Ted's Mozilla Blog

So, after much wrangling, we have turned on profile-guided optimization on our Windows nightly build machine. The immediate impact is that we got faster, by about 10% on some of our benchmarks. We also exposed at least one tricky layout bug that relied on undefined order of evaluation, but dbaron fixed it. Big thanks to Rob Sayre and everyone else that made this possible!

Next up is probably going to be turning this on on our Linux nightly build machine. I think we’ve resolved the issues there, but we’re going to wait until after beta 4 for that. Apparently we shipped Firefox 1.0 nightlies with PGO, so it should be ok, although that was back with gcc 3.3 or so.

We’d like to do this on Mac, but that still needs some work. I’m hopeful that we’ll get there before Firefox 3 ships.

I always get a kick out of seeing Firefox being used on TV shows.  My wife was watching an episode of Nip/Tuck that she had taped (yes, taped, we don’t own a TiVo) and one of the characters was looking up some information on the web.  I did a double take and made her rewind and sure enough, they were using Firefox.  Even better, it was on a Mac!  In your face, Safari!  Of course this is Nip/Tuck so the characters have found out they’re related and are looking at a pro-incest website, but that’s one of the least edgy bits given the show content.

After a lot of hard work by Ben, we’ve finally released an updated MozillaBuild package, version 1.2.  This version contains a number of fixes and additions over the previous 1.1, including:

• Support for Visual C++ 2008
• Support for the Vista Platform SDK
• A non-ancient version of ssh
• A l10n start script that doesn’t require a compiler or SDK
• A fix for that annoying ClearType bug that drew lines next to every character you typed
• Included a Unicode version of NSIS

You can see some of the details in the tracking bug.  We also finally created a MozillaBuild component in bugzilla, so you can file bugs in mozilla.org : MozillaBuild if you have problems or requests.  Big thanks to everyone who contributed code and testing to make this release possible!

Thanks to the combined efforts of a few people, the Tinderbox build logs for our unit test machines now suck much less.  You can now click on “View Brief Log” and get a summary of test failures right at the top, instead of searching through the full log for various failure strings.  In addition, if you click down to the errors in the body of the log, the test files are linkified to bonsai for you.  Awesome!

In Firefox 3, the bookmarks and history systems are getting an overhaul, collectively known as “Places.”  I was skeptical of this work when it first started, and I’m sure I wasn’t alone in that view.  I rarely used bookmarks, preferring to just search Google to recall a page I had previously visited.  Just how relevant was any overhaul of bookmarks going to be to my modern browsing?

Well, I was wrong.  Places is here, and it’s awesome.  It has improved my browsing workflow so much that I can’t even use Firefox 2 anymore.  In Firefox 3, the urlbar now autocompletes what you’ve typed from both URLs and page titles from your history and bookmarks.  Maybe that doesn’t sound impressive, but in practice it’s fantastic.  I tend to visit a lot of pages on the same site (mostly bugzilla), and with the old urlbar autocomplete, it was hard to find things from bugzilla.  With the new autocomplete, all you need to do is type in part of a unique word from the page title or URL, and it will be found.  In addition, if you bookmark a page (just by clicking the handy star there) it will get weighted into your autocomplete results.

I hear there’s still more work to be done, including improving the formatting of the results, but it’s already vastly improved my day-to-day browsing.  Hats off to the Places team!

So, some time ago dbaron got RLk down to 0 bytes on our leak test box.  Sometime after that, we deployed the new Linux reference platform, only to have that go back up to 8 bytes.  Turns out it was my fault, a string wasn’t being freed in the crash reporter code.  The crash reporter must not have been enabled on the previous reference platform.  I’ve made amends and fixed this, and I also checked in rhelmer’s patch to make the leak test boxes turn orange if RLk goes above zero, so we should be able to hold the line on this per our test failure policy.  For comparison, on the 1.8 branch we leak up to 45KB(!) per test run.

I’ve had enough people tell me that I ought to have a weblog, so here it is.  I’ll be writing somewhat infrequently about the various things I’m working on.  This way, when we meet at a party, instead of me explaining all the cool things I work on, you can say, “Hey, I read about all the cool things you did on your blog!” and then we’ll high-five and it will be awesome.