XPath vs find

[serrufSe]

Hi, Scanny,

Profiling the script that used python-docx revealed that the bottleneck is "first_child_found" method from "docx.oxml.xmlchemy._OxmlElementBase" class. The problem is in using a method that extends ElementTree API method "find".
Based on this article — http://www.ibm.com/developerworks/library/x-hiperfparse/, i replaced "find" method to xpath-request and gained performance boost. What do you think about this? Should we replace all "find" and "findall" usages with xpath-requests?

Thanks

[scanny]

What is the script you mention that exhibited bottlenecks?

[serrufSe]

Something like this

new_docx = Document()
i = 0
while i < 1000:
    new_docx.add_paragraph('one')
    new_docx.add_paragraph('two')
    new_docx.add_paragraph('three')
    i += 1

I used PyChar profiler

[scanny]

and to where did you localize the bottleneck?

[serrufSe]

http://postimg.org/image/phfcweav5/ with "find" and http://postimg.org/image/3pkewoymr/ with "xpath"

[scanny]

My question was where in the code did you find the bottleneck; What lines of code are consuming more time than strictly necessary?

[serrufSe]

https://github.com/python-openxml/python-docx/blob/master/docx/oxml/xmlchemy.py#L713 it takes about 80 % of total time.

[scanny]

Hmm, very interesting!

How do you change that line of code to produce the performance gains you've observed?

[serrufSe]

I just replaced

self.find(qn(tagname))

to

self.xpath('%s[1]' % tagname)

[genme]

Ping

[scanny]

I'm seeing slight performance degradation in the test suite when using XPath instead of find. (3.475s for XPath vs 3.361s for find)

I'm wondering if a broader test sample than just add_paragraph() is required. If it were such a decisive improvement I'd expect to see more change (and in the other direction) when running the test suite, since it tests the entire code base.

Also, that IBM article is from many moons ago. lxml has had a lot of revisions since then.