- Nhc-users - Haskell.org

Doubts in bytecode
by Arunkumar S Jadhav 17 Mar '05

17 Mar '05

Hello, I've two questions which are unrelated. I'll ask them one by one. Q1) I've this function: myFoo :: (Int -> Int -> Int) -> Int -> Int -> Int myFoo f x y = f (x+y) (x-y) And it compiles to the following bytecode (I've pasted just the byte code part, just ignore other details). bytes2word(NEEDHEAP_I32,PUSH_HEAP,HEAP_CVAL_I3,HEAP_ARG_ARG) , bytes2word(2,3,PUSH_HEAP,HEAP_CVAL_I4) , bytes2word(HEAP_ARG_ARG,2,3,PUSH_P1) , bytes2word(0,PUSH_P1,2,PUSH_ZAP_ARG_I1) , bytes2word(ZAP_ARG_I2,ZAP_ARG_I3,ZAP_STACK_P1,4) , bytes2word(ZAP_STACK_P1,3,EVAL,NEEDHEAP_I32) , bytes2word(APPLY,2,RETURN_EVAL,ENDCODE) Okay, as I've come to understand, this is how things work out. 1. Construct the graph for (x + y) and push its root onto stack. PUSH_HEAP, HEAP_CVAL_I3, HEAP_ARG_ARG, 2, 3 2. Construct the graph for (x - y) and push its root onto stack. PUSH_HEAP,HEAP_CVAL_I4, HEAP_ARG_ARG, 2, 3 3. Now push the first argument onto heap, make top of stack point to it and evaluate it then apply its result to top two elements on stack. (I understand that arguments have to be ZAPped before we start evaluating them, so I know what ZAP is doing here) PUSH_ZAP_ARG_I1, ZAP_ARG_I2,ZAP_ARG_I3,ZAP_STACK_P1,4 ZAP_STACK_P1,3,EVAL,NEEDHEAP_I32, APPLY,2 Now in between all these contents of stack that were pointing to two graphs (i.e x+y and x-y) are being replicated on the stack and then one of the copies (of both the graphs) is being zapped as can be seen from the code snippet below. PUSH_P1, 0, PUSH_P1,2, ,ZAP_STACK_P1,4, ZAP_STACK_P1,3 This has caught me off guard. I just couldn't figure out why is this being done ? i.e Why copy the contents of stack and then zap one of them. Also what all are the uses of ZAP nodes apart from black hole detection. Do ZAP nodes help in garbage collection too ? Now second doubt. Q2) As Malcolm explained in detail this is the purpose of CONSTR macro CONSTR(c,s,ws) Construct a tag (i.e. a header for a data node) where there is a mixture of pointers and basic values amongst the data items I compiled various examples but till now I haven't seen a single example where CONSTR is used for a mixture of pointers and basic values. It has always been for basic values. I'll illustrate with an example. data MyData a = One | Two a | Three a (MyData a) (MyData a) | Four a a a a a intToData :: Int -> MyData Int intToData n = Three 12 (Two 13) One Now the constant table for function "intToData" contains these entries. , CONSTR(2,3,0) , CONSTR(0,0,0) , CONSTR(1,1,0) Clearly, they correspond to the constructors "Three", "One" and "Two" respectively. I expected to see something like CONSTR(2,3, 2) for "Three" constructor because it's second and third field are supposed to be pointers. But this is not the case. Can you please give an example where CONSTR is actually used to construct a mixture of pointer and basic data nodes. Thanks for your patience in going through this. Any help regarding this will be highly appreciated. Regards, ------------------------------------ Arunkumar S Jadhav, Masters Student, KReSIT, IIT-Bombay, India Ph: +91-22-25764967 http://www.it.iitb.ac.in/~arunk ------------------------------------ I exist because I work.

