Skip to content

Meeting 2015 08 24

Jump to bottom
Josh Matthews edited this page Aug 24, 2015 · 1 revision

Servo Meeting 2015-08-24

Vidyo links:

Reminder: use http://benjamin.smedbergs.us/weekly-updates.fcgi/

Agenda items

  • Should we enable the experimental flag for wpt/css tests? (Ms2ger)
  • Gecko use counters (jdm)
  • SSL libraries (mbrubeck)

Attending

  • cimes, laleh, mbrubeck, larsberg, jdm, jack, pcwalton, ms2ger, mrobinson, frewsxcv

Last week

Experimental flags for tests

  • jack: Is Simon there?
  • larsberg: probably hosting the CSSWG stuff
  • jack: mbrubeck, have you tried running with -e in the test framework?
  • mbrubeck: None of my stuff is behind -e. We have support in the reftest harness for -e, but not in WPT.
  • jack: Unsure if we should have -e on by default. But all of the test harnesses should have the option.
  • mbrubeck: Yes, even if just for development.
  • jack: Once we have the support, we can schedule a run with it on the CI hardware. Not sure there's any reason not to run it with -e...
  • mbrubeck: One tricky part is that if we want to track failures, we'd need a second whole set of test expectations. Another bazillion metadata files... At least for the subset of things that test flexbox, we'd need another one. Maybe we could have conditional expectations a la our per-platform tests?
  • jdm: Sounds hard. Especially when we ask people to run & update manifests automatically based on those outcomes.
  • jack: Maybe chat with jgraham about the feature and experimental testing? If he thinks it's a good idea, then maybe we should. I worry about these optional flags and ending up like Cargo where you update something and then the other configs don't get properly updated. Who wants to file the bugs? Sounds like an E-Easy bug... ( https://github.com/servo/servo/issues/7335 ) We can schedule a run after that lands. After we have results, we can talk about it again.
  • mbrubeck: Paul already filed an issue, which is in the notes (above).

Gecko use counters

  • jdm: Gecko recently added the ability to have use counters for DOM properties and CSS properties and webidl methods/attributes etc. If there are things we'd like to measure the usage of, we can file bugs to add those measurements to Gecko.
  • pcwalton: That will be great! Chrome use counters don't support the webidl methods. Looking at chromium's numbers for CSS properties helped us prioritize our work. I'd like to know things about SVG. Perhaps content-editable features, too. Have a theory that SVG support is not that difficult because most sites just use simple paths and that's it, but don't have any data. Same with content-editable - nobody knows what' actually used.
  • jack: Couple are also known parallelism hazards - mixed writing modes plus uncleared floats. Not sure if the system can handle it.
  • pcwalton: Tricky to do in Firefox, because Gecko doesn't determine that data for floats. Kinda have to replicate a chunk of Servo's layout in Gecko.
  • jack: Just measure the use of float and clear?
  • pcwalton: Matters where they are in the page. e.g., is the clear inside the float?
  • jack: The raw number of pages that use float but not clear is not useful?
  • pcwalton: It'll tell us something... but now sure how it's related to parallelizability. Could gather it easily in Servo because we have that info available.
  • jack: How about mixed writing modes?
  • pcwalton: Probably easier. Weird flexbox, too, like vertical inside of horizontal. But not as key as SVG. My gut tells me flexbox is just used as a simple subset.
  • mbrubeck: Also true because most browsers only supported a subset of flexbox until recently.
  • pcwalton: Also, directional + flexing covers most needs.
  • jack: All those things are already there, right? So we can go get some of it out of the analysis?
  • jdm: I believe they just have the infra to record now and are waiting for people to request said recording.
  • jack: Please request SVG!
  • pcwalton: Would also love to see a big list of DOM APIs, ranked by use, so that we can see where Servo's at on getting them all. Really helped for CSS and our race to web compatibility.
  • jack: Slightly confused... why wouldn't you just change the code generator to generate them all at once?
  • jdm: Annotations to every webidl file.
  • mbrubeck: Also, gigabytes of data per page visited.
  • jack: How are they planning to use this support? For Servo, we clearly want the full GB of data across the Alexa top-100. Not quite ready for targeted queries, at least for Servo.
  • mbrubeck: https://github.com/w3c/touch-events/issues/29 we were discussing a method, and the chrome team mentioned there are numbers on webidl...
  • pcwalton: There are two forms. CSS properties, which covers all of them, and then these other ones are ad-hoc for answering specific questions.
  • mbrubeck: So, like Gecko.
  • pcwalton: I discovered them, but then found out they weren't super useful because they just answered one-off questions for the Chrome team (mainly about deprecating stuff).
  • jack: Ick, lots of deprecated webkit features in the top 10...
  • pcwalton: Not a good use of our time.
  • jack: Depends. Lots of the complaints about mobile web compat are related to us having a policy about not implementing those things...
  • larsberg: Speaking of content-editable, there is an editing working group meeting Friday Simon will be at...
  • pcwalton: I mainly want to know what we should implement for maximum web compat.

SSL libraries

  • mbrubeck: Mainly relaying a weekend IRC convo. Briansmith of pkix cert. validation fame is working on a new port of pkix to Rust + C. Servo doesn't have cert validation yet apart from the basic stuff built into OpenSSL 1.0, which really doesn't work. Brian says his new thing will be public this week-ish. Based on BoringSSL, which is a stripped-down OpenSSL. So, we might think about moving Hyper from OpenSSL to BoringSSL (or when/who). I know Corey was interested in this.
  • jack: There are some folks in Mozilla working on Rust crypto stuff. Don't know if it's public... but it's not for replacing the cert bits. I'll ping them and see if they're talking with Brian, since they are possibly working on complimentary things. brson, do you have any insight into Rust & crypto stuff?
  • brson: klutzy or yurume has a project, but it's not under active development.
  • jack: I'll try to round up everybody that people have mentioned :-)
  • mbrubeck: Briansmith has already pushed out ring, which has the primitives and a start of the Rust port. Based on the BoringSSL primitives.
  • jack: Removing all the bad-news fallbacks?
  • mbrubeck: BoringSSL is already doing so. Here's the current project: https://github.com/briansmith/ring
Clone this wiki locally