- Glasgow-haskell-users

Re: FW: GHC API: Problems with implementing ":reload" for "ghc --make"
by Ian Lynagh 30 Nov '06

30 Nov '06

Hi Brian, Sorry for the delayed response. > My goal is to write a program GhcRemake that works like "ghc --make." > However, instead of terminating after compilation is done, I want the > program to stay open and wait for me to hit <ENTER>. When I hit > <ENTER>, GhcRemake rebuilds the project, just as if I had called "ghc > --make" again with the same arguments. The benefit of such a program > is that GhcRemake should be able to cache a lot of data in memory > between invocations and hopefully be able to do the subsequent > re-makes much faster. Nice idea! > My program seems to work fine when run on itself and when run on > Cabal. But, these two packages seem to be too small to notice any > reduction of building time. So, I decided to test my program by > building the GHC API with it. Unfortunately, it seems like every build > after the first one in the session does the dependency analysis badly, > and things get recompiled unnecessarily. It looks like the problem is caused by recursive module imports. I've added a bug (#1027) with a smaller example of it. Thanks Ian

3 2

Common subexpression elemination (CSE)
by Christian Maeder 29 Nov '06

29 Nov '06

the following code does not run as fast as expected: modexp a e n = if e <= 0 then 1 else if even e then mod (modexp a (div e 2) n * modexp a (div e 2) n) n else mod (a * modexp a (e - 1) n) n it gets only fast if written as: modexp2 a e n = if e <= 0 then 1 else if even e then let d = modexp2 a (div e 2) n in mod (d * d) n else mod (a * modexp2 a (e - 1) n) n I wonder, why the common subexpression "modexp a (div e 2) n" is not eliminated in the first case. Does CSE work at all? For testing the exponent "e" should have at least 10 digits. Cheers Christian P.S. Other alternatives are slower, too modexp1 a e n = if e <= 0 then 1 else mod (a * modexp1 a (e - 1) n) n modexp3 a e n = mod (a ^ e) n

6 8

Re: Common subexpression elemination (CSE)
by Dinko Tenev 29 Nov '06

29 Nov '06

On 11/28/06, Mark T.B. Carroll <mark(a)ixod.org> wrote: > > "Dinko Tenev" <dinko.tenev(a)gmail.com> writes: > (snip) > > How exactly can CSE cause space leaks, and what does this have to do > > with strictness? > > I puzzled over Lennart's comment too, and the theory I formed was that > you might have the same expression used in different parts of a > function, where the compiler hasn't realised that the remaining > reference to the expression is in a part of the function that isn't > going to be executed this time, so it is holding on to the expression's > value even though it is never going to be used. (For instance, with a > recursive function, all the previous calls may be holding on to their > version of the value until the call stack has come back out past them > again.) > > -- Mark > > I see this as a special case of Lennart's point -- you're talking about references that are practically dead (just not burried yet :) The compiler should probably be able to kill these early by a CPS transform, shouldn't it? -- Cheers, Dinko

2 1

Using GHC API
by C.M.Brown 29 Nov '06

29 Nov '06

Hi, I am currently in the process of porting some of the Haskell Refactorer (HaRe) over to ghc 6.6. Part of HaRe requires the API and until now I've been content with using th 6.5 API. However, since I've started the switch I've noticed some strange problems and the latest is I am getting the following error when trying to find the type of an expression: <interactive>:1:0: Can't find interface-file declaration for Main.main Probable cause: bug in .hi-boot file, or inconsistent .hi file Use -ddump-if-trace to get an idea of which file caused the error I have attached the code and would appreciate if someone could point me in the right direction. Basically I just want to find the type of a particular expression within a file. This used to work but now seems to fail. The code is compiled with: ghc --make -package ghc-6.6 -o main hasktags.hs and run with ./main Main.hs The file main can have anything in it really - I'm just using it to find a type of some function called main. Thanks for you help! Chris Brown.

3 3

bignums, gmp, bytestring, .. ?
by Claus Reinke 28 Nov '06

28 Nov '06

it seems that haskell versions of bignums is pretty much gone from more recent discussions of gmp replacements. now, I assume that there are lots of optimizations that keep gmp popular that one wouldn't want to have to reproduce, so that a haskell variant might not be competitive even if one had an efficient representation, but - do all those who want to distribute binaries, but not dynamically linked, need bignums? - it would be nice to know just how far off a good haskell version would be performance-wise.. - what would be a killer for numerical programming, might still be quite acceptable for a substantial part of haskell uses? of course, the real gmp replacement project might be going so well that a haskell version would be obsolete rather sooner than later, and i certainly don't want to interfere with that effort. all that being said, it occurred to me that the representations and fusions described in the nice "rewriting haskell strings" paper would be a good foundation for a haskell bignum project, wouldn't they? http://www.cse.unsw.edu.au/~dons/fps.html http://hackage.haskell.org/trac/ghc/wiki/ReplacingGMPNotes has anyone been looking into this option? just another thought, claus ps. while I'm at it: claiming that "array fusion .. has received comparatively little attention" sounds a bit dangerous to me, and the references are all too limited - even if you meant "in the Haskell world" (and PADL is no Haskell event). you might want to try searching with some other search terms, keeping in mind that most of the work will have been done for not so declarative languages: it isn't my area, but "loop fusion" and "with-loop folding" come to mind, and I'm sure that all those array language and numerical programming folks would be rather, well, disappointed?, to see their efforts ignored like that.

