[Xapian-discuss] Python segmentation fault when abstracting value range processor

[Olly Betts]

On Wed, Mar 04, 2009 at 11:05:33AM +0000, James Aylett wrote:
> ISTR that we have a way of tackling this kind of
> problem, but I could be misremembering. If no one else pipes up, I'll
> add it to trac.

There's already a ticket covering this:

http://trac.xapian.org/ticket/186

Essentially our reference counting mechanism doesn't work for
user-subclassable classes.  Moving to something from boost would
probably address this, as we could more sanely expose the reference
counting mechanism, but is quite a major incompatible change.

Actually, another option might be to keep a reference to the Python VRP
proxy object in the Python QueryParser proxy object.  I don't know how
feasible that is.

Perhaps it would be better to deliberately leak such objects for now
- that's what Search::Xapian does at the moment.

Cheers,
    Olly