September 2002 - Glasgow-haskell-users

RE: creating a library from haskell files and a c wrapper in Linux
by Simon Marlow 10 Sep '02

10 Sep '02

> tis 2002-09-10 klockan 16.08 skrev Martin Norbäck: > > Now I have done the following: > > > > # ghc -fglasgow-exts -c Adder.hs > > # gcc -c -DDLL_IMPORT= -I/usr/lib/ghc-5.04/include dllMain.c > > # ld -shared -o adder.so Adder.o Adder_stub.o dllMain.o > > -L/usr/lib/ghc-5.04 -lHSbase -lHSbase_cbits -lHSrts -lHSbase -lgmp > > # gcc test.c -L. -ladder > > # env LD_LIBRARY_PATH=. ./a.out > > > > and then I get the following: > > > > a.out: fatal error: LOOKS_LIKE_GHC_INFO- is incorrectly defined > > > > What to do next, I wonder... > > Here are the files used so you can test for yourself. Oh, and adder.so > above should be libadder.so. You can't make a dynamic library containing Haskell code on Unix, sorry. What actually is it you're trying to do? Cheers, Simon

1 0

RE: Time out
by Simon Marlow 10 Sep '02

10 Sep '02

> Ashley Yakeley wrote: > > | > timeOut :: Int -> IO a -> IO (Maybe a) > : > | Sounds rather unsafe. What happens to the IO action > | after the specified time? > > Good point. Actually, a function of the type: > > timeOut :: Int -> a -> IO (Maybe a) > > would suffice for my purposes, but I tried to be a bit more > general when I posed my question. It's only unsafe if your code isn't written to cope with asynchronous exceptions. Doing this in most cases just amounts to using combinators like Control.Exception.bracket to do the exception handling for you. In FFI code you have to be more careful, however. Cheers, Simon

2 1

creating a library from haskell files and a c wrapper in Linux
by Martin Norbäck 10 Sep '02

10 Sep '02

I'm trying to create a (dynamic) library from a bunch of Haskell modules and some C code that calls functions exported from Haskell via the FFI. The problem is that I don't know how to link everything into a library. The first problem I had was that DLL_IMPORT which was found in some header files is undefined in Linux, so I just passed -DDLL_IMPORT= to gcc. The next problem is that I can't get ghc to make a library file for me, the --mk-dll flag just makes ghc output nothing, and I can find no other flag to make ghc create a library instead of a binary. I'll try to find out the right libraries to pass to ld, but it would be much easier if ghc could do it by itself, as it can do under Windows. Regards, Martin -- Martin Norbäck d95mback(a)dtek.chalmers.se Kapplandsgatan 40 +46 (0)708 26 33 60 S-414 78 GÖTEBORG http://www.dtek.chalmers.se/~d95mback/ SWEDEN OpenPGP ID: 3FA8580B

1 2

RE: FFI breaking with -O?
by Simon Marlow 10 Sep '02

10 Sep '02

> I have a variety of C interfaces which are generated with hsc2hs, in > particular a ncurses and openssl wrapper. they work perfectly when > compiling normally, but when i use the -O or -O2 option, I end up with > at the least a ton of C warnings on the /tmp/ghc*.hc file and in some > cases a fatal error. as far as i can tell they relate to some odd > interaction with the C preprocessor and the generated haskell > code. I am > not compiling with -fvia-C but it appears to be going via C > anyway. Does > anyone have any ideas why this might be the case or a > workaround for the > problem? Perhaps if this is the case, I can force it to not > compile via > C even when optimizing? > John > > (the offending code is part of 'ginsu' my gale client implemented in > haskell http://repetae.net/john/computer/ginsu/ ) This "shouldn't" happen, although there are occasionally good reasons why it does. Are you including the right header files using the -#include option? Can you send us some sample code to test, or should I try the whole of ginsu? You can force non-via-C by saying -fasm. Put this flag *after* -O, because -O implies -fvia-C. Cheers, Simon

3 2

Re: Time out
by Ashley Yakeley 10 Sep '02

10 Sep '02

At 2002-09-10 03:03, Koen Claessen wrote: >What is the recommended way of implementing the function: > > timeOut :: Int -> IO a -> IO (Maybe a) > >which takes a maximum number of seconds (say), an IO action, >and tries to run the IO action in the specified time. If it >succeeds, it returns just the answer, otherwise it returns >Nothing. Sounds rather unsafe. What happens to the IO action after the specified time? -- Ashley Yakeley, Seattle WA

2 1

RE: FFI breaking with -O?
by Simon Marlow 10 Sep '02

10 Sep '02

> okay, I figured out putting the -fasm after the -O and that seems to > cause the problem to go away supporting my theory. > For an example of something that breaks, the following will create an > error when compiled via-C and not otherwise. > > #include <ncurses.h> > foreign import ccall touchwin :: Window -> IO CInt > > for the ncurses included with red hat linux, i think it has > something to > do with the fact that touchwin is both #defined in terms of other > routines AND part of the ncurses library as its own function. > I imagine > they did this to allow inlining when compiling C as well as external > bindings and function pointers to refer to the function. many other > routines give warnings which are #definied in similar ways. Ah yes, the function above is #defined as follows: #define touchwin(win) wtouchln((win), 0, getmaxy(win), 1) where getmaxy() is a macro which expands to an indirect structure selection on its argument. In GHC's via-C code, the argument is not declared as a pointer to the right structure, so C compiler complains. I think the right thing to do here is to have your own header file which just #undefs all these macros, and #include it after ncurses.h. Cheers, Simon

