[Swig-user] Wrapped type passing between two SWIG Python modules

[Andy B. <an...@in...>]

Hi,

I have a problem on one sort of target system, which probably results
from passing a mutually wrapped C++ type between two different
SWIG-generated Python modules. I say "probably", because all I see as an
error is a "TypeError: expected a pointer" printout, resulting from an
exception thown in the SWIG module. I'd appreciate if you can
collectively either solve the problem or just explain what's happening
more clearly than I can currently appreciate!

I don't want to burden you with app specific details, but creating
pseudonyms for classes would be more likely to generate typos than
leaving them unchanged, so please accept that the two SWIG-wrapped C++
classes are

 * HepMC::GenEvent -> hepmc.GenEvent
 * Rivet::AnalysisHandler -> rivet::AnalysisHandler

These are mapped in two completely distinct packages, i.e. the runs of
"swig -c++ -python ... " are totally independent. In the client code,
then, there is a routine "analyze" on a rivet.AnalysisHandler object,
which takes a hepmc.GenEvent object as its non-"self" argument. On the
32 bit Ubuntu Linux systems I've developed and tested on, this works
absolutely fine --- I can pass the hepmc object around between other
modules and everything is fine. On a Scientific Linux 64 bit system,
though, I get the above-mentioned TypeError when calling

  myAnalysisHandler.analyze(myGenEvent)

Digging a bit into the offending function's definition (in
rivet_wrap.cc), I find this in the source code:

  static PyObject *_wrap_AnalysisHandler_analyze(PyObject *self,
      PyObject *args) {
    ...
    if(!PyArg_ParseTuple(args,(
      char *)"OO:AnalysisHandler_analyze",&obj0,&obj1)) goto fail;
    if ((SWIG_ConvertPtr(obj0,(void **) &arg1,
      SWIGTYPE_p_Rivet__AnalysisHandler, SWIG_POINTER_EXCEPTION | 0 ))
      == -1) SWIG_fail;
    if ((SWIG_ConvertPtr(obj1,(void **) &arg2,
      SWIGTYPE_p_HepMC__GenEvent, SWIG_POINTER_EXCEPTION | 0 ))
      == -1) SWIG_fail;
    if (arg2 == NULL) {
      PyErr_SetString(PyExc_TypeError,"null reference"); SWIG_fail;
    }
    (arg1)->analyze((HepMC::GenEvent const &)*arg2);
    ...
  }

Since one of the SWIG_POINTER_EXCEPTION lines is probably responsible
for the Python TypeError, I was interested to find how the SWIGTYPE_...s
were defined. In rivet_wrap.cc, it was this:

  #define  SWIGTYPE_p_HepMC__GenEvent swig_types[14]

and in hepmc_wrap.cc:

  #define SWIGTYPE_p_HepMC__GenEvent swig_types[5]

I'm now unsure of how it *ever* worked (I admit, I was amazed when this
appeared to work automatically in the first place!), since the
swig_types indices are completely different. But then, I know nothing of
how the type mappings are actually used, and it's hopefully a bit
cleverer than just matching indices ;) Can anyone advise on what the
expected behaviour is, why the problem is appearing on one sort of
system (I'm not sure if it's a 32/64 bit issue), and how I can fix it?
Any of the above would be helpful, but the last one more than most!

Thanks,
Andy