hackery (was: dynamic linking)

hackery (was: dynamic linking)

[Tomas By]

Hello again everybody,

After some experimentation, it now seems clear that this has nothing
to do with garbage collection or Mercury. It is probably just a silly
programming error on my part.

Here follows full source code for a simple example that illustrates
the problem I am having. The important bit is the call to the Mercury
defined procedure `testproc_c.' With that call commented out, the
module loads fine.

The intention is that the `mytype' data is manipulated only in
Mercury, and passed around in the Scheme interface as a C pointer.

Would be glad if somebody could explain what I am missing.

(Guile 1.8.7)

/Tomas

---------- mytest.scm ------------------------------------------------

(define-module (mytest))
(export mytest)
(load-extension "libguile-mytest" "init_mytest")

---------- mylib.m (Mercury) -----------------------------------------

:- module mylib.

:- interface.
:- use_module io.
:- import_module list,bool.

:- type mytype ---> stuff(int,float,bool).

:- pred testproc(list(string),mytype,io.state,io.state).
:- mode testproc(in,out,di,uo) is det.

:- implementation.

:- pragma foreign_export("C",testproc(in,out,di,uo),"testproc_c").

testproc(_,X,!IO) :- X = stuff(0,0.0,no).

:- end_module mylib.

---------- mytest.h --------------------------------------------------

#ifndef MYDEF_H
#define MYDEF_H
#include <libguile.h>
SCM testproc_wrapper(SCM,SCM);
void init_mytest(void);
#endif /* MYDEF_H */

---------- mytest.c --------------------------------------------------

#include "mylibrary.h"
#include "mylib.mh"
#include "mytest.h"
#include <mercury.h>
#include <libguile.h>

SCM testproc_wrapper(SCM s,SCM x)
{
  MR_Word temp = MR_list_empty();
  MR_Word* p;

  testproc_c(temp,p); /* CALL TO MERCURY PROCEDURE */

  SCM_SET_SMOB_DATA(x,(scm_t_bits)p);

  return SCM_BOOL_T;
}

void init_mytest()
{
  void *dummy;
  char* args[1] = {(char*)""};

  mercury_init(1,args,&dummy);
  
  (void)scm_make_smob_type("mytype",0);
  
  scm_c_define_gsubr("mytest",2,0,0,testproc_wrapper);
}

---------- mylib.mh --------------------------------------------------

/*
** Automatically generated from `mylib.m'
** by the Mercury compiler,
** version rotd-2009-12-28, configured for i686-pc-linux-gnu.
** Do not edit.
*/
#ifndef MYLIB_MH
#define MYLIB_MH

#ifdef __cplusplus
extern "C" {
#endif

#ifdef MR_HIGHLEVEL_CODE
#include "mercury.h"
#else
  #ifndef MERCURY_HDR_EXCLUDE_IMP_H
  #include "mercury_imp.h"
  #endif
#endif
#ifdef MR_DEEP_PROFILING
#include "mercury_deep_profiling.h"
#endif

#ifndef MYLIB_DECL_GUARD
#define MYLIB_DECL_GUARD

#endif
void testproc_c(MR_Word, MR_Word *);

#ifdef __cplusplus
}
#endif

#endif /* MYLIB_MH */

---------- Makefile --------------------------------------------------

nothing: lib

CC = gcc
LD = gcc
MMC = mmc
MMAKE = mmake

GUILE_CFLAGS=$(shell guile-config compile)
GUILE_LIBS=$(shell guile-config link)

MERCURY_CFLAGS=$(shell $(MMC) --output-cflags)
MERCURY_LIBS=-L/usr/lib -L/usr/local/mercury-rotd-2009-12-28/lib/mercury/lib \
 -L/usr/local/mercury-rotd-2009-12-28/lib/mercury/lib/asm_fast.gc \
 -Wl,-rpath,/usr/local/mercury-rotd-2009-12-28/lib/mercury/lib \
 -Wl,-rpath,/usr/local/mercury-rotd-2009-12-28/lib/mercury/lib/asm_fast.gc \
 -lmer_std -lmer_rt -lm -lgc

