Strange SWIG behaviour

ruby-talk
1

Hi,

   I'm doing some experiments with SWIG (trying to make a Ruby interface for
MuSE low-level sound classes), and I'm having some problems.

   One of them is really weird: it seems that there are problems with extern
"C" declarations (from C++) and with functions _named_ "error" (namely, that
when you try to call the function, the interpreter aborts, without exceptions
or anything).

   Here is my tiny example:

-- errortest.c / errortest.cxx ------ 8< -------------------------------------
#include <stdio.h>

void error(const char *fmt, ...)
{
   printf("I won't be called\n");
}
------------------------------------- >8 -------------------------------------

-- errortest.i ---------------------- 8< -------------------------------------
%module errortest

%{
#include "errortest.h"
%}

#ifdef __cplusplus
extern "C" {
#endif

void error(const char *format, ...);

#ifdef __cplusplus
}
#endif
------------------------------------- >8 -------------------------------------

-- errortest.h ---------------------- 8< -------------------------------------
#ifdef __cplusplus
extern "C" {
#endif

void error(const char *format, ...);

#ifdef __cplusplus
}
#endif
------------------------------------- >8 -------------------------------------

-- Makefile ------------------------- 8< -------------------------------------
BASENAME = errortest
WRAPPER_BASENAME = $(BASENAME)_wrap
SWIG = swig
INTERFACE = $(BASENAME).i
CXXSRCS = errortest.cpp
SRCDIR = .
OBJS = $(CXXSRCS:.cpp=.o)
INCLUDES = -I.
LINK =
CPP = 1
GCC = $(if $(CPP),g++,gcc)
SWIG_OPTS = $(if $(CPP),-c++,)
SRC_EXT = $(if $(CPP),cxx,c)

all: $(BASENAME).so

$(BASENAME).so: $(BASENAME).o $(WRAPPER_BASENAME).o
  $(GCC) -shared $(OBJS) $(WRAPPER_BASENAME).o $(LINK) -o $@

$(BASENAME).o: $(SRCDIR)/$(BASENAME).$(SRC_EXT)
  $(GCC) -fpic -fPIC -DHAVE_CONFIG_H -c -shared $(INCLUDES) $<

$(WRAPPER_BASENAME).o: $(WRAPPER_BASENAME).$(SRC_EXT)
  $(GCC) -fpic -fPIC -DHAVE_CONFIG_H -c -shared $(INCLUDES) -I/usr/lib/ruby/1.8/i386-linux $(WRAPPER_BASENAME).$(SRC_EXT)

$(WRAPPER_BASENAME).$(SRC_EXT): $(INTERFACE)
  $(SWIG) $(SWIG_OPTS) -ruby $(INTERFACE)

clean:
  $(RM) $(BASENAME).so $(BASENAME).o $(WRAPPER_BASENAME).o $(WRAPPER_BASENAME).$(SRC_EXT)

.PHONY: all clean
------------------------------------- >8 -------------------------------------

When I try to use it:

------------------------------------- 8< -------------------------------------
zoso@velutha:~/tmp/swig$ ruby -rerrortest -e "puts Errortest::error('some error')"
ruby:
zoso@velutha:~/tmp/swig$
------------------------------------- >8 -------------------------------------

When I rename the "error" function to "error2", it works great :-?

   Also, if I remove the extern "C" declarations (from both errortest.{h,i}, I
don't know if it makes sense leaving it in errortest.i) but leave the function
as "error", it works, too (note that I tried both with C and C++, hence the
errortest.c{,xx} file and the $(CPP) variable in the Makefile).

   Can anyone shed some light on this?

···

--
Esteban Manchado Velázquez <zoso@foton.es> - http://www.foton.es
EuropeSwPatentFree - http://EuropeSwPatentFree.hispalinux.es

2

Esteban Manchado Velázquez wrote:

Hi,

   I'm doing some experiments with SWIG (trying to make a Ruby interface for
MuSE low-level sound classes), and I'm having some problems.

   One of them is really weird: it seems that there are problems with extern
"C" declarations (from C++) and with functions _named_ "error" (namely, that
when you try to call the function, the interpreter aborts, without exceptions
or anything).

I played around with this a little yesterday afternoon and saw exactly the problem you desribe on Gentoo Linux. I ran gcc -E looking for a global 'error' pulled in from some header file, but found nothing. I'm stumped.

You might try it without SWIG--just write your own version of _wrap_error() using the Ruby C API. I suspect it'll have the same problem, but if not, you'll know the problem is with SWIG.

Steve

3

Esteban Manchado Velázquez wrote:
>Hi,
>
> I'm doing some experiments with SWIG (trying to make a Ruby interface
> for
>MuSE low-level sound classes), and I'm having some problems.
>
> One of them is really weird: it seems that there are problems with
> extern
>"C" declarations (from C++) and with functions _named_ "error" (namely,
>that
>when you try to call the function, the interpreter aborts, without
>exceptions
>or anything).

I played around with this a little yesterday afternoon and saw exactly
the problem you desribe on Gentoo Linux. I ran gcc -E looking for a
global 'error' pulled in from some header file, but found nothing. I'm
stumped.

   Thanks for trying :slight_smile:

You might try it without SWIG--just write your own version of
_wrap_error() using the Ruby C API. I suspect it'll have the same
problem, but if not, you'll know the problem is with SWIG.

   OK, I'll try that. Thanks again!

···

On Tue, Jul 26, 2005 at 02:11:59AM +0900, Steven Jenkins wrote:

--
Esteban Manchado Velázquez <zoso@foton.es> - http://www.foton.es
EuropeSwPatentFree - http://EuropeSwPatentFree.hispalinux.es

4

The only conclusion I've reached is that a function named "error" can't be
declared in C (and I guess this goes for _every_ Ruby C extension), probably
because it's declared somewhere. The problem is, I don't know where and I
don't have the slightliest idea of where to find it (in the interpreter
code?).

   As I was told by a friend, it probably works in C++ (without the extern "C"
thing) because of C++ function name mangling.

   I guess I'll have to remove the extern "C" declarations from MuSE, to be
able to make the Ruby interface :-/

   Best regards, and thanks again for trying...

···

On Tue, Jul 26, 2005 at 02:11:59AM +0900, Steven Jenkins wrote:

Esteban Manchado Velázquez wrote:
>Hi,
>
> I'm doing some experiments with SWIG (trying to make a Ruby interface
> for
>MuSE low-level sound classes), and I'm having some problems.
>
> One of them is really weird: it seems that there are problems with
> extern
>"C" declarations (from C++) and with functions _named_ "error" (namely,
>that
>when you try to call the function, the interpreter aborts, without
>exceptions
>or anything).

I played around with this a little yesterday afternoon and saw exactly
the problem you desribe on Gentoo Linux. I ran gcc -E looking for a
global 'error' pulled in from some header file, but found nothing. I'm
stumped.

You might try it without SWIG--just write your own version of
_wrap_error() using the Ruby C API. I suspect it'll have the same
problem, but if not, you'll know the problem is with SWIG.

--
Esteban Manchado Velázquez <zoso@foton.es> - http://www.foton.es
EuropeSwPatentFree - http://EuropeSwPatentFree.hispalinux.es