November 2007 - Glasgow-haskell-users

ANNOUNCE: Important bug fix for regex-pcre ByteStrings.
by ChrisK 20 Nov '07

20 Nov '07

Greetings, There are new version 0.82 and 0.93 of regex-posix. If you use regex-posix with Data.ByteString then you should upgrade to obtain a fix for a crash error. There are new version of regex-pcre available on hackage and the two darcs repositories: http://darcs.haskell.org/packages/regex-pcre/ for "stable" version 0.82 and http://darcs.haskell.org/packages/regex-unstable/regex-pcre/ for "unstable" version 0.93 The new version regex-pcre-0.82 replaces the broken version 0.81 The new version regex-pcre-0.93 replaces the broken version 0.92 I apologize for the error. I took the address of the ByteString to give to the pcre library by using unsafeUseAsCStringLen. I had not appreciated that this can return a null pointer for an empty ByteString. My code caught this null pointer and reported it, however, and so a good error message sent in by a user allowed me to quickly find and hopefully fix it. For version 0.92 the error also applied to Lazy ByteStrings. In case anyone is curious, the fixed code checks the pointer returned by unsafeUseAsCStringLen and replaces a null pointer to a non-null pointer to a private private bytestring "myEmpty" with the length set to 0 by "trim": > {-# INLINE asCStringLen #-} > asCStringLen :: ByteString -> (CStringLen -> IO a) -> IO a > asCStringLen s op = B.unsafeUseAsCStringLen s checked > where checked cs@(ptr,_) | ptr == nullPtr = B.unsafeUseAsCStringLen myEmpty (op . trim) > | otherwise = op cs > myEmpty = B.pack [0] > trim (ptr,_) = (ptr,0) This fix is motivated by desiring to ensure asCStringLen never passes a null pointer to "op" and that no memory copy operations are used. The safe "useAsCString(Len)" do not appear capable of returning a null pointer. Only "unsafeUseAsCString(Len)" on an empty bytestring might return a null pointer. -- Chris

1 0

The order if bindings generated by GHC
by Victor Nazarov 19 Nov '07

19 Nov '07

I use STG-bindings generated by GHC during CoreToSTG phase. What is the order of this bindings is it random or does it correspond to original source code or does it reflect the dependency structure of the program? If I define the following in my program: data Numeral = Zero | Succ Numeral zero = Zero one = Succ zero ... ten = Succ nine the order of zero .. ten definition will remain the same, but the special Zero binding will be generated and added to the end of the bindings list. But Zero is used by binding of one (strangely enough) so order doesn't reflect dependency. What is the order of bindings? If I need the order reflecting dependency, should I sort it myself by SRT tables or there is an easier way? -- vir http://vir.comtv.ru/

3 4

The order if bindings generated by GHC
by Victor Nazarov 19 Nov '07

19 Nov '07

> ==================== STG syntax: ==================== > Foo.zero = NO_CCS Foo.Zero! []; > SRT(Foo.zero): [] > Foo.one = NO_CCS Foo.Succ! [Foo.Zero]; > SRT(Foo.one): [] > Foo.ten = NO_CCS Foo.Succ! [Foo.one]; > SRT(Foo.ten): [] > Foo.Zero = NO_CCS Foo.Zero! []; > SRT(Foo.Zero): [] > Foo.Succ = \r [eta_s68] Foo.Succ [eta_s68]; > SRT(Foo.Succ): [] > In this output the dependecy is as follows: Foo.zero `depends_on` [] Foo.one `depends_on` [Foo.Zero] -- this is the source of my question Foo.ten `depends_on` [Foo.one] Foo.Zero `depends_on` [] Foo.Succ `depends_on` [] Foo.one here saves the Foo.Zero as an argument of constructor application, i. e. saves Foo.Zero's address in one of the allocated closure's field. So I think Foo.Zero must be defined before Foo.one. -- vir http://vir.comtv.ru/

1 0

GHC 6.8.1 on powerpc OS X 10.5
by Chris Kuklewicz 19 Nov '07

19 Nov '07

So I decided to try and route around the ld/collect2 problems with ghc-6.8.1 not working on powerpc OS X 10.5 OS X 10.5 (Leopard) uses XCode 3.0 which has a gcc-4.0 toolchain. So I try to get another gcc & binutils from macports. This failed. All of gcc40 gcc41 gcc42 and gcc43 failed to compile successfully with macports. So I switched to trying fink, which offers gcc-4.2.2 and this did compile and install. Hooray. Unfortunately, using ghc-6.6.1 (from when I ran OS 10.4 Tiger) with gcc-4.2.2 does not work when compiling ghc-6.8.1 (at least on this platform). > ------------------------------------------------------------------------ > == make boot -r; > in /Users/chrisk/Documents/projects/haskell/build/ghc-6.8.1/ghc-6.8.1/compiler > ------------------------------------------------------------------------ > ../utils/mkdirhier/mkdirhier stage1 > for i in utils basicTypes types hsSyn prelude rename typecheck deSugar coreSyn vectorise specialise simplCore stranal stgSyn simplStg codeGen main profiling parser cprAnalysis ndpFlatten iface cmm nativeGen; do \ > ../utils/mkdirhier/mkdirhier stage1/$i; \ > done > Creating stage1/ghc_boot_platform.h... > Done. > ../utils/genprimopcode/genprimopcode --primop-tag < prelude/primops.txt > primop-tag.hs-incl > /bin/sh: line 1: 59384 Illegal instruction ../utils/genprimopcode/genprimopcode --primop-tag < prelude/primops.txt > primop-tag.hs-incl > make[1]: *** [primop-tag.hs-incl] Error 132 > make[1]: *** Deleting file `primop-tag.hs-incl' > make: *** [stage1] Error 1 Running "../utils/genprimopcode/genprimopcode --primop-tag" produces "Illegal instruction" There were other odd messages during the make,such as in this early output: > ------------------------------------------------------------------------ > == make boot -r; > in /Users/chrisk/Documents/projects/haskell/build/ghc-6.8.1/ghc-6.8.1/compat > ------------------------------------------------------------------------ > ../utils/mkdependC/mkdependC -f .depend -I. -Iinclude -I../includes -- -O -I. -Iinclude -D__GHC_PATCHLEVEL__=1 -I../libraries/base/cbits -I../libraries/base/include -- cbits/directory.c cbits/rawSystem.c cbits/unicode.c > /usr/local/bin/ghc -M -optdep-f -optdep.depend -osuf o -H16m -O -w -I. -Iinclude -Rghc-timing -ignore-package Cabal -ignore-package filepath -I../libraries -fglasgow-exts -no-recomp Compat/Directory.hs Compat/RawSystem.hs Compat/Unicode.hs Distribution/Compat/Directory.hs Distribution/Compat/Exception.hs Distribution/Compat/Map.hs Distribution/Compat/RawSystem.hs Distribution/Compat/ReadP.hs Distribution/Compat/TempFile.hs Distribution/Compiler.hs Distribution/GetOpt.hs Distribution/InstalledPackageInfo.hs Distribution/License.hs Distribution/Package.hs Distribution/ParseUtils.hs Distribution/Simple/Program.hs Distribution/Simple/Utils.hs Distribution/System.hs Distribution/Verbosity.hs Distribution/Version.hs Language/Haskell/Extension.hs System/FilePath.hs System/FilePath/Posix.hs System/FilePath/Windows.hs Trace/Hpc/Mix.hs Trace/Hpc/Tix.hs Trace/Hpc/Util.hs > <<ghc: 36490248 bytes, 5 GCs, 130248/130248 avg/max bytes residency (1 samples), 16M in use, 0.01 INIT (0.01 elapsed), 0.22 MUT (4.47 elapsed), 0.06 GC (0.43 elapsed) :ghc>> > make all > /usr/local/bin/ghc -H16m -O -w -I. -Iinclude -Rghc-timing -ignore-package Cabal -ignore-package filepath -I../libraries -fglasgow-exts -no-recomp -c System/FilePath/Posix.hs -o System/FilePath/Posix.o -ohi System/FilePath/Posix.hi > gcc: unrecognized option '-no-cpp-precomp' > > /var/folders/CT/CTQA-fMz2RWZwk+BYmhr7U++-Eg/-Tmp-//ghc64877_0/ghc64877_0.s:6606:0: > section difference relocatable subtraction expression, "_s2tx_0_alt" minus "_s2tx_info" using a symbol at the end of section will not produce an assembly time constant > > /var/folders/CT/CTQA-fMz2RWZwk+BYmhr7U++-Eg/-Tmp-//ghc64877_0/ghc64877_0.s:6606:0: > use a symbol with a constant value created with an assignment instead of the expression, L_const_sym = _s2tx_0_alt - _s2tx_info The '-no-cpp-precomp' and 'section difference ...' and 'use a symbol ...' messages repeat often. I looked at trac, and can only find the closed ticket http://hackage.haskell.org/trac/ghc/ticket/1427 relating to gcc-4.2 issues. Does anyone have suggestions? hints? bad jokes? -- Chris K

