Bug 10265 – SVG needs a Page Load Test (PLT)

• Component: SVG
• Status: RESOLVED FIXED

[Eric Seidel (no email)]

SVG needs a PLT

This would not be super hard to do.  All this would require is someone sitting down with some public domain SVG files (like http://www.openclipart.com/) for a while and finding 50 or so "representative" files (some large, some small, hopefully covering most of the features of SVG.

To make a plt file just means make a new text file named "mytest.pltsuite" with a bunch of urls in it (preferably file://) and adding it into the Safari.app resources folder.

Then under the debug menu, you'll see a "Show Page Load Test Window" menu item.  Open that to run your newly created PLT file.

We just need an arbitrary PLT file so that we can start to keep performance numbers for SVG as we continue to change things.

[Eric Seidel (no email)]

Another option would be to have a PLT for SVG which just used the w3c test files.  That's probably easiest to begin with.  "real world" SVGs are probably more representative however.

It would even be fine to include animation (script or declarative) which would need to run to completion.  fixed-length animations will increase the length of running the tests, but will not decrease their accuracy.

One further obstacle which will need to be overcome is how to install the plt suite properly on any machine.  Internally I wrote a little sed script to do that, however that script has not yet made its way into the public repository yet.

Eventually it would be nice to get all of the safari performance testing scripts into the public repository.  That just hasn't happened yet (and since I've left we'll have to convince a current safari team member to do so).

[Eric Seidel (no email)]

bug 10266 talks about moving more internal performance testing scripts into the public webkit repository.

[Eric Seidel (no email)]

An example pltsuite file was attached to 6010.

[Eric Seidel (no email)]

Created attachment 12049 [details]
patch including just the support files

Merry Christmas, WebKit. :)

[Brady Eidson]

Comment on attachment 12049 [details]
patch including just the support files

r=me with two comments
1 - In principle, this looks great but as far as being a "representative set of svgs" I can't really say - perhaps olliej or hyatt can better judge that
2 - I'd like to see a wait 30 after the first test then wait 10's for the cached runs.  This is as opposed to the wait 10 and wait 5's I see here

[Eric Seidel (no email)]

I'm not sure the numbers are correct.  (PLT may not work for XML correctly, not sure.) Autovicki still will need to be made available (something better than the silly run-pageloadtest I hacked together).  Also, we might want to revise the set of SVGs we use.  Once we get things working, a blog post announcing the test and asking for submissions could be useful.

Landed as r18427.

[Eric Seidel (no email)]

*** Bug 5038 has been marked as a duplicate of this bug. ***