Issue1342
Created on 2009-05-07.20:21:22 by kfitch42, last changed 2009-05-10.20:01:33 by zyasoft.
File name |
Uploaded |
Description |
Edit |
Remove |
testET.py
|
kfitch42,
2009-05-07.20:21:22
|
|
|
|
msg4655 (view) |
Author: Kevin Fitch (kfitch42) |
Date: 2009-05-07.20:21:21 |
|
$ wget http://www.w3schools.com/XML/cd_catalog.xml
$ python testET.py
0.012904882431
$ jython testET.py
5.10699987411
I have done some testing on a variety of files using jython 2.5b4 and
current trunk (r6311) on ubuntu 8.04 (both 64bit and 32bit), and
slowdown of a factor of 300 to 400 is pretty common.
|
msg4657 (view) |
Author: Philip Jenvey (pjenvey) |
Date: 2009-05-07.20:32:20 |
|
Sébastien - can you look into this?
|
msg4659 (view) |
Author: (boisgera) |
Date: 2009-05-08.08:02:23 |
|
Yes, I'm on it. I was expecting a poor performance for the Jython version
of ElementTree (the expat module being pure Jython, not Java), but
definitely not 5 secs for a 200-line xml file.
|
msg4660 (view) |
Author: (boisgera) |
Date: 2009-05-08.11:48:22 |
|
I think that I've found the bottleneck: shame on me for using 'eval' for
convenience ... expect a fix in the next few days.
|
msg4662 (view) |
Author: Philip Jenvey (pjenvey) |
Date: 2009-05-08.19:20:11 |
|
Thanks Sébastien. FYI we're aiming to release a release candidate on about
Sunday
|
msg4663 (view) |
Author: (boisgera) |
Date: 2009-05-09.10:17:25 |
|
fixed in r72 of http://code.google.com/p/jython-elementtree. Expect a
slowdown factor of 20 instead (rough estimate) with respect to CPython.
A benchmark has been added to measure this performance:
benchmark$ python run.py
ElementTree benchmark -- Python 2.5 on linux2
alice.fodt : 0.11465 s
tiny.xml : 0.00010 s
lion.svg : 0.00396 s
cd_catalog.xml: 0.00957 s
benchmark$ jython run.py
ElementTree benchmark -- Python 2.5b3 on java1.6.0
alice.fodt : 0.81000 s
tiny.xml : 0.00200 s
lion.svg : 0.06600 s
cd_catalog.xml: 0.09600 s
|
msg4666 (view) |
Author: Philip Jenvey (pjenvey) |
Date: 2009-05-10.19:11:29 |
|
Thanks a bunch Sébastien, speedup applied in r6330
|
msg4667 (view) |
Author: Jim Baker (zyasoft) |
Date: 2009-05-10.20:01:32 |
|
When I run the original test by kfitch42 with 100 iterations (to allow
hotspot to kick in), Jython is now about 7x slower. No doubt there is
still room for optimizations, but this is not so bad.
However, that's the python version of ET. cElementTree on CPython is
about 100x faster still! So there is significant room for improvement by
profiling and writing the core of our Jython implementation in Java.
|
|
Date |
User |
Action |
Args |
2009-05-10 20:01:33 | zyasoft | set | nosy:
+ zyasoft messages:
+ msg4667 |
2009-05-10 19:11:29 | pjenvey | set | status: open -> closed resolution: fixed messages:
+ msg4666 |
2009-05-10 06:53:36 | pjenvey | set | priority: high assignee: pjenvey |
2009-05-09 10:17:26 | boisgera | set | messages:
+ msg4663 |
2009-05-08 19:20:12 | pjenvey | set | messages:
+ msg4662 |
2009-05-08 11:48:23 | boisgera | set | messages:
+ msg4660 |
2009-05-08 08:02:23 | boisgera | set | messages:
+ msg4659 |
2009-05-07 20:32:20 | pjenvey | set | nosy:
+ pjenvey, boisgera messages:
+ msg4657 |
2009-05-07 20:21:22 | kfitch42 | create | |
|