Re: [Jython-users] elementtree performance

Roussi Roussev, 06.04.2010 23:41:
> I am experimenting with elementtree and jython 2.5.1 but its performance
> is far from lxml. What would it take to get the jython xml performance
> on-par with the C library?

Writing a Java XML parser that runs as fast as the C based parsers, I 
guess. Both cElementTree and lxml.etree quite easily beat the existing Java 
parsers in terms of performance - at least according to the last benchmarks 
I read, cross-platform benchmarks are really rare. Most of the time, the 
platform is fixed anyway, so there's little benefit from looking around.


> Has anyone tried to optimize it? Any
> fundamental roadblocks? Could someone provide pointers to an in-depth
> discussion if it has happened before?

I don't remember any major discussion on this, not even when I brought it 
up on the list. It is true that the ElementTree adaptor for Java isn't 
particularly tuned. I would expect that you could get it to run a lot 
faster by running it through a profiler and turning some screws here and there.

Another thing to consider is the approach that lxml has taken: write a 
Python wrapper around a native DOM tree. But that's certainly a lot more 
work (and likely also a lot less memory friendly) than just tuning 
ElementTree for Jython.

Stefan