups. However, if we restrict the operations to insert and lookup then ER is restored. (And those two operations is all I need.) I tried several ways of implementing it but those monadic types just kept cropping up in the map interface. I'd appreciate any ideas or pointers. Regards, Lajos Nagy

3 2

RE: GHC-6.4.1 on FreeBSD-amd64 still not ported
by Simon Marlow 17 Oct '05

17 Oct '05

On 12 October 2005 17:34, Wilhelm B. Kloke wrote: > Simon Marlow <simonmar(a)microsoft.com> schrieb: > >> Just doing 'make -k' in ghc/rts should leave all the .hc files >> behind, because you already have the -keep-hc-files flag in your >> command line, so GHC won't delete the intermediate .hc files even >> when the subsequent compilation stage fails. > > This doesn't work. The commands don't leve a .hc file behind, even > with make -k. My apologies: the failure was happening before generation of the .hc file. The error is generated by this in Cmm.h: #if SIZEOF_mp_limb_t != SIZEOF_VOID_P #error mp_limb_t != StgWord: assumptions in PrimOps.cmm are now false #endif SIZEOF_mp_limb_t comes from DerivedConstants.h, and SIZEOF_VOID_P comes from ghcautoconf.h (both in ghc/includes). Both of these files should be from the target system for a cross-compile; I strongly suspect that one of them has been overwritten by the host version in your tree. Sorry that this process is somewhat flaky; we've never got around to making the build system really do cross-compilation properly, and the fact that it usually only needs to be done once for any given platform means there isn't a lot of motivation to do the work. Cheers, Simon

6 6

GHC build failure
by Krasimir Angelov 16 Oct '05

16 Oct '05

When I am trying to build GHC from CVS it fails with: deSugar/DsMeta.hs:288:13: Constructor `ConDecl' should have 6 arguments, but has been given 4 When checking the pattern: ConDecl con [] (L _ []) details When checking the pattern: L loc (ConDecl con [] (L _ []) details) In the definition of `repC': repC (L loc (ConDecl con [] (L _ []) details)) = do con1 <- lookupLOcc con repConstr con1 details HsDecls.ConDecl has been changed in revision 1.112 but DsMeta isn't updated.

1 0

Network Exception
by Arias 14 Oct '05

14 Oct '05

Hi, I've a problem, I'm using this code on GHC compiler version 6.4.1: --- BEGIN --- module Main where import System.IO import Network main = withSocketsDo $ do handle <- connectTo "localhost" ( PortNumber 8080 ) hSetBuffering handle LineBuffering hClose handle --- END --- The problems is that it throws an exception at connectTo, the exception is: getServiceEntry: does not exist (no such service entry) The server and the port is correct, I have apache listening on port 8080 and if I put http://localhost:8080/ on browser it loads a web page. Do somebody knows what it happens and how to fix it? thanks.

3 2

RE: Profiling and Data.HashTable
by Simon Marlow 14 Oct '05

14 Oct '05

On 14 October 2005 15:17, Ketil Malde wrote: > I have a program that uses hash tables to store word counts. It can > use few, large hash tables, or many small ones. The problem is that > it uses an inordinate amount of time in the latter case, and > profiling/-sstderr shows it is GC that is causing it (accounting for > up to 99% of the time(!)) > > Is there any reason to expect this behavior? > > Heap profiling shows that each hash table seems to incur a memory > overhead of approx 5K, but apart from that, I'm not able to find any > leaks or unexpected space consumption. I'm not certain that this is your problem, but hash tables are by definition mutable objects, and mutable objects don't work too well with generational garbage collection, GHC's in particular. Basically every GC, even the minor ones, will have to scan all the mutable objects. GHC doesn't make any attempt to record which mutable objects have actually been mutated since the last GC. This is a compromise: it makes writing to a mutable object faster, at the expense of GC time later if you have lots of mutable objects. We haven't tried the alternative approach, but it's almost certainly a case of swings and roundabouts: programs with lots of infrequently-mutated data would benefit, programs with a few often-mutated objects would lose. I guess we currently optimise for the "few often-mutated objects" case. Cheers, Simon

