Meeting 2014 01 27
Lars Bergstrom edited this page Jan 27, 2014
·
1 revision
- build fontconfig on linux to ensure we have a thread-safe version?
- what's JSManaged blocked on?
- reftest & contenttest statuses (larsberg)
-
- linux strategy?
- larsberg, jdm, metajack, pcwalton, ms2ger, dherman, azita
- jack: We fixed the bug, so we might as well have our own version for Linux. We already build it for Android, so we should just upgrade it and build it on both. Probably a reasonable thing in the short-term; I can do this today. Opinions on what we should do on these platform dependencies that are both Linux & Android? In Linux for both, android if only one?
- jdm: Good.
- jack: Moving to Linux instead.
- jdm: Finishing the conversion to use RefCell internally for proper borrow checking. Also on deciding how to solve the safety problem of multiple JSManaged instances on the stack (GC hazards there). Then, also on some investigations of performance implications (maybe just something to track).
- pcwalton: Just follow up on perf. Is the macro approach for rooting OK?
- jdm: Haven't looked into it yet. Not sure if it's needed before we have precise rooting. One strategy for now is to disable the JS GC while running Rust code and re-enable it when we call back into Rust code.
- pcwalton: I'm fine with that, as long as it's not a burden to add exact rooting in the future.
- jack: Seemed liek we could fix all this if we had a modern spidermonkey and rust. Land JSManaged before that.
- pcwalton: Exact rooting is orthogonal to jsmanaged.
- jdm: Don't have an easy way to add RC'd roots. Can do addobject(root) and get a pointer, but if you call add and then remove it, it just removes it.
- jack: Know there's a perf problem?
- jdm: Not implicitly copyable, so I have to add clone() calls everywhere and I need a benchmark to see waht that looks like.
- jack: Happened before?
- jdm: It was implicitly copyable.
- jack: Weren't @mut's being copied and the rest? So it should be similar?
- pcwalton: Yes, it should be, unless there's a bug.
- jack: #[deriving(Encode)] also related?
- jdm: I have a patch that should work to unblock that.
- jack: Considering that ms2ger is interested, I assume it would improve the lives of those on the content side. Any idea of a timeline?
- jdm: Hard to say - I have Gecko things this week. I may put it up and take volunteer contributions. ~2 weeks.
- jack: If you do push it up, if you could write a little mail to dev-servo to ask for specific help, I can help find such a volunteer.
- larsberg: reftests are now turned on and enabled. three of the reftests have regressed and i decided to disable those particular ones. but now we won't break acid1. i did find out that that one of the samsung PRs has some platform differences that I need to track down. I'm looking into that not. I'll open issues for the three broken reftests. on the contenttest side, shutdown continues to be a little bit fragile. pcwalton's performance work has exposed another shutdown hang. i'll have to invest some time into patching our shutdown. -jdm: do you know it's shutdown related?
- larsberg: yes, it's freezing in the headless compositor recv_opt waiting for a message that will never arrive. i'll turn on message debugging to determine the sequence of messages that we are getting. now that we've lost task failure, we hang in those cases, and we'll have to harden servo against that.
- jack: i would like to get these enabled on linux as well since platform differences are starting to appears. what is blocking this? does gecko have a software/headless mode?
- jdm: gecko does not have headless mode.
- larsberg: do they start x?
- jdm: i guess so.
- pcwalton: we might look into how webkit does headless mode.
- jack: that is blocked on a software compositor.
- jack: can use llvmpipe to do it (http://www.mesa3d.org/llvmpipe.html )
- jack: ping bhearsum if he has suggestions and see if the reftests just works. since we check return codes
- larsberg: it's also possible it was fixed with my work on cleaning up nvidia startup etc.
- jack: It's the fontconfig issue. I'll land that change today or tomorrow so that it works again for people.
- larsberg: Can we share it now?
- jack: We will pull out the column of potential tasks and put it up on the wiki Roadmap shortly.
- jdm: bz is in town for the gecko workweek. questions?
- pcwalton: I have some. And it's in SF, so I will find him.
- jack: Maybe ask about JSManaged? And maybe run the planning stuff by him, particularly the ordering of big features?