[SciPy-dev] Inclusion of cython code in scipy

[Brian Granger]

On Thu, Apr 24, 2008 at 10:38 AM, Prabhu Ramachandran
<prabhu@aero.iitb.ac.in> wrote:
> Brian Granger wrote:
>  >>  > C++ STL containers are truly great for things like this.  I would
>  >>  > write a simple c++ header file that defines the Particle class, create
>  >>  > an std::vector<Particle> to hold them and wrap the whole thing into
>  >>  > Cython.  I have already done stuff like this (templated c++ with STL)
>  >>  > with cython and it works great.  Furthermore, you end up with an
>  >>  > actual C/Python extension type that 1) is super fast 2) has its own C
>  >>  > api that can be called from other C extensions.
>
>  For some strange reason I still haven't gotten this email but Brian,
>  that is the point I was making, STL has all the needed container classes
>  for this and this is precisely what I use.  I just use SWIG to wrap this
>  C++ code and it works great.  Much of the STL is already wrapped via
>  SWIG and with numpy.i it interfaces quite well with numpy, in addition
>  you can inject doxygen generated documentation almost automatically so
>  your generated wrapper code is fully documented.
>
>  I've also explored using D and PyD and it makes for a fantastic
>  combination.  D seems to be (gdc) about 1.7 times slower than C++ but is
>  really nice to work with.  PyD's support for numpy is lacking but could
>  be fixed.
>
>  The point is (and what I understand of what Nathan was saying), pyrex
>  and cython let you wrap the code to an extent but do not provide the
>  underlying solution.  People were arging that Cython may be used in
>  place of C, and I think our point is that it isn't there yet and for
>  people used to C or C++ it is much easier to just use those instead.

I do think that in many cases, Cython _can_ be used quite well instead
of C/C++.  But, for the example you gave, the C++ STL is a great tool
and probably the tool I would use if I really needed the performance.
But, for someone that doesn't know or want to deal with C/C++, Cython
does provide a nice way of writing the hypothetical Particle class as
a C/Python extension type.  Here is the Cython code:

cdef class Particle:
  double public x, y, z
  double public px, py, pz
  double public mass, charge

The resulting extension type would be nearly as fast as the
corresponding c++ class (I will test this), and is already wrapped
into Python.  Of course, the user would need to store these objects in
a Python list rather than a c++ std::vector, but depending on what
they are doing, it could be totally fine.  For a user who would
otherwise write this class in pure Python, Cython represents a
_massive_ improvement.  One cool possibility would be to make an
stl::vector of these python extension types, which I think Cython
would allow.

>
>  > Unfortunately, the context I have done this in is proprietary code for
>  > my employer.  But, I will try to put together a simple example this
>  > weekend and post it to launchpad.  The documentation on these
>  > capabilities of Cython isn't great right now for sure, so some example
>  > code would sure help.
>
>  So why don't we do this together.  I'll put together my tiny example of
>  a shootout with Python C++ and D and lets see how they stack up.  If you
>  would be so kind you can add the cython/pyrex parts to this.  Thanks.
>
>  cheers,
>  prabhu
>
>
> _______________________________________________
>  Scipy-dev mailing list
>  Scipy-dev@scipy.org
>  http://projects.scipy.org/mailman/listinfo/scipy-dev
>