RemoteSearchable not implemented... why?

Howdy.  I recently came across a situation where it would
have been
useful to have the RemoteSearchable API available, but I see
in the
README (ht
tp://svn.osafoundation.org/pylucene/trunk/README)
that's
it's the one thing not yet implemented.  I was thinking
about taking a
shot at writing a simple Python XMLRPC implementation of
Searchable to
fill the same need, and I thought I'd ask why
RemoteSearchable wasn't
implemented in case there's any relevant pitfalls.
_______________________________________________
pylucene-dev mailing list
pylucene-devosafoundation.org
http://lists.osafoundation.org/mailman/listinfo/pylu
cene-dev

On Mon, 20 Mar 2006, John Reese wrote:

> Howdy.  I recently came across a situation where it
would have been
> useful to have the RemoteSearchable API available, but
I see in the
> README (ht
tp://svn.osafoundation.org/pylucene/trunk/README)
that's
> it's the one thing not yet implemented.  I was
thinking about taking a
> shot at writing a simple Python XMLRPC implementation
of Searchable to
> fill the same need, and I thought I'd ask why
RemoteSearchable wasn't
> implemented in case there's any relevant pitfalls.

In order to implement RemoteSearchable, you have to figure
out how to 
serialize and unserialize python SWIG-wrapped java objects.
I'm in the process 
of removing SWIG from the PyLucene picture so this could
become simpler...

Why didn't I implement RemoteSearchable ? The amount of
work involved wasn't 
worth it. There was interest in this earlier (about a year
ago, see the list's 
archives) but it seems that the effort was fruitless.

Andi..
_______________________________________________
pylucene-dev mailing list
pylucene-devosafoundation.org
http://lists.osafoundation.org/mailman/listinfo/pylu
cene-dev

Andi Vajda [vajdaosafoundation.org] said:

> >Howdy.  I recently came across a situation where it
would have been
> >useful to have the RemoteSearchable API available,
but I see in the
> >README (ht
tp://svn.osafoundation.org/pylucene/trunk/README)
that's
> >it's the one thing not yet implemented.  I was
thinking about taking a
> >shot at writing a simple Python XMLRPC
implementation of Searchable to
> >fill the same need, and I thought I'd ask why
RemoteSearchable wasn't
> >implemented in case there's any relevant pitfalls.
> 
> In order to implement RemoteSearchable, you have to
figure out how to 
> serialize and unserialize python SWIG-wrapped java
objects. I'm in the 
> process of removing SWIG from the PyLucene picture so
this could become 
> simpler...

Why are you removing SWIG from the picture?  I haven't seen
anything come across my subversion feeds.

-matthew
_______________________________________________
pylucene-dev mailing list
pylucene-devosafoundation.org
http://lists.osafoundation.org/mailman/listinfo/pylu
cene-dev

On Tue, 21 Mar 2006, Matthew O'Connor wrote:

> Why are you removing SWIG from the picture?  I haven't
seen
> anything come across my subversion feeds.

Because the code generated by SWIG is needlessly thick and
slow. The 5,000 
line PyLucene.i file translates into almost 60,000 lines of
C++ and 2,000 
lines of Python code. *None* of that generated Python code
is actually needed.
All of that can be done in C++ directly, leaving out one
layer of python calls 
altogether.

For example, python -m timeit "from PyLucene import
Object" "Object()"
runs in 12.4 usec per loop with the new code vs 33.2 usec
with SWIG-generated 
code.

Also, each and every release of SWIG breaks PyLucene and
I've had to fix it 
once too many.

I've become a lot more familiar with the Python C API over
the past year 
and SWIG's help is really no longer needed.

You haven't seen anything in subversion yet because I
haven't checked in 
anything so far. I spent a few days on this so far and have
a few more to go.

Andi..
_______________________________________________
pylucene-dev mailing list
pylucene-devosafoundation.org
http://lists.osafoundation.org/mailman/listinfo/pylu
cene-dev