SWIG and Typesystems

<h2>SWIG and Typesystems</h2>

<p>
For the past seven years, a considerable amount of effort has gone
into the development of SWIG and its support for various programming
languages.  Currently, more than a <a href="guilty.html">dozen developers</a> work on the system
and there are thousands of users.  However, almost all of SWIG's development
has been relatively ad-hoc---primarily driven by feature requests from
users.  As SWIG's original author, coordinating all of this chaos has
always been a bit of challenge.  In fact, it's been rather difficult to even
describe what SWIG "is" and "how it works" without using terms like
"magic" and, well, "more magic."   Needless to say, this isn't the most
academic way to look it ;-).

<p>
A little over a year ago, I became interested in the relationship
between SWIG and work in the area of type systems.  In order
to support advanced C++ features like namespaces and templates, a
significant amount of very difficult development work focused on the
SWIG type system.  Moreover, this work resulted in a lot of old SWIG
features being folded into type system extensions.  Because of this
work, I have now come to view SWIG as being mostly driven as an
extension of the C++ type system rather than an extension of a C++
parser.  This difference is subtle, but it is the only way to really
understand how SWIG works at a fundamental level.

<p>
To the best of my knowledge, no one working on automatic C/C++ wrapper
generators has really approached the wrapping problem from the
standpoint of type systems.  Instead, everyone seems to focus on the problem
of <em>parsing</em> C++, which, although important, is not enough to
really understand what is going on.  

<p>
Starting in the summer of 2003, the NSF will be funding a 
<a href="http://www.fastlane.nsf.gov/servlet/showaward?award=0237835">research
project</a> at the University of Chicago to explore the relationship
between type-systems and wrapper generation in more detail.  A lot of
this work will involve SWIG and the semantics of its underlying type
system.  Moreover, this work will be exploring some exciting new
features such as software contracts and improved mixed-language
debugging support.  We're also going to try and make connections
between SWIG and related work in the area of programming languages.

<p>
What does this mean for the future of SWIG?  Well, my hope is that
this work will make the system more capable, more reliable, and more
useful than ever.  I think everyone will be pleasantly surprised with
some of the changes.

<p>
As always, I'd like to acknowledge everyone who has contributed to SWIG
over the years---thank you for your support!

<p>
Cheers,
<p>
Dave Beazley <br>
March 23, 2003