2 1

Build failure of nhc98-1.18 on Mac OS X 10.3.8 with gcc 3.3
by Gregory Wright 14 Mar '05

14 Mar '05

Hi, I've encountered a build failure of nhc98 1.18 on Mac OS X 10.3.8 using Apple's gcc 3.3. The build fails with a bus error, as can be seen in the attached log. GHC is not installed; the build is using gcc. The build was done with 'ulimit -s unlimited' set and also used gnu make 3.79. The build is part of the darwinports infrastructure on Mac OS X, which builds software from source in the manner of the FreeBSD ports collection. The build commands were: CC=gcc ./configure --buildwith=gcc --mandir=/opt/local/share/man ulimit -s unlimited && make Can anyone tell me if I've done something blindingly stupid? Please cc me as I am not subscribed to the nhc lists. Thanks! Best Wishes, Greg Wright

1 0

ANNOUNCE: nhc98-1.18 and hmake-3.10
by Malcolm Wallace 11 Mar '05

11 Mar '05

To complement today's fresh release of ghc-6.4.... ------------------------- nhc98-1.18 + hmake-3.10 ------------------------- http://haskell.org/nhc98 http://haskell.org/hmake We are pleased to announce a new release of the nhc98 compiler and the hmake compilation manager. nhc98 is portable to almost any 32-bit unix-like platform, and can be easily bootstrapped with just a C compiler if no existing Haskell compiler is available. Hmake plays happily with ghc and hbc as well as nhc98. This release is mainly a maintenance refresh, fixing numerous bugs, and adding support for many of the new hierarchical libraries that are also supplied with ghc-6.4 and Hugs. Main bugfixes: * Resolved configuration bugs when using ghc-6.2 and ghc-6.4 with hmake. * Build failures associated with gcc-3.3 and upward have been fixed. Main feature updates: * Supplied with these standard library packages: base, parsec, haskell-src, QuickCheck, HaXml, HUnit, Cabal * FFI: foreign import "dynamic" is now supported, and C header-files named in an import are now respected. * cpphs: The compiler and hmake now use cpphs rather than your C compiler for preprocessing. This removes problems with string gaps, primes in identifiers, and so on. * Source now supplied with a small test suite to help verify whether the compiler works on new platforms. This release of nhc98 has been tested with the following machines and build compilers: ix86-Linux (Slackware 10 and RedHat 7.2 built with ghc-5, ghc-6, hbc, gcc-2.95, gcc-3.3, nhc98) ix86-Cygwin (Windows 2000 built with gcc-3.3, ghc-6) sparc-solaris2 (SunOS 5.9 built with ghc-6, gcc-2.95, nhc98) powerpc-Darwin7 (MacOS X 10.3 using ghc-6, gcc-2.95, gcc-3.3) Credits: Malcolm Wallace (release manager) Ian Lynagh (gcc-3.3 bugfixes, hmake-config add-dyn) Wolfgang Thaller (MacOS gcc-3.3 bugfixes) Simon Marlow (ghc configuration fixes) Sven Panne (warning police, hbc build) numerous library authors, distribution packagers, and bug reporters.

1 0

Data nodes
by Arunkumar S Jadhav 03 Mar '05

03 Mar '05

Hello, As per my understanding following hold true about macroes for data nodes. CONSTR(e, s, ws): This is used to construct header for the data nodes which contain more than one constructors e.g List, Tree etc CONSTRW(s, e) and CONSTRP(s, e) : Are used to construct header for data nodes which contain single constructor e.g Pair. I just wrote a program : ---------------------------------- data MyPair a = ThisPair a a a selectThird (ThisPair 1 x y) = ThisPair 1 1 1 selectThird (ThisPair _ _ y) = ThisPair y y y ------------------------------------ When I saw the corresponding bytecode, I saw (To my surprize) CONSTR(0, 3, 0) being used to construct the data node in the function selectThird (i.e "ThisPair 1 1 1" etc). What is happening here ? If this is the case, then when exactly are the meaning of CONSTR() and CONSTRW(), CONSTRP() Regards, ------------------------------------ Arunkumar S Jadhav, Masters Student, KReSIT, IIT-Bombay, India Ph: +91-22-25764967 http://www.it.iitb.ac.in/~arunk ------------------------------------ I exist because I work.

