[C++-sig] create boost array from a Numpy PyArray without copying data? (original) (raw)

Sebastian Walter walter at mathematik.hu-berlin.de
Wed Jan 21 09:40:00 CET 2009

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1

Ravi schrieb:

On Tuesday 20 January 2009 04:27:02 Sebastian Walter wrote:

Thanks for the hint :), Your code is a little to heavyweight for my purposes, but I learned quite a lot how boost::python converters work.

And it also gave me the idea that finally solved my problem! Good. // global to guarantee that the data is not freed after testfunctionthatdoesntcopy returns double x[] = {1.,2.,3.}; In this case, you run into the problem of lifetime management. Unless you use globals, how do you guarantee that once all users of the array are done with it, the array is deleted? The array must be deleted from either the python side (via numpy) or from the C++ side (by yourself). Handling this problem is the cause for much of my code (which may still be too heavyweight for you). Yes, doing it as I posted above caused a memory corruption. Apparently Python frees the memory of the Numpy array when it's not needed anymore in Python. Apparently I have to increase the refcount manually when I return a raw PyObject. So I wrote getter functions like that:

PyObject* get_rhs(){
    return incref((PyObject *) _array_my_rhs);
}

I'm not sure that this is the way it should be done. But it solved the memory corruption problem.

With heavyweight I meant that I don't want to use ublas nor do I want to write a converter. I simply don't have the time to check out the inner workings of boost::python. And writing a converter requires a fair amount of template meta programming + knowledge of boost::python.

What I'm trying to accomplish is the following: I have a DAE-solver which is implemented in C++. A DAE is basically an ODE + algebraic equations: \dot xd = f(t,xd,xa) 0 = g(t,xd,xa) where xd/xa are the differential resp. algebraic state variables.

In C++ I give the integrator a function pointer to a function: void f(double *t, double *x, double *rhs);

The integrator is responsible for the memory management of *t, *x, *rhs and just calls the function f. The function f should then convert the raw pointers to numpy arrays and then call the implementation of f which is in Python.

void f(double *t, double *x, double *rhs){

  1. create PyArray with PyArray_FromDimsAndData(1, &Nx, PyArray_DOUBLE, (char*) t);
  2. then call a callable python object with those PyArrays as argument. }

Regards, Ravi

Cplusplus-sig mailing list Cplusplus-sig at python.org http://mail.python.org/mailman/listinfo/cplusplus-sig

-----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.4-svn0 (GNU/Linux) Comment: Using GnuPG with SUSE - http://enigmail.mozdev.org

iD8DBQFJdt9g9PBA5IG0h0ARAlY4AJ0T0M2ljLLjegLYikBeRuDeaJelUgCdEF4q sORViEHb8BmzhNSyhdmNWZA= =eSRA -----END PGP SIGNATURE-----

More information about the Cplusplus-sig mailing list