[Swig-devel] [Patch] Feature to override Java's "swigCPtr" so "this pointer" null checks become possible before dereferencing

[Art C. <ac...@vl...>]

Hi,

First off, Swig rocks... this is merely a suggestion to make it
hopefully rock a little more in Java.  Please let me know your
thoughts

In the Java module of Swig (as of 1.3.37) the string "swigCPtr" is
assumed as the name of the variable that holds the native pointer to
the C Object, and this string is inserted directly in the Java proxy
class when calls are made requiring a "this" pointer.

This works great, but swig also auto-generates a delete() method that
will set swigCPtr to 0, and auto-generates C++ method calls that
essentially do (swigCPtr)->func(). If someone has called delete() in
the Java wrapper, and then a class (non static) method, they'll get a
core-dump as opposed to a more friendly error message.

That said, it would be inappropriate to assume that everyone wants to
do something more intelligent (and consequently less efficient) that
always assume swigCPtr is valid, so I propose the following change:

Allow people to override the string used to find the current native
pointer from the JNI proxy class, but default to "swigCPtr".

For example:

%pragma(java) jniclassthisptrname="mySafeGetCPtr()".

That way if no one overrides this "jniclassthisptrname" setting
they'll get today's efficient (but potentially unsafe) "swigCPtr". If
they do override it, they can intercept deferences of swigCPtr before
it causes the core-dump in native code and instead raise an exception.

The attached patch does exactly that for Revision 10894. It should be
100% backwards compatible.

The main reason I'd like this feature is, I admit it, to protect me
(as the maintainer of a swig-based library) from stupid mistakes by my
users.  Yes, they should NEVER use an object after they call delete(),
and yes, everytime I point that out to them they agree.  Still the
sight of their JVM core-dumping always brings them to me screaming,
and I'd like to instead return a nice Java exception so my users are a
little, well, calmer :)  Also, if there is a way to do this with Swig
without patching swig, please let me know (but I've tried all sorts of
typemaps, and keep coming back to the issue that I can't tell from a
typemap if I'm in a static or non-static method).

Thanks in advance,

- Art Clarke

p.s.
I also added this to the sourceforge tracker:
https://sourceforge.net/tracker2/?func=detail&aid=2198889&group_id=1645&atid=301645