Three Monkeys, Three Typewriters, Two Days

October 22, 2008

Loadgroup callbacks

One interesting change we made with SVG external document references is that the notificationCallbacks of a loadgroup are no longer necessarily something you can get a docshell from. I think this was already the case for images, but it will now generally be true for all external SVG resources and everything loaded from such documents. The reason for setting it up this way was to make sure people don't by accident treat an external resource load as being a load in the display document. That said, the load is of course associated with the display document...

So the question is whether there are extensions that rely on getting a docshell or some such from the notification callbacks, and if so what they need it for and what they'd expect to get for the external resource documents. I have no idea how to find this information, unfortunately, so I welcome any pointers.

Posted by bzbarsky at 9:29 AM

October 21, 2008

Fixing scroll restoration on no-store pages

If someone wants to take a brief dive into Gecko code, I'd love to see a patch for bug 215405 as described in comment 131. I can promise speedy reviews!

Posted by bzbarsky at 11:57 AM

October 14, 2008

HTML (and a bit of Python) help needed

The hgweb annotation output on is painfully slow. On large files it's two orders of magnitude slower than our old Bonsai output.

I've filed bug 459823 on getting this fixed. There is a corresponding bug on mercurial which has lots of analysis and some comments about what needs fixing and how to start.

The upshot is that the template for the annotation output needs to be changed (HTML coding plus a bit of Python, I assume) to output HTML that doesn't require O(N^2) behavior to render. Sadly, I'm a bit swamped for time to work on this right now, but if someone wants a way to seriously help out Gecko development without touching C++ code, this is it!

Posted by bzbarsky at 9:04 AM

October 7, 2008

Regexps and writing tests

David Mandelin's blog post about one of the sunspider subtests is chock-full of fun information on regexps. It also highlights some more perils of writing performance tests, especially when there are optional features involved, or when the tests don't do a good job of testing whether the behavior is correct in addition to being fast.

Posted by bzbarsky at 7:07 AM