2 1

Re: Network Exception
by Arias 14 Oct '05

14 Oct '05

Thanks for the reply. I don't know if the problem is the same, but when I try to run this code: main = withSocketsDo $ do host <- getHostName putStrLn host the getHostName throws this exception: getHostName: failed (Successful WSAStartup not yet performed (WSANOTINITIALISED)) :( I'm using Windows XP Professional Sigbjorn Finne escribió: > Hi, > > that error message is a bit confusing, to say the least: > getProtocolByName identifies itself as getServiceEntry > when failing. In this case, I'm quite sure, your snippet > fails because (getProtocolByName "tcp") isn't successful. > I've no idea why your /etc/protocols doesn't contain an entry > for 'tcp'. > > The error message in said function has now been fixed; thanks > for the report. > > --sigbjorn > > ----- Original Message ----- From: "Arias" <arias(a)elleondeoro.com> > To: <glasgow-haskell-users(a)haskell.org> > Sent: Friday, October 14, 2005 06:47 > Subject: Network Exception > > >> Hi, >> >> I've a problem, I'm using this code on GHC compiler version 6.4.1: >> >> --- BEGIN --- >> module Main where >> import System.IO >> import Network >> >> main = withSocketsDo $ do >> handle <- connectTo "localhost" ( PortNumber 8080 ) >> hSetBuffering handle LineBuffering >> hClose handle >> --- END --- >> >> The problems is that it throws an exception at connectTo, the >> exception is: >> >> getServiceEntry: does not exist (no such service entry) >> >> The server and the port is correct, I have apache listening on port >> 8080 and if I put http://localhost:8080/ on browser it loads a web page. >> >> Do somebody knows what it happens and how to fix it? thanks. >> _______________________________________________ >> Glasgow-haskell-users mailing list >> Glasgow-haskell-users(a)haskell.org >> http://www.haskell.org/mailman/listinfo/glasgow-haskell-users > > > >

1 0

Help wanted! Win32 and HGL for GHC on Windows
by Simon Peyton-Jones 14 Oct '05

14 Oct '05

Dear GHC users This is an appeal for help with the libraries/Win32 package [on Windows, obviously] libraries/HGL package [on Windows] Here's the situation: * Win32 provides access to the native Windows API, which is obviously very useful for people writing Haskell on Windows. HGL, the "Haskell Graphics Library", is the graphics package used by Paul Hudak's book "The Haskell School of Expression". * The libraries/Win32 package is supposed to supercede the old hslibs win32 package. Ross Paterson ported the latter to become the former, but the libraries/Win32 package had not received very much testing, and Ross can't do any more himself. We aren't confident that it's reliable. (Sadly, the two packages have the same name (modulo case) which has given rise to much confusion. But that's another matter.) * The libraries/HGL package works fine on Unix, where it sits on top of the libraries/X11 package. But it does not work properly on Windows, where it sits on top of libraries/Win32. It is not clear whether the problem is with the HGL package or the Win32 package. So the current situation is that, for GHC on Windows, the new hierarchical libraries world lacks (a) a reliable Win32 package and (b) a HGL package that works at all. This is sad, especially as we are trying to nuke hsllibs altogether. Question: would anyone be willing to fix the libraries/Win32 package; and (better still) look after it thereafter? We'd be delighted to give commit rights to anyone who was willing to do this. [So far as we know the two libraries/ packages work ok in Hugs, so there probably isn't that much to do.] Many thanks Simon

3 3

RE: Default name of target executable
by Simon Marlow 14 Oct '05

14 Oct '05

On 13 October 2005 18:57, Tomasz Zielonka wrote: > On 10/11/05, Tomasz Zielonka <tomasz.zielonka(a)gmail.com> wrote: >> On 10/11/05, Simon Marlow <simonmar(a)microsoft.com> wrote: >>> Ok, let's close this bikeshed. Someone want to send us a patch? >> >> I will try to do this > > On which branch of GHC should I be working? > There are some differences between HEAD and STABLE > in the relevant areas. We always work on HEAD first, and port changes to STABLE if appropriate. Cheers, Simon

1 0

RE: Default name of target executable
by Simon Marlow 13 Oct '05

13 Oct '05

On 11 October 2005 11:04, Duncan Coutts wrote: > I'd tend to agree. > > It'd mean one less magic incantation to tell students when teaching > practicals (and one less thing for them to get wrong) if they can do: > > ghc --make Main.hs > > rather than > > ghc --make Main.hs -o Main > > (that's one advantage of ghci for teaching students, that 'ghci > Blah.hs' "Just Works"tm) Ok, let's close this bikeshed. Someone want to send us a patch? Cheers, Simon

2 2