LIBS = $(MERCURY_LIBS) $(GUILE_LIBS)

mylibrary.o : mylib.m
	$(MMC) --generate-standalone-interface mylibrary mylib.m

mytest.o : mytest.c
	$(CC) -c $(GUILE_CFLAGS) $(MERCURY_CFLAGS) mytest.c

mylib.mh : mylib.m
	$(MMC) mylib.m

lib : mytest.o mylibrary.o
	$(LD) -shared -fPIC mylibrary.o mytest.o -o libguile-mytest.so $(LIBS)

install : mytest.scm libguile-mytest.so
	sudo cp mytest.scm /usr/share/guile/1.8/
	sudo cp libguile-mytest.so /usr/lib/

---------- without the call to testproc_c ----------------------------

$ guile
guile> (use-modules (mytest))
guile>

---------- with call to testproc_c -----------------------------------

$ guile
guile> (use-modules (mytest))

Backtrace:
In unknown file:
    ...
   ?: 13  (begin (if # #) (make-modules-in # full-name))
   ?: 14* (if (or # #) (try-load-module name))
   ?: 15  [try-load-module (mytest)]
   ?: 16  (or (begin (try-module-linked name)) (try-module-autoload name) ...)
   ?: 17* [try-module-autoload (mytest)]
   ?: 18  (let* (# # # #) (resolve-module dir-hint-module-name #f) (and # #))
    ...
   ?: 19  (letrec ((load-file #)) (dynamic-wind (lambda () #) (lambda () #)
...) ...)
   ?: 20* [dynamic-wind #<procedure #f ()> #<procedure #f ()> #<procedure #f ()>]
   ?: 21* [#<procedure #f ()>]
   ?: 22* (let* ((file #)) (cond (# => #) (# => #)))
   ?: 23  [#<procedure #f (full)> "/usr/share/guile/1.8/mytest.scm"]
   ?: 24  [with-fluid* #<fluid 7> #f #<procedure #f ()>]
   ?: 25* [#<procedure #f ()>]
   ?: 26* [load-file #<primitive-procedure primitive-load> ...]
   ?: 27* [save-module-excursion #<procedure #f ()>]
   ?: 28  (let (# #) (dynamic-wind # thunk #))
   ?: 29  [dynamic-wind #<procedure #f ()> #<procedure #f ()> #<procedure #f ()>]
   ?: 30* [#<procedure #f ()>]
   ?: 31* [primitive-load "/usr/share/guile/1.8/mytest.scm"]
In /usr/share/guile/1.8/mytest.scm:
   3: 32* [load-extension "libguile-mytest" "init_mytest"]

/usr/share/guile/1.8/mytest.scm:3:1: In procedure dynamic-link in expression
(load-extension "libguile-mytest" "init_mytest"):
/usr/share/guile/1.8/mytest.scm:3:1: file: "libguile-mytest", message: "file
not found"
ABORT: (misc-error)
guile>

----------------------------------------------------------------------

Re: hackery (was: dynamic linking)

[Linas Vepstas]

On 15 February 2010 11:43, Tomas By <tomas.by@fcsh.unl.pt> wrote:
> Hello again everybody,

Just a wild guess -- i'm focusing on the "file not found" aspect of the
error message:

what does ldd libguile-mytest.so show?  Are all dependencies resolved?

If you compile the following, will it run?

#include "mylibrary.h"
#include "mylib.mh"
#include "mytest.h"
#include <mercury.h>

main()
{
 MR_Word temp = MR_list_empty();
 MR_Word* p;

 testproc_c(temp,p);
return 0;
}

i.e. is this enough to compile it:

cc main.c  -lguile-mytest.so -L .

or do you need something more?

--linas

disclaimer: I am a guile novice

Re: hackery (was: dynamic linking)

[Tomas By]

On Tue, February 16, 2010 16:20, Linas Vepstas wrote:
> what does ldd libguile-mytest.so show?  Are all dependencies resolved?

I think so.

> If you compile the following, will it run? [...]
> i.e. is this enough to compile it: [...]

Mercury needs to be initialized, and the C compiler needs to know about
the locations of the Mercury .h/.o files, but I compiled it and got a
Mercury error, manifesting itself as a segmentation fault.

So it looks like it is a C/Mercury interface problem, not Guile.

(thanks anyway)

/Tomas