Kommentare zu: XML parser performance in PyPy

Von: Stefan Behnel

Stefan Behnel — Tue, 17 Jul 2012 11:04:27 +0000

Right, PyPy support is a feature that is still under development.

Von: Valery

Valery — Tue, 17 Jul 2012 10:58:03 +0000

For lxml, you””ll need the latest github source versions from both lxml and Cython as well as the dev packages of libxml2 and libxslt.
0.16 (released 2012-04-21) will not work?

Von: Stefan Behnel

Stefan Behnel — Thu, 28 Jun 2012 16:36:52 +0000

Hi Maciej, I agree, the memory usage might hint at a leak - couldn””t look into it yet. If you want to give it a try, I put the code and data files here:

http://lxml.de/etbench.tar.bz2

For lxml, you””ll need the latest github source versions from both lxml and Cython as well as the dev packages of libxml2 and libxslt.

Von: fijal

fijal — Tue, 26 Jun 2012 21:36:34 +0000

Hi Stefan

I think the pypy memory usage is not really acceptable. I would be happy to look how to improve the situation (maybe you cannot, but who knows). It”’’s definitely not just “jitted code”.

Feel free to post a request to pypy-dev or even a bug (but no promises when I””ll have time to look at it)

Cheers,
fijal

Von: Stefan Behnel

Stefan Behnel — Tue, 26 Jun 2012 09:33:22 +0000

I updated the numbers to more correctly reflect the PyPy behaviour with and without JIT warming. Note the huge memory usage of PyPy after JIT warming.

Von: Rene Dudfield

Rene Dudfield — Tue, 26 Jun 2012 04:31:32 +0000

Woo! I””m very happy pypy is doing well with xml, and that lxml is on pypy now! Rejoice.

Can you publish results without warming up the jit? This is a more reasonable case for some of my tools. cmd line tools that, Parse xml, do something, quit process. I think your tests would be a good to guide on performance for long running python web servers that parse xml though.