Reputation: 1897
I'm having some trouble coupling C++ (98) with python 3. I have some base classes in C++ which I'd like to extend in Python. Some methods in question are pure virtual on the C++ side and will thus be implemented on the Python side.
Currently, I can call the abstract methods from C++ and, over swig, the specialization gets called in Python. Cool. I'm having trouble handing over parameters to Python..
Minimal complete example to simplify my problem:
// iBase.h
#pragma once
#include <memory>
typedef enum EMyEnumeration{
EMyEnumeration_Zero,
EMyEnumeration_One,
EMyEnumeration_Two
}TEMyEnumeration;
class FooBase{
protected:
int a;
public:
virtual int getA() = 0 ;
};
class Foo : public FooBase{
public:
Foo() {a = 2;}
int getA(){return a;}
};
class iBase{
public:
virtual void start() =0;
virtual void run(std::shared_ptr<FooBase> p, TEMyEnumeration enumCode) = 0;
};
On the swig side:
// myif.i
%module(directors="1") DllWrapper
%{
#include <iostream>
#include "iBase.h"
%}
%include <std_shared_ptr.i>
%shared_ptr(FooBase)
%shared_ptr(Foo)
%feature("director") FooBase;
%feature("director") iBase;
%include "iBase.h"
Run swig as:
swig -c++ -python myif.i
swig -Wall -c++ -python -external-runtime runtime.h
Compile myif_wrap.cxx -> _DllWrapper.pyd
Create an *.exe with the following code, it will load up the _DllWrapper.pyd library (make sure it's in the same directory!). Also, copy DllWrapper.py generated by swig to the exe directory.
//Main_SmartPtr.cpp
#include "stdafx.h"
#include <Python.h>
#include <windows.h>
#include <string>
#include <memory>
#include "iBase.h"
#include "runtime.h"
using namespace std;
int _tmain(int argc, _TCHAR* argv[])
{
string moduleName = "ExampleSmartPtr";
// load *.pyd (actually a dll file which implements PyInit__<swigWrapperName>)
auto handle =LoadLibrary("_DllWrapper.pyd");
// getting an instance handle..
Py_Initialize();
PyObject *main = PyImport_AddModule("__main__");
PyObject *dict = PyModule_GetDict(main);
PyObject *module = PyImport_Import(PyString_FromString(moduleName.c_str()));
PyModule_AddObject(main, moduleName.c_str(), module);
PyObject *instance = PyRun_String(string(moduleName+string(".")+moduleName+string("()")).c_str(), Py_eval_input, dict, dict);
//calling start() in the Python derived class..
//PyObject *result = PyObject_CallMethod(instance, "start", (char *)"()");
// trying to call run in the Python derived class..
shared_ptr<Foo> foo = make_shared<Foo>();
EMyEnumeration enumCode = EMyEnumeration_Two;
string typeName1 = "std::shared_ptr <FooBase> *";
swig_type_info* info1 = SWIG_TypeQuery(typeName1.c_str());
auto swigData1 = SWIG_NewPointerObj((void*)(&foo), info1, SWIG_POINTER_OWN);
string typeName2 = "TEMyEnumeration *";
swig_type_info* info2 = SWIG_TypeQuery(typeName2.c_str());
auto swigData2 = SWIG_NewPointerObj((void*)(&enumCode), info2, SWIG_POINTER_OWN);
auto result = PyObject_CallMethod(instance, "run", (char *)"(O)(O)", swigData1, swigData2);
return 0;
}
Create a new Python file and put it in the exe's directory:
#ExampleSmartPtr.py
import DllWrapper
class ExampleSmartPtr(DllWrapper.iBase):
def __init__(self): # constructor
print("__init__!!")
DllWrapper.iBase.__init__(self)
def start(self):
print("start")
return 0
def run(self, data, enumCode):
print("run")
print("-> data: "+str(data))
print("-> enumCode: "+str(enumCode))
print (data.getA())
return 1
The output of running the exe is :
__init__!!
run
-> data: (<DllWrapper.FooBase; proxy of <Swig Object of type 'std::shared_ptr< FooBase > *' at 0x00000000014F8B70> >,)
-> enumCode: (<Swig Object of type 'TEMyEnumeration *' at 0x00000000014F89F0>,)
How can one 'dereference' the enumCode to a simple int? How does one call print (data.getA()) in python class run()? In its current form it doesn't print anything..
Upvotes: 0
Views: 1211
Reputation: 1897
It seems somebody else tried the exact same thing !
What I did was compile the *.pyd with -DSWIG_TYPE_TABLE=iBase.
Then I added this to the main application on the cpp side:
iBase *python2interface(PyObject *obj) {
void *argp1 = 0;
swig_type_info * pTypeInfo = SWIG_TypeQuery("iBase *");
const int res = SWIG_ConvertPtr(obj, &argp1,pTypeInfo, 0);
if (!SWIG_IsOK(res)) {
abort();
}
return reinterpret_cast<iBase*>(argp1);
}
and called the implementation form python like this:
auto foo = make_shared<Foo>();
TEMyEnumeration enumCode = EMyEnumeration_Two;
python2interface(instance)->run(foo, enumCode);
To finish off, I compiled the C++ implementation again with -DSWIG_TYPE_TABLE=iBase.
Works like a charm!
Upvotes: 0
Reputation: 4725
This is not really an answer, but I read the discussion Discussion from 2005 and it makes sense, that it shouldn't be possible. If you on the Python side, do the following, you get your enumeration 'dereferenced' to a simple integer.
import ExampleSmartPtr
instance = ExampleSmartPtr.ExampleSmartPtr()
swigData1 = ExampleSmartPtr.DllWrapper.Foo()
swigData2 = ExampleSmartPtr.DllWrapper.EMyEnumeration_Two
instance.run(swigData1,swigData2)
This will print
__init__!!
run
-> data: <DllWrapper.Foo; proxy of <Swig Object of type 'std::shared_ptr< Foo > *' at 0x7f8825c0b7e0> >
-> enumCode: 2
I think that the issue is that two different vtables are into play. The original C++ vtable and that of the Swig Object
. Just curious, in what scenario is it of interest to use a Python descendant of a C++ class from within C++?
Upvotes: 1