2 1

GUI ?
by John Velman 01 Mar '05

01 Mar '05

Please excuse me if this is obvious, or if it's obvious how to get one of the GUI packages listed on the Haskell website to work with NHC, but the ones I've looked into all seem (to me) to be dependent on GHC. Is there a GUI package that works with nhc? Thanks, John Velman

3 3

NHC98 & Cygwin
by Lapo Luchini 24 Jan '05

24 Jan '05

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Hi! Yesterday I "casually discovered" the existance of the Haskell language, became intrigued by it, erad the "Genet introduction", and gone to the NHC98 homepage to being "experimenting"... there I found a pre-compiled Cygwin version, that suggested to use Cygwin as downloadable with setup.exe (thanks god you didn't decide to distribute it "binary" with cygwin1.dll, as many do: having multiple cygwin1.dll in the system is indeed a problem). Now I wonder why don't you create an "official" Cygwni package installable directly with setup.exe? Whouldn't it be handy? Do you lack the "specific know-how" (in that case I could help), the will to manage yet another format or something else? Anyway I'm (locally) trying to build a cygwin package compatible with Cygwin's specs, so that it could be included in the distribution... then it's up to you if you want to mantain it or if you prefer someone else to do it (in that case I could volunteer, but I still must decide ^^). Lapo - -- Lapo Luchini lapo(a)lapo.it (PGP & X.509 keys available) http://www.lapo.it (ICQ UIN: 529796) -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.0 (Cygwin) Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org iEYEARECAAYFAkHyQL8ACgkQaJiCLMjyUvsCdwCcC+aW+g2hQU6MfpEv+Rv+wvyN zlMAoJQTXeeUuxwkh5GAuR3WhOtWXwQD =7kuM -----END PGP SIGNATURE-----

2 1

RE: [nhc-users] openFd?
by Simon Marlow 04 Jan '05

04 Jan '05

On 04 January 2005 14:29, Malcolm Wallace wrote: > David Roundy <droundy(a)abridgegame.org> writes: > >> I'm wondering if there is any equivalent to GHC.Handle.openFd which >> I can use with nhc98? It's nice being able to convert a file >> descriptor into a Handle... > > I had a quick look at the documentation for openFd, and it wasn't > very informative - no text, just a signature: > > openFd :: FD -> Maybe FDType -> FilePath -> IOMode > -> Bool -> Bool -> IO Handle It's an internal function not intended for general comsumption, hence the lack of documentation. David: System.Posix.fdToHandle is the right way to do this. On Windows, there ought to be equivalent functionality exposed by the Win32 library (I suspect there isn't though, because on Windows we use file descriptors too, but we really should be using native Windows HANDLEs instead). > However, my guess is that these library functions are actually a > thin skin over some fairly standard C library, in which case you > can probably write a little bit of FFI code to interface to the same > functionality. GHC's IO library is mostly Haskell; there are only a few tiny inline C functions underneath. > nhc98's representation of a Handle as a wrapper over C's FILE* type is > "struct FileDesc", defined in src/runtime.h, and an example of some > FFI stuff to manipulate a Handle can be seen in > src/prelude/IO/OpenFile.hs and src/runtime/Builtin/cOpen.c So with nhc98 you'll need to convert the file descriptor into a FILE* first (using fdopen()?) before you can get a Handle. Cheers, Simon

2 1

openFd?
by David Roundy 04 Jan '05

04 Jan '05

Hello all, I'm wondering if there is any equivalent to GHC.Handle.openFd which I can use with nhc98? It's nice being able to convert a file descriptor into a Handle... For the curious, I'm looking into how hard it might be to make darcs compilable using nhc98, and this is the first sticking point--actually the second one... the first was trivial. openFd is already handled via the configure script, so the interface doesn't need to be the same, as long as the functionality is equivalent, so I can write a wrapper. -- David Roundy http://www.darcs.net

3 2

apply1 and $
by Arunkumar S Jadhav 15 Dec '04

15 Dec '04

Hello All, Byte codes for functions "apply1" and "$" are exactly same (correct me if I'm am wrong). Just "apply1" is used while building a graph and "$" is used (also while building a graph) while interpreting the byte code "APPLY n". Can anyone tell why is this distinction being made ? Because as I see it, apply1 could've been used in place of "$" and the difference wold've been nothing. Regards, ------------------------------------ Arunkumar S Jadhav, Masters Student, KReSIT, IIT-Bombay, India Ph: +91-22-25764967 http://www.it.iitb.ac.in/~arunk ------------------------------------ I exist because I work.

2 1

pointers < 2^31?
by Ian Lynagh 28 Nov '04

28 Nov '04

Hi, Am I confused or does nhc98 assume that all pointers are < 2^31? On m68k malloc is giving me stuff above 0xc0000000 and I get a segfault the first time mutator.c:460 if(GET_TAG(nodeptr) & VAP_TAG && !CINFO_NEED(GET_CINFO(nodeptr))) { is reached. Thanks Ian

2 2