Re: [Swig-devel] 'talby' perl problems, perl5 diffs

[David F. <fr...@tu...>]

Hi William,

>>>>> "wf" == William Fulton writes:

>> Sometime later today, I'll try to copy a snapshot of the libkml
>> interface project I'm working on to the swig tracker.... presuming I
>> can figure out how to use it, of course.  :-)

wf> See my previous email re the SWIG testing framework needs to be
wf> used and not libkml.

Yes, I'll be developing a swig-specific test case, too.  But, I also
needed to develop a libkml-specific test case, and that's what I chose
to work on first (because it was easier to morph an existing test
case, and I didn't know about the swig test case when I started).

>> I'm now in the process of testing for memory leaks using valgrind
>> and I see a few difficulties...  In other words, I'm pretty sure
>> that the changes I've made aren't related.

wf> Tracking down memory leaks in these SWIG wrapped smart pointers is not 
wf> easy... I've had to do this for shared_ptr. I put in a wrapper around 
wf> shared_ptr in the end to reference count it all, you could consider 
wf> doing something similar, see the li_boost_shared_ptr.i testcase.

Yep, that's a good way to do this.  Fortunately, intrusive_ptr *is* a
reference count with just a tiny bit of extra machinery (though,
technically the reference count is kept in the core object, not the
intrusive_ptr object).  So, it's pretty easy to see how many
references there are.  What's harder is finding out, from the various
extension languages, how many references it has to the intrusive_ptr
object (or any object).

I think what's happening in this situation is that perl is not gaining
ownership of the pointer for some reason.  I haven't looked into this
fully, but I think the SWIG_POINTER_OWN flag isn't being honored.  If
that is, indeed, the case, this will take more work to solve, and
require a more intrusive solution... which means that I should move
back to the 'talby' branch if, indeed, this is about ready to be
adopted.  Hmmm....

Tx.

--
David Fletcher				Tuscany Design Automation, Inc.
fr...@tu...			3030 S. College Ave, Suite 102
        				Ft. Collins, CO 80525 USA