1 0

Time out
by Koen Claessen 10 Sep '02

10 Sep '02

Hi all, What is the recommended way of implementing the function: timeOut :: Int -> IO a -> IO (Maybe a) which takes a maximum number of seconds (say), an IO action, and tries to run the IO action in the specified time. If it succeeds, it returns just the answer, otherwise it returns Nothing. Do I have to start two threads, one that waits and then throws an exception, the other that evaluates the IO action? Or is there a better way? (I thought of asynchronous exceptions...) /Koen. -- Koen Claessen http://www.cs.chalmers.se/~koen Chalmers University, Gothenburg, Sweden.

1 0

FFI breaking with -O?
by John Meacham 10 Sep '02

10 Sep '02

I have a variety of C interfaces which are generated with hsc2hs, in particular a ncurses and openssl wrapper. they work perfectly when compiling normally, but when i use the -O or -O2 option, I end up with at the least a ton of C warnings on the /tmp/ghc*.hc file and in some cases a fatal error. as far as i can tell they relate to some odd interaction with the C preprocessor and the generated haskell code. I am not compiling with -fvia-C but it appears to be going via C anyway. Does anyone have any ideas why this might be the case or a workaround for the problem? Perhaps if this is the case, I can force it to not compile via C even when optimizing? John (the offending code is part of 'ginsu' my gale client implemented in haskell http://repetae.net/john/computer/ginsu/ ) -- --------------------------------------------------------------------------- John Meacham - California Institute of Technology, Alum. - john(a)foo.net ---------------------------------------------------------------------------

1 0

RE: Packages (again)
by Simon Marlow 09 Sep '02

09 Sep '02

> (adding the linker flags to ld_opts does not make a > difference for the rest of the story) > > To make everything work with ghc, I do the following: > > * gcc -c foo.c -o foo.o > > * ghc -c Foo.hs -o Foo.o > > * ar -rsv libHSFoo.a foo.o Foo.o > > Now, I can compile programs with "ghc -package foo". Great! > > However, when I try ghci, it does not work. So, first, I > make a .o version of libHSFoo.a: > > * gld -r --whole-archive -o HSFoo.o libHSFoo.a > > But, when I start "ghci -package foo", it does not find the > library libbar! It does not even find it when I say "ghci > -L/path/to/libs -lbar -package foo"! You should add "bar" to the 'extra_libraries' field of the package config, because otherwise GHCi doesn't know about the dependency. > No worries, we can make ghci read a .so file instead: > > * rm HSFoo.o > > * gcc -shared Foo.o foo.o -L/path/to/libs -lbar -o libHSFoo.so > > Starting ghci now loads libHSFoo.so, and everything works > wonderfully. I'm highly surprised if this really works. Can you actually call functions from the Foo module without it crashing? Cheers, Simon

1 0

RE: ghc 5.04 -- make -prof -auto-all under WinNT
by Simon Peyton-Jones 09 Sep '02

09 Sep '02

Dear Win32 GHC folk It does seem that GHC 5.04 profiling was utterly broken. It now seems that this might have been to do with the version of gcc we were using. It seems to work with 5.04.1 and a new version of gcc, which (though Sigbjorn's good offices) will come inside 5.04.1's distribution. So our claim is that 5.04.1 will work for profiling. We hope to get it out this week. Can you try again then? Thanks Simon | -----Original Message----- | From: andreas.marth(a)daimlerchrysler.com | [mailto:andreas.marth@daimlerchrysler.com] | Sent: 15 August 2002 16:58 | To: glasgow-haskell-users(a)haskell.org | Subject: ghc 5.04 -- make -prof -auto-all under WinNT | | Hallo! | | I tried to compile the following little program (hallo.hs) with ghc-5.04 --make | -prof -auto: | | import System (getArgs) | | | main :: IO () | main= do args <- getArgs | if args == [] then error "Aufruf: Filename Dimension1 Dimension2 | ..." | else prog (head args) (tail args) | | prog :: FilePath -> [String] -> IO () | prog file dimZ = do cont <- readFile file | putStr (file ++ "\n") | putStr (show dimZ ++ "\n") | let result = "Hello world!" | writeFile (file++".out") ( result) | | The system is WinNT and compiling succeeded. But I got an a.out file while I | should get a file hallo.exe (according to ghc --help). | I renamed the program to hallo.exe and tried it. When I don't give any | arguments it behaves like expected. If I give it one or more arguments it | crashes. | | When I compile it only with ghc --make hallo.hs I also get an a.out file but | this behaves well when renamed to hallo.exe. | | What is the problem? And what can I do to use the profiling? | | Thank, | Andreas

1 0