5 8

GHC 6.8.1 on Mac OS X 10.4 (Tiger)
by Christian Maeder 17 Nov '07

17 Nov '07

Dear Intel Mac Users, I've also created a binary distribution of GHC 6.8.1 for Mac OS X 10.4 (Tiger) that is based on gmp3 and readline5 libraries under /opt/local/ (without framework dependencies). You can set the DYLD_LIBRARY_PATH if your gmp3 and readline5 libraries are elsewhere. http://www.informatik.uni-bremen.de/agbkb/forschung/formal_methods/CoFI/het… Christian I wonder if that distribution would run under Leopard, too.

2 1

ghc trac wiki, rss feeds not activated?
by Claus Reinke 17 Nov '07

17 Nov '07

a lot of ghc-related discussion takes place on ghc's trac wiki these days, as does the development of ghc docs. so i'd like to follow wiki changes as an rss feed. this is supported, according to http://hackage.haskell.org/trac/ghc/wiki/TracRss http://hackage.haskell.org/trac/ghc/wiki/TracTimeline unfortunately, i haven't seen that xml/rss icon on any ghc trac wiki page, and the url modifier is not otherwise documented, either. is that an optional feature that hasn't been activated? could rss feeds for the ghc trac wiki please be made available? claus

2 2

GHC 6.8.1 on FreeBSD 7/amd64
by Brian P. O'Hanlon 17 Nov '07

17 Nov '07

Hello, all. I successfully built GHC 6.8.1 on my FreeBSD 7.0_BETA amd64 machine. I was able to bootstrap it from the GHC 6.6.1 for FreeBSD 6/amd64 which was posted to this list a while ago (and was running in FreeBSD 6 compatibility mode), as well as happy 1.15 which I compiled with that same compiler. I did have to make a minor modification; in the configure script, the machine type needed to be changed from amd64-*-freebsd* to x86_64-*-freebsd* (the same needed to be done for happy when I compiled that earlier). The C compiler was the stock gcc 4.2 compiler. GHCi does not work, but I did not really expect that it would since it did not in 6.6.1. I do not know how useful this might be to anyone else; mine is a rather esoteric setup (I am really only even using FreeBSD 7 because FreeBSD 6 does not support my video card or SATA controller), but if people are interested in binaries, I should be able to do that. -Brian -- Brian P. O'Hanlon, <brianpo+(a)cmu.edu>, 7039012836, KG4QXK Carnegie Mellon University - Math, Computer Science, and English Chairman, Spring Carnival; Station Manager, W3VC; Co-President, Math Club; Social Director, Computer Club

