Why SWIG has been selected as C++ wrapper for Python for GNURadio?

Hi All

This is a question intended for the Software designers and maintainers
of
the GNURadio

Is there a specific reason / advantage of using SWIG as a C++ wrapper
over
using other avaialble wrappers for C++ as Boost.Python , Cython etc

in the context of GNURadio ?

Thanks!
Joe

On 06/20/2014 10:34 AM, Joe D wrote:

This is a question intended for the Software designers and maintainers
of the GNURadio

Is there a specific reason / advantage of using SWIG as a C++ wrapper
over using other avaialble wrappers for C++ as Boost.Python , Cython
etc … in the context of GNURadio ?

First, you need to realize that the decision to use SWIG was made a long
way back, when there were less alternatives. At the time, one important
argument was that Python wasn’t the only scripting language that was to
be supported.
At this point, we know a lot of SWIG’s quirks, and there is little
reason to change. It does all we need (although it could do somethings a
bit faster, the build times are still annoying), and it integrates well
with our code structure.
To be honest: Few of us are 100% happy with SWIG, but it’s the devil we
know :slight_smile:

However, while there are alternatives to SWIG worth considering, Cython
is not suitable for what we need. I haven’t looked into Boost.Python
closely, though.

M

Hi Joe,

Disclaimer: I haven’t been around when SWIG was introduced into GNU
Radio. These are somewhat educated guesses, but since I’ve had some
fun with SWIG, I wondered myself and this is what I rationalized.

  • SWIG was around at the time GNU Radio was rewritten, which must have
    been around 2004. I think the choice to use SWIG was made before that
    [1]. SWIG has been around since '96, so I guess it was expected to be
    somewhat stable/reliable. Boost.Python was started in '03, and if the
    guys behind it decided that they had to conceptually rework something,
    the second year it exists sounds like a dangerous time to use it. ctypes
    is (I think) python-2.5esque, so this couldn’t have been around back in
    the day, and only does basic C wrapping.
  • It does what GNU Radio needed: Generate python wrapper classes without
    the developers needing to actually write code; defining the interface
    was enough for a “normal” block to be SWIG’ed, and specialized wrapper
    classes only had to be written for some of the core things so that
    things like connect()ing blocks would work. It is arguable if writing
    and maintaining the SWIG .i files (especially since we’re talking about
    the autohell era) was worth the trouble of avoiding that, but for the
    only comparable competitor Boost.Python, you had to define wrapper
    classes, and references to the classes wrapped and their members.
    Pyrex/Cython now is a complete different language and would have made
    GNU Radio something much more multi-paradigmatic, whilst adding even
    more to the “active” codebase that had to be maintained by hand.
  • It was expected that if SWIG wasdone right, GNU Radio could end up
    with multiple scripting language interfaces [2]. Nowadays, people are
    content to use python, especially since we now (feels like yesterday)
    have support to write gr::block classes in python, which of course
    needed a lot of gateway work, so (to my knowledge) there haven’t been
    efforts to make use of more than the C++ and Python SWIG
    capabilities[3].

Greetings,
Marcus

[1]
http://lists.gnu.org/archive/cgi-bin/namazu.cgi?query=SWIG&submit=Search!&idxname=discuss-gnuradio&max=20&result=normal&sort=date%3Aearly
[2]
http://lists.gnu.org/archive/html/discuss-gnuradio/2003-03/msg00135.html
[3] After all, for scripting python works really nice and is easy to
learn. If you wanted to use a fast Matrix operation in Fortran, there is
no stopping you from doing that from C++, which is the programming
language of choice for performance-critical blocks, anyway, as long as
your Fortran and C++ compilers are ABI-compliant. If you wanted


Discuss-gnuradio mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio

On Fri, Jun 20, 2014 at 5:42 AM, Marcus Müller
[email protected]
wrote:

decided that they had to conceptually rework something, the second year it
competitor Boost.Python, you had to define wrapper classes, and references

choice for performance-critical blocks, anyway, as long as your Fortran and
in the context of GNURadio ?

Thanks!
Joe

Martin and Marcus nicely answered the question for you.

Just wanted to say that while we’re not entirely impressed with some
features of SWIG, the fact that it’s working well for us and the
disruption
we would cause switching it out is not worth it. We’d have to make sure
that the changes would result in a perfect 1-to-1 match with what we’re
exporting using swig.

Tom


Discuss-gnuradio mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio