|
|
Boost Testing : |
From: Douglas Gregor (doug.gregor_at_[hidden])
Date: 2005-06-01 22:13:01
On Jun 1, 2005, at 12:52 AM, Misha Bergal wrote:
> Doug Gregor <dgregor_at_[hidden]> writes:
>
>> Perhaps we could optimize the XSLT enough to make it feasible again?
>
> I think it can be easily done. Here are some ideas:
>
> * the test data can be generated using
> xsl_reports/test/generate_test_results.py
>
> * I would make sure that issues.xsl generates correct results first.
There were a few things I needed to tweak because the testing system
has changed a little since issues_page.xsl was maintained, but it's
been pretty simple to hack.
> * Then I would use a real regression logs and commenting out parts of
> XSLT code to find out what the bottlenecks are.
After some poking around, I found this little resource leech:
- <xsl:key
- name="a"
- match="."
- use="concat( @library, '@', @test-name )"/>
It was probably just some experimental code dropped in at one point,
but yanking this out turned a 70-second xsltproc run into a 6-second
run. The full issues.html generation takes 5-10 minutes now, instead of
90.
> Don't hesitate to ask more questions - we very much appreciate you
> looking into this.
And I appreciate the help!
I've committed some changes to the XSLT and a little tweak to
boost_wide_reports.py. The issues page should build much faster now and
I've worked out most (not all!) of the kinks. If all goes well on your
end, could you re-enable issue page generation? Then I'll get down to
writing my spam-o-matic :)
Doug