May 2002 - Glasgow-haskell-users

RE: Building GHC 5.02.3 from bootstrap.
by Simon Marlow 03 May '02

03 May '02

> In spite of all the warnings, I tried to build ghc 5.02.3 > from scratch. > The first step, > > ./distrib/hc-build > returned the error > >make[1]: *** No rule to make target `Addr.o', needed by > `libHSlang.a'. Stop. > >make: *** [all] Error 1 hmm, do you have Addr.hc in hslibs/lang? > I'm actually building on a standard Linux x86 platform. The > resaon I want to > build from scratch is to incorporate GHC into the portage > source tree for > the Gentoo distribution, http://www.gentoo.org The way I'm doing this for the FreeBSD port now is to have the build process download a pre-built cut-down binary distribution of GHC purely for bootstrapping. We used to bootstrap from .hc files in the ports tree but after a while it was just too fragile and hard to get right. Cheers, Simon

1 0

Porting guide
by Simon Marlow 02 May '02

02 May '02

Hi Folks, I've just updated the Building Guide to include a section on porting GHC to a new platform. Please take a look: http://www.haskell.org/ghc/docs/latest/building/sec-porting-ghc.html Disclaimer: I haven't actually followed the steps in doing an unregisterised bootstrap from beginning to end, so there's sure to be some inaccuracies in there. Cheers, Simon

1 0

last result in ghci
by Serge D. Mechveliani 02 May '02

02 May '02

Dear GHC, Do you provide in your interactive system a denotation for the last result? (`:?' command does not show it). For example, Prelude> 1 1 Prelude> % + 1 2 ... ----------------- Serge Mechveliani mechvel(a)botik.ru

3 2

RE: profiling crashes my code
by Simon Peyton-Jones 02 May '02

02 May '02

| Okay, I've only just toyed with profiling myself, but one | observation I've made is that it's not a good idea to mix | modules compiled with profiling and modules compiled without. | | That is, always 'make clean' (i.e. 'rm -f *.o *.hi') before | you build with a new set of options. Absolutely. GHC should complain if you mix profiled and non-profiled code but it currently doesn't. You'll certainly get disaster if you mix them. (The format of heap objects changes.) I bet that's your problem. S

1 0

profiling crashes my code
by Hal Daume III 02 May '02

02 May '02

I'm in a profiling kind of mood (read: am sick of my program taking too long to run), but when I ghc with -prof -auto-all (or -auto or whatever), when I try to run my program it seg faults. I don't want to attach all the code, since it's really long, but I *have* tested profiling on smaller programs and it works fine. (Note that the program crashes regardless of whether I put +RTS -p or not.) Here are some "unusual" things I'm doing; hopefully someone will be able to say "oh, there's your problem": - using the ghc binary package (which using FastMutInt -- I don't really know what this does, but it looks scary) - i'm only compiling the main program with profiling -- it imports a whole lot of modules which are compiled with -O2, including binary which is compiled using -cpp also - i use the Word8 type extensively - i use unsafePerformIO (but only once per run of the code) - i use UArray Int Double (i think this also uses upIO) - uhm...i do a lot of file io :) That's all I can think of at the moment. It dies regarless of whether I use -O, -O2, or neither... -- Hal Daume III "Computer science is no more about computers | hdaume(a)isi.edu than astronomy is about telescopes." -Dijkstra | www.isi.edu/~hdaume

3 2

problems with ld
by Hal Daume III 01 May '02

01 May '02

This isn't a ghc specific question as I've just found a similar problem with gcc, but I was hoping someone here could lend me a hand...I recently installed RedHat 7.2 on my x86 laptop and then immediately installed ghc 5.02.3 from rpm. Compilation works fine as does interpretation but linking doesn't and it seems to be a problem with ld. It complains "/usr/bin/ld: cannot find -lc". A similar thing happens in C, except in that ld cannot find -lc. I'm wondering if anyone has any idea what this means and how to fix it (I'm not a *nix hacker by any means). I've attached outputs from ghc and gcc when run with -v; the important error is at the very bottom. The test programs where, for haskell, "main = print 5" and for c "int main() { printf("Hello world"); }". Thanks for any help! - Hal -- Hal Daume III "Computer science is no more about computers | hdaume(a)isi.edu than astronomy is about telescopes." -Dijkstra | www.isi.edu/~hdaume

2 1

profiling (again)
by Hal Daume III 01 May '02

01 May '02

i'm using ghc 5.02.1 for sparc solaris -- Hal Daume III "Computer science is no more about computers | hdaume(a)isi.edu than astronomy is about telescopes." -Dijkstra | www.isi.edu/~hdaume

1 1

RE: No NoFib
by Simon Marlow 01 May '02

01 May '02

> I can't find a tarred up release of the NoFib suite, I can only find > it in the CVS repository. Is this intentional? > > The link from GHC's home page points to a very old page at Glasgow > which doesn't give a link to the software. (Perhaps the link in the > paper still works - I haven't tried). We haven't done releases of the NoFib suite for a long time - there doesn't seem to be a need now you can get the sources from CVS. We generally tag NoFib at the same time as a GHC release, so to get the NoFib sources that correspond to GHC 5.02.3, for example, 'cvs co -rghc-5-02-3 nofib'. Cheers, Simon

2 1

No NoFib
by Kevin Glynn 01 May '02

01 May '02

Hi, I can't find a tarred up release of the NoFib suite, I can only find it in the CVS repository. Is this intentional? The link from GHC's home page points to a very old page at Glasgow which doesn't give a link to the software. (Perhaps the link in the paper still works - I haven't tried). k

1 0

announce: GHC internals library (version 0.1)
by Bernard James POPE 01 May '02

01 May '02

Hi all, Announcing the GHC internals library, version 0.1. Essentially this provides a means for polymorphically deconstructing values that reside on the GHC heap, all from the comfort of your Haskell program. For those who are familiar with the HugsInternals library that comes with Hugs, it offers similar support to getCell and classifyCell, in a GHC-centric way. What might you use this for? I don't know, but I'm using it for debugging, maybe you can too, or alternatively, you might find another use. You can find it here: http://www.cs.mu.oz.au/~bjpop/code.html For your curiosity, here's _one_ thing you can do with it: > import GhcInternals > > main :: IO () > main = do let val = zip [1..12] (iterate not True) > print $ take 3 val > metaVal <- reify False val > putStrLn $ prettyPrintMeta metaVal [] Running this program you get: [(1,True),(2,False),(3,True)] (: ((,) 1 True) (: ((,) 2 False) (: ((,) 3 True) THUNK))) How does it work? Basically, using some odd looking FFI code it peaks at the heap representation of a value, and constructs a meta representation of it. The deconstruction is only done one constructor at a time, although provision is made for the whole structure to be inspected (more or less) at once. Cycles and sharing in general are not yet supported, but future versions will attempt to solve this, perhaps with some tinkering with Stable Pointers. You can choose to inspect the structure lazily (ie no further evaluation) or strictly (ie force the structure to evaluate one constructor at a time). Functions can be printed, but mostly you'll get some funny string that the compiler generated. This might be improved in later versions of the library. Currently it relies on compiling your program with -prof to make sure the names of constructors are retained in the heap. No modifications are required to your compiler, it should compile out of the box without any hackery to GHC itself. Works with version 5.03 of GHC, but you need a very current version to avoid exposing some bugs in the Stable Pointer implementation. Cheers, Bernie.

2 1