Setepenre
Setepenre

Reputation: 2053

Python function as Argument in C++

I am trying to expose eigen3 in python using Boost.Python.

I cannot find a way to expose the function unaryExpr (const CustomUnaryOp &func=CustomUnaryOp())

What I would like to have is something that allow me to something like that:

python

import libMatrix as mat
a = mat.Matrix(10, 10)
mat.unary_expr( lambda x : 1)

Do you have any idea ?? It may look like that:

void unary_expr(Matrix const& self, PyObject* callable_object)
{
   cpp_callable = ??(callable_object)
   self.unaryEpxr(cpp_callable);
}

=== What I tried: ==========================================

1) I tried to use a simple callback definition

typedef double(*UnaryExprType)(double);
void unary_expr(Matrix const& self, UnaryExprType a);
    {
       self.unaryEpxr( a );
    }

but boost does not convert the python function into a UnaryExprType.

2) I tried to implement a struct PythonCallBack, Nevertheless, it does not work, I have got an error that the python signature did not match the c++ signature.

struct PythonCallBackBase
{
public:
    virtual ~PythonCallBackBase()   {}
    virtual double operator() (double const & x) {  return 0;   }
};

struct PythonCallBack : PythonCallBackBase, boost::python::wrapper<PythonCallBackBase>
{
public:
    typedef boost::python::wrapper<PythonCallBackBase> wrap;

    double default_op(double const & x)
    {
        return 0;
    }

    double operator() (double const & x)
    {
        if (boost::python::override f = wrap::get_override("__call__"))
            return f(x);
        return PythonCallBackBase::operator ()(x);
    }
};

void unary_expr(Matrix const& self, PythonCallBack a)
{
   self.unaryEpxr( a );
}

Error message

ArgumentError: Python argument types in
    Matrix.unary_expr(Matrix, Boost.Python.class)
did not match C++ signature:
    unary_expr(Eigen::Matrix<double, -1, -1, 0, -1, -1>, PythonCallBack)
    unary_expr(Eigen::Matrix<double, -1, -1, 0, -1, -1>, double (*)(double))

Upvotes: 0

Views: 203

Answers (1)

Tanner Sansbury
Tanner Sansbury

Reputation: 51971

Boost.Python is designed to minimize the need to interact with PyObject, and one can often simple use boost::python::object in the same manner they would with an object in Python. For instance, if func is a boost::python::object that refers to a lambda x: 1, then here is the following Boost.Python usage with annotated Python comments:

// >>> func = lambda x: 1
boost::python::object func = ...;
// >>> result = func(42)
boost::python::object result = func(42);
// >>> assert(1 == result)
assert(1 == boost::python::extract<int>(result));

In this case, as the C++ code is likely expecting for the functor's return value to be a C++ type rather than the generic boost::python::object, one can use a wrapper type to adapt the functor.

/// @brief Auxiliary type that adapts a Boost.Python object to a
///        unary functor with an explicit return type.
template <typename Arg, typename Result>
class py_unary_functor
{
public:

  typedef Arg argument_type;
  typedef Result result_type;

  py_unary_functor(boost::python::object object)
    : object_(object)
  {}

  result_type operator()(argument_type a1)
  {
    return boost::python::extract<result_type>(object_(a1))();
  }

private:
  boost::python::object object_;
};

Here is a complete minimal example:

#include <boost/python.hpp>

/// @brief Mockup matrix class.
struct matrix
{
  template <typename CustomUnaryOp>
  void unaryExpr(CustomUnaryOp func)
  {
    value = func(value);
  }

  double value;
};

/// @brief Auxiliary type that adapts a Boost.Python object to a
///        unary functor with an explicit return type.
template <typename Arg, typename Result>
class py_unary_functor
{
public:

  typedef Arg argument_type;
  typedef Result result_type;

  py_unary_functor(boost::python::object object)
    : object_(object)
  {}

  result_type operator()(argument_type a1)
  {
    return boost::python::extract<result_type>(object_(a1))();
  }

private:
  boost::python::object object_;
};

/// @brief Auxiliary function used to adapt matrix::unaryExpr.
void matrix_unaryExpr(
  matrix& self,
  boost::python::object object)
{
  py_unary_functor<double, double> func(object);
  return self.unaryExpr(func);
}

BOOST_PYTHON_MODULE(example)
{
  namespace python = boost::python;
  python::class_<matrix>("Matrix")
    // Expose auxiliary function.
    .def("unaryExpr", &matrix_unaryExpr)
    .add_property("value", &matrix::value, &matrix::value)
    ;
}

Interactive usage:

>>> import example
>>> matrix = example.Matrix()
>>> matrix.value = 21
>>> assert(21 == matrix.value)
>>> matrix.unaryExpr(lambda x: x*2)
>>> assert(42 == matrix.value)

Upvotes: 1

Related Questions