Meeting 2014 01 06
farodin91 edited this page Jan 8, 2014
·
1 revision
- Servo workweek final dates, location, hotel? (jack)
- hotel: hotel vitale
- Rust upgrade (jack)
- ARM buildbot status (Rust + Servo) (jack)
- Some notes from initial page load perf investigations (larsberg)
- jack: Registration forms went out for Mozilla folks, etc. to register for the travel and hotel.
- azita: The workweek has been approved. Did you get the location?
- jack: It sounds like it will be in the San Francisco office. The hotel is:
- azita: Let's get the total number. I was hoping for maybe a conference room for us offsite from the SF office.
- dherman: I agree. That might be easier.
- azita: Is it easier for Samsung to have the meeting at the hotel for the week? Rather than the SF office?
- ryan: That is good.
- azita: I'll see if we can do that.
- ryan: Is the hotel near the office?
- jack: I believe so. I will send more information when I send out the registration form. We can do that. Dates are January 21st-24th (January 20th is a federal holiday in the US). Once we know the location, we will send that out as well.
- jack: Rooms are already reserved. There will be a group rate that you can get. If you need more rooms, our travel agent can find more rooms as well.
- azita: The sooner we can get the total count, we can find out if we can book a conference room. Can you get a total number of attendees this week?
- ryan: Yes. We will get it and get back to you soon.
- jack: This rust upgrade has been painful. There are two more outstanding bugs in rust blocking us. One is in debug info, which we have a good test case for. THe other is something that we can work around having to do with temporaries. A partial fix has landed in Rust, but we have hit another failure case for it. Rust issue #10683. Not blocking us, because we can work around the problem. Last thing after the debug info is ther eis to make sure that the android build still works and then we can land the upgrade.
- jack: Talked with Mozilla release engineering. We will convert one linux builder to an android cross-compile build. It will happen in the next few weeks. After that, we will gate servo checkings on android cross-compiles. There will be no tests runs yet. In the meantime, I will ensure Rust the upgrade builds with android. Brian, what's the status of rust on android in the compiler?
- brson: Unwinding was broken but got fixed yesterday and works again on rust. In the meantime, there were another couple of regressions that I fixed. I am working through the process of making it reliable. We test on every commit, but it does not get it out. I am working on Rust 0.9 release, but will be working on the Android bot as well.
- jack: Anything blocking 0.9?
- brson: No. I am trying to get the documentation cleaned up. I am moving the wiki stuff to official docs. It should go out in the next day or two.
- jack: Yet again, we will probably miss the official release by a couple of commits or so in Servo. Does Samsung have any comments on the Android stuff?
- lars: started investigating perf issues with Benoit Girard. the important thing is the initial set of numbers we came up with. there's been research on where load time is spent. evenly split betwen javascript, layout, rendering, and OS calls. the aggregator doesn't reflect real pages. real pages tend to be bound in one place. this is better for us because it's easier to parallelize those chunks of work. there are a lot of opportunities for optimization contrary to literature. i'm doing this work using firefox as the baseline. i want to try it with firefox for android as well.
- pcwalton: just seconding Benoit is awesome. he wrote the world's best profiler for windows, which is integrated into firefox. we should look at using that in rust in the future.
- lars: he's eager for us to do that work. not only for page load work but also on de-janking.
- dherman: where is hte dataset coming form.
- lars: i'm collecting it on big web sites. there are datasets in papers, but those are aggregated from say the alexa top 100. that loses all the distributions of hte individual pages.
- lars: i've mainly been hitting it with the alexa top 25 and while logged in. not logged in is so trivial so it's not useful to look at those. Benoit says that there are a lot of performance outliers that he sees coming in from firefox telemmetry. hopefully our parallel wins will be even better there.
- jack: is there a reason they study the pages in aggregate instead of individually?
- lars: rolling it up like that supported the theses of the paper.
- ryan: we've been working on acid2 tests. we're trying to finish acid2 by march. we have assigned teams
- ryan: when buildbot on rust and servo?
- jack: servo very soon, build only. Rust will come once the tests pass on the bot, probably shortly after we release 0.9.
- brson: I will work on it at the end of this week, after 0.9 is released (Thursday or Friday).
- jack: WIll any of those bugs block a rust upgrade with servo?
- brson: I don't know what they are at this point. The unwinding one was really bad, though.
- jack: We still haven't picked it yet.
- brson: Most of the test suite passes. Nothing major.
- jack: Then we should be able to get Android working immediately with the Rust upgrade, hopefully.
- brson: RIght now, something deadlocks somewhere in the test suite. It has been happening for about a month now.
- jack: Like a mutex?
- brson: It just freezes. Probably something with one of our hundreds and hundreds of hand-written atomics.
- jack: Assembler?
- brson: They use intrinsics.
- pcwalton: Modes: acquire, release, sequentially consistent. They all work fine on x86 because there's no difference, but on ARM there is, so things may go haywire.