Numeric C extension: bug or error ?

[Philippe Grosse]

In article <qnku16u3t26.fsf at arbutus.physics.mcmaster.ca>,
cookedm+news at physics.mcmaster.ca (David M. Cooke) wrote:

> At some point, veronique.grosse at wanadoo.fr (Philippe Grosse) wrote:
> > Hi,
> >
> > I don't understand this strange behaviour:
> >
> > I compile this code :
> [code clipped; it looks good]
> ...
> >    /* registration table  */
> >    static struct PyMethodDef testMethods[] = {
> >       {"return_vector", return_vector, 1},       /* method name, C
> >                                                     funcptr, always-tuple */
> >       {NULL, NULL}                   /* end of table marker */
> >    };
> 
> Instead of the magic constant "1", use METH_VARARGS.

I have tried METH_VARARGS: same result

> 
> ...
> > Very simple: this module takes a Numeric array (vector) as argument and
> > send this array back to python...
> >
> > If I compile it under macOSX, the result in python is:
> >
> >>>> import test
> >>>> from Numeric import *
> >>>> v=array([1.0,2.0,3.0,4.0],Float)
> >>>> v
> > array([ 1.,  2.,  3.,  4.])
> >>>> test.return_vector(v)
> > array([ 1.,  2.,  3.,  4.])
> >
> > but in linux the result is:
> >
> >>>> import test
> >>>> from Numeric import *
> >>>> v=array([1.0,2.0,3.0,4.0],Float)
> >>>> v
> > array([ 1.,  2.,  3.,  4.])
> >>>> test.return_vector(v)
> > array([ 1.+0.j,  2.+0.j,  3.+0.j,  4.+0.j])
> 
> This isn't what I get; I get the same result under linux as you get
> from Mac OS X. Have you checked that the copies of the code are the
> same?

Yes, I'm sure, the codes are exactly the same.

I use this command to compile under linux:

[grosse at intercalcul dev_fortran]$ gcc -DNDEBUG -O3 -fomit-frame-pointer
-pipe -mcpu=pentiumpro -march=i586 -ffast-math -fno-strength-reduce -fPIC
-g -I/usr/include/python2.2 -c test.c -o test.o

[grosse at intercalcul dev_fortran]$ gcc -shared -g -o test.so test.o

I can understand where is the problem, I'm totaly confused.

Thank you for your help