7 12

Tuning -funfolding-use-threshhold
by dons＠cse.unsw.edu.au 27 Nov '06

27 Nov '06

Does this thread on haskell-cafe, about getting better bit shifting results, indicate we should be tweaking the default unfolding-use threshold? Do the recent bit shifting/inlining patches improve things? http://thread.gmane.org/gmane.comp.lang.haskell.cafe/16849/focus=16849 I'm thinking that users shouldn't need to resort to manual -funfolding-use-threshhold settings? -- Don

2 1

Passing arrays between Haskell and C without overhead
by Brian Hulley 26 Nov '06

26 Nov '06

Hi, For a 3d graphics library, I'd like to have a 4x4 matrix of floats which I can pass from Haskell to C without having to copy all 16 floats each time, something like: data Mat44f = Mat44f Float# Float# ... Float# foreign import ccall c_setWorldTransform :: Ptr Mat44f -> IO () setWorldTransform :: Mat44f -> IO () setWorldTransform mat = withSafePtr mat c_setWorldTransform where withSafePtr :: a -> (Ptr a -> IO b) -> IO b would supply a pointer directly to the Haskell heap representation of (a) and would ensure that (a) doesn't get garbage collected until the IO action is completed. (Note that this is not the same as Foreign.Marshal.Utils.with, which uses alloca and poke to copy the matrix - I'm trying to avoid this overhead of copying.) However there is no such function afaics so the only solutions at present seem to be to either make Mat44f an instance of Storable and use alloca (which involves copying all 16 floats just so that C can read it even though I'd expect the heap representation to already be identical since there is only one constructor and all fields are unboxed) or else pass the 16 floats individually by: foreign import ccall c_SetWorldTransform' :: Float# -> Float# -> ... -> Float# -> IO () setWorldTransform (Mat44f m11 m12 m13 m14 ... m44) = c_SetWorldTransform' m11 m12 m13 ... m44 In this second case the C function will have to manually put these 16 floats back into an array so that it will have a pointer to a temporary array of 16 floats to use internally (eg to pass to the DirectX api which will then make its own copy of it), so in both cases there is a wasteful deconstruction then reconstruction of the original array of 16 floats stored on the Haskell heap. Similar problems arise when trying to pass an array from C back to Haskell, but in this case it's clear some copying will be needed so that one ends up with a value stored on the Haskell heap - the "double copy" problem only arises in the direction from Haskell to C. I've also thought about just implementing all the math functions in C and using mallocForeignPtrArray so that the data type would be: newtype Mat44f = Mat44f (ForeignPtr CFloat) While this would make passing between C and Haskell very fast, it seems a pity to have to sacrifice the ability to code the math ops directly in Haskell in a functional style with all the attendant optimizations one could expect from GHC (using ForeignPtr requires unsafePerformIO, withForeignPtr, and lots and lots of peek/pokeElemOff's in a monadic style - rather gross compared to pattern matching etc) So my questions are: 1) Is there any reason why there can't be a function (withSafePtr) to avoid all this marshalling overhead when we know (or specify via a pragma) that Haskell stores the data already in a form compatible with C, or is there some other way to represent a 4x4 matrix such that it can be manipulated in Haskell and passed without overhead to C? 2) Does withSafePtr already exist somewhere? 3) Would it be a good idea to make Storable one of the classes which can be automatically derived? Thanks, Brian. -- http://www.metamilk.com