2 1

GHC 6.8.1 SpecConstr
by kenny lu 17 Nov '07

17 Nov '07

Hi, Recall the example from Simon's paper appearing in ICFP 2007. myLast :: [a] -> a myLast [] = error "" myLast [x] = x myLast (x:xs) = myLast xs which returns the last element in a list. Applying the idea in the paper, we should rewrite the function as follows, myLast :: [a] -> a myLast [] = error "" myLast (x:xs) = myLast' x xs where myLast' x [] = x myLast' x (y:ys) = myLast' y ys which avoids redundant pattern tests. We tried to compile the "unspecialized" version of myLast function with GHC 6.8.1 with -O2 -ddump-simpl flags and observe the core code as follows, ==================== Tidy Core ==================== lvl_r72 :: forall a_a5P. a_a5P [GlobalId] [Str: DmdType b] lvl_r72 = \ (@ a_a5P) -> GHC.Err.error @ a_a5P (GHC.Base.[] @ GHC.Base.Char) Rec { Last.$smyLast :: forall a_a5P. a_a5P -> [a_a5P] -> a_a5P [GlobalId] [Arity 2] Last.$smyLast = \ (@ a_a5P) (ipv_s6i :: a_a5P) (ipv1_s6j :: [a_a5P]) -> case ipv1_s6j of wild_X9 { [] -> ipv_s6i; : ipv2_X6p ipv3_X6r -> Last.myLast @ a_a5P wild_X9 } Last.myLast :: forall a_a5B. [a_a5B] -> a_a5B [GlobalId] [Arity 1 Str: DmdType S] Last.myLast = \ (@ a_a5P) (ds_d68 :: [a_a5P]) -> case ds_d68 of wild_B1 { [] -> lvl_r72 @ a_a5P; : x_a5I ds1_d69 -> case ds1_d69 of wild1_X9 { [] -> x_a5I; : ipv_s6i ipv1_s6j -> Last.myLast @ a_a5P wild1_X9 } } end Rec } ==================== Tidy Core Rules ==================== "SC:Last.myLast0" [0] forall {@ a_a5P ipv_s6i :: a_a5P ipv1_s6j :: [a_a5P]} Last.myLast @ a_a5P (GHC.Base.: @ a_a5P ipv_s6i ipv1_s6j) = Last.$smyLast @ a_a5P ipv_s6i ipv1_s6j It seems that $smyLast is the specialized version of myLast And the last tidy core rule says that we should replace the application (myLast (x:xs)) by ($smyLast x xs). We are surprised by the development of the second pattern of $smyLast. In the second pattern, function myLast is applied to wild_X9 which is the reference of ipv_s6j. However, according to the running example, it seems to make more sense if this application is replaced by a recursive application ($smyLast ipv_3X6r) instead. Could you guys please enlighten me? Regards, Kenny

4 3

Problem with functional dependencies
by Daniel Gorín 16 Nov '07

16 Nov '07

Hi I have some code that uses MPTC + FDs + flexible and undecidable instances that was working fine until I did a trivial modification on another part of the project. Now, GHC is complaining with a very confusing (for me, at least) error message. I've been finally able to reproduce the problem using these three small modules: > {-# LANGUAGE MultiParamTypeClasses #-} > {-# LANGUAGE FunctionalDependencies #-} > {-# LANGUAGE FlexibleInstances #-} > module M1 > > where > > data M n = M > data F n = F > > class C m f n | m -> n, f -> n where > c :: m -> f -> Bool > > instance C (M n) (F n) n where > c _ _ = True > {-# LANGUAGE MultiParamTypeClasses #-} > {-# LANGUAGE UndecidableInstances #-} > {-# LANGUAGE FlexibleInstances #-} > module M2 > > where > > import M1 > > newtype F'= F' (F N) > > data N = N > > instance C m (F N) N => C m F' N where > c m (F' f) = c m f > module M3 > > where > > import M1 > import M2() > > data N' = N' > > go :: M N' -> F N' -> Bool > go m f = c m f Now, when trying to compile M3 (both in 6.6.1 and 6.8.1) I get: M3.hs:11:0: Couldn't match expected type `N'' against inferred type `M2.N' When using functional dependencies to combine C m M2.F' M2.N, arising from the instance declaration at M2.hs: 13:0 C (M N') (F N') N', arising from use of `c' at M3.hs:11:9-13 When generalising the type(s) for `go' It is worth observing that: - M2 compiles fine - No type defined in M2 is visible in M3 - if the "import M2()" is commented out from M3, it compiles fine - if, in M3, N' is placed by N (needs to be imported), everything compiles again Normally, it takes me some time to digest GHC's type-classes-related error messages, but after some reflection, I finally agree with them. This time, however, I'm totally lost. I can't see any reason why N' and M2.N would have to be unified, nor why this code should be rejected. Any help would be much appreciated! Daniel

2 3

6.8.1: source code in docs?
by Greg Fitzgerald 16 Nov '07

16 Nov '07

GHC 6.8.1's docs no longer link to the source code in the upper right-hand corner of each module. Intentional? http://haskell.org/ghc/docs/latest/html/libraries/ Thanks, Greg

1 0