2 2

Hacking on GHC interactively with GHCi
by Clemens Fruhwirth 25 Nov '06

25 Nov '06

Hello everyone, I would like to hack on GHC interactively. My aim is to load ghc into ghci and start hacking a source file in one emacs buffer, while the other hosts an inferior-haskell session connected to GHCi. I really like this kind of development style and found it to be way more productive than "edit/compile/test/restart from edit". To context of the following is the compiler/ subtree, as I'm primarily interested in that. I tried to: * load all GHC compiler source files into ghci, fails because of occassional unboxed types (bytecode generated can't deal with them) * load all .o files (with unboxed types) and load a modified source in interpreted mode. This doesn't work well because for that approach you need to mix the source with the object in the same directory (copy compiler/**/*.[l]hs to stage1 or stage2). This, fails because there seem to be some sources missing, I have not investiged where the .o-boot files come from like compiler/stage2/basicTypes/OccName.o-boot. With source files, GHCi just ignores the existing .o file (resumable because it can't check whether the .o file is more recent than the source file) * load "ghci -package ghc" and load main/Main.hs. This work pretty well, but only in comparsion to the other approaches -- that means loading does not fail. Invoking ":main -c foobar.hs" freezes the ghci session. With this approach I can not change any module withing the packaged ghc. Is that true? But that's what I want to do (in particular to the parser part). So, before I start to work towards a particular end of these 3 approaches, is there anyone how develops ghc in an interactive way using ghci with partial :reload-ing of modules? The GHC user guide has a good illustration of what I'm looking for http://www.haskell.org/ghc/docs/latest/html/users_guide/ghci-compiled.html Any hint is appreciated. -- Fruhwirth Clemens - http://clemens.endorphin.org for robots: sp4mtrap(a)endorphin.org

3 3

Gtk2Hs and GHC 6.6
by Michael T. Richter 24 Nov '06

24 Nov '06

OK, my last post on this was put into moderator limbo for being too large and then never released nor rejected. Here's try number two. I won't attach the full build output because that's apparently what put it into moderator limbo in the first place. I'm trying to build Gtk2Hs under GHC 6.6. I've built it successfully under 6.4.1 and 6.4.2 -- no muss, no fuss, no bother. Something big seems to have changed in 6.6, however, that breaks the Gtk2Hs build in a bad way. Here are some representative errors: First, we see that it did find the right GHC version: checking for ghc... /usr/local/bin/ghc checking version of GHC... 6.6 checking for ghc-pkg... /usr/local/bin/ghc-pkg But now we get some strange package problems. Each of these is a representative. The full build has dozens of these -- I think I've got each flavour though: tools/c2hs/base/general/Binary.hs:59:7: Could not find module `Data.FiniteMap': Use -v to see a list of the files searched for. make[1]: *** Deleting file `tools/c2hs/c2hsLocal.deps' tools/c2hs/base/admin/Common.hs:54:0: Failed to load interface for `Config': it is a member of package ghc-6.6, which is hidden make[2]: *** [tools/c2hs/base/admin/Common.o] Error 1 make[2]: Leaving directory `/home/michael/Development/Gtk2Hs/gtk2hs-0.9.10' make[1]: *** [glib/System/Glib/Types.hs] Error 2 make[1]: Leaving directory `/home/michael/Development/Gtk2Hs/gtk2hs-0.9.10' <no location info>: can't find file: glib/System/Glib/Types.hs make: *** Deleting file `glib/libHSglib_a.deps' glib/System/Glib.hs:12:0: Failed to load interface for `System.Glib.UTFString': Use -v to see a list of the files searched for. make[1]: *** [glib/System/Glib.o] Error 1 make[1]: Leaving directory `/home/michael/Development/Gtk2Hs/gtk2hs-0.9.10' make: *** [all] Error 2 So I'm left with the question: what am I doing wrong (if anything)? Or what do I need to do to get Gtk2Hs to compile under 6.6? -- Michael T. Richter Email: ttmrichter(a)gmail.com, mtr1966(a)hotpop.com MSN: ttmrichter(a)hotmail.com, mtr1966(a)hotmail.com; YIM: michael_richter_1966; AIM: YanJiahua1966; ICQ: 241960658; Jabber: mtr1966(a)jabber.cn "I would not flinch from sacrificing a million lives for India's liberty!" --Mahatma Gandhi

2 1

evaluate to lazy?
by Andreas Marth 22 Nov '06

22 Nov '06

Hi! With the following code: module Guess where import Prelude hiding (catch) import Control.Exception (evaluate, catch) guess :: [String] -> IO String guess sl = do res <- catch (evaluate (concat $ sl ++ [error "some error message", "blah blah blah"])) (\e -> return ("error#"++show e)) return res -- only for demonstration coded like this normally a call to sysalloc comes here what will a) guess [] b) guess [""] c) guess [" "] return? The idea was to return the string, catch every error if any occour, convert it into a string and prefix it with "error#" and return this string then. The reason to do this is to create a stable DLL with the error handling in non haskell land. At the moment every exception raised crashes the whole system, which is unacceptable. Unfortunately the c) case still raises an exception. I think at least the library description needs a hint that 'evaluate (" "++error "some error message")' does not raise the error (which I find strange!) and hence catch won't catch it. Is this behaivior really desired or should we consider it a bug? Thanks, Andreas PS: I circumvent this issue by using 'rnf' and 'using' from Control.Parallel.Strategies

3 4