what event loop does servo use?

something custom

fabrice: is it embedded into servo or does it exist as a separate crate?

qoxncyha: internal to servo I'm quite sure

fabrice: any chance you can point me to the source?

qoxncyha, which event loop are you interested in?

the constellation or the script thread?

or the layout thread?

are there different ones?

yes, they're all different

i think workers also have their own event loop

maybe.. layout thread?

what's the difference?

okay, then that's under components/layout_thread/lib.rs

the difference is that they're different implementations

qoxncyha, github.com/servo/servo/blob/master/…omponents/layout_thread/lib.rs#L607

qoxncyha, more accurately, it should be the run() function where we actually loop

i see

handle_request is where we receive messages from other threads

emilio: I'll look at the test & resources issue today. Can you expand more on the "test" feature you mentioned?

Paul: I was imagining to put the fake test resource loader behind a `test` feature used for unit tests

Paul: but I didn't think too hard about it, so there may be better approaches

emilio: that's better that having to register the resource thing manually…

emilio: thank you. I'll look at this today.

Paul: right, doing it manually is super tedious and hard to get right

Paul: because adding a pref read somewhere else usually doesn't involve modifying unit tests

i'm baaaaaack

jdm: \o/

how's our CI situation?

crowbot: infrastructure report

jdm: Everything is borken

jdm: welcome back! time for me to sleep though

heh

jdm: welcome back!

We have a mostly functioning History API now

woo!

jdm: welcome back :)

:)

jdm: hellooooo

Let's say I run a test for components/foo. components/foo depends on components/bar. Is `#[cfg(test)]` true in components/bar?

emilio: re #20710 - trying to figure out how to run test-specific code under components/embedder_traits when testing code under components/net.

jdm: yay, welcome back :)

Paul: hmm, so I thought we had a crate specific for net unit tests, but we don't seem to anymore

Paul: I think putting the feature in `[dev-dependencies]` in the cargo.toml in `components/net` _should_ work

Paul: #[cfg(test)] is not true in components/bar according to the quick testcase I just made

emilio: ok - I'll introduce a feature and enable it for tests see how that goes

jdm: thanks

ha

Paul: if you have a library with in-crate unit tests it’ll be compiled twice, once with cfg(test) and once without. Dependencies are not recompiled

\o/ jdm

jdm: welcome back! Have fun trawling through your INBOX.

SimonSapin: fyi I have a patch to remove all the build code and ports/geckolib code, just in case you feel like doing it :P

jdm: "Mark all read" is your fiend :)

*friend

nevar

SimonSapin: also, thanks for fixing the template :)

emilio: I’ll redirect that decision to our corporate overlords

SimonSapin: hah, yeah... Hopefully when I'm back from PTO the MR all-hands has already happened and we have a decision on what to do about all this mess. Oh well

huh, the amazon perf test became way more variable around april 6

jdm: CI was on fire, but I'll tell you on Wed. Things are ok now.

jdm: you mean the web archive tests?

jdm: I'm just here to give you teasers and frustrate you not telling the full epic story =)

ajeffrey: yeah

jdm: yes, I noticed that.

ajeffrey: it's been 3-6s forever, and suddenly it's either 3s or 12s

jdm: did we introduce a new lower-capacity linux machine?

(the tests aren't pinned to a particular machine)

(or maybe it just misses you)

yet another instance where it would be useful to record the name of the machine that runs the tests

jdm: oh are we not recording that. Oops :/

ajeffrey: that information is surprisingly difficult to acquire

jdm: we could do some archaeology on the logs to work out which machine ran linux-nightly on which dates.

jdm: hmm, apparently linux-2 is missing a file needed to do the upload to AWS: build.servo.org/builders/linux-nightly

ajeffrey: yeah, aws is missing on that machine

jdm: oops. A different problem though...

jdm: hmm, one of the highs is Apr 24, and close to a low is Apr 22, both on linux-6.

ajeffrey: my guess lies at the feet of servo/servo #20507, which is the only significant PR in github.com/pulls?utf8=%E2%9C%93&q=i…018-04-05..2018-04-07+user%3Aservo+

jdm: Hmm, not sure how that PR could have a perf impact.

cbrewster: ^

ajeffrey: jdm interesting, does any history traversal happen in those perf tests?

cbrewster: no, it just loads the page from a web archive.

cbrewster: there'll be some redirects but that's about it.

I mean, the easiest investigation here is probably just seeing if we can reproduce the bimodal perf behaviour with 4/5 and 4/8 nightly builds and then look for a regression range if so

jdm: I'll see if I can get bimodal behaviour locally.

jdm: bad news: perf trouble. Good news: we have actual perf measurements that caught this!

yep!

next step is building a tool to run daily that files issues when it discovers perf trouble :D

Argh, we now have so many build artefacts at servo-builds.s3.amazonaws.com that the listing is truncated!

Yay for the aws command line!

jdm: Ah, running that test a few times with the most recent nightly shows that the amazon home page is very nondeterministic,

I suspect it's randomly generating asset ids to load?

ajeffrey: the saved home page?

(e.g. doing some load balancing on the client side)

jdm: yes

jdm: the asset randomization might also be date-based which would explain why I didn't see this behaviour when I created the test.

ajeffrey: by the way, I'm super pleased to see the progress on bindgen-based SM

jdm: thanks!

I think that's a sensible way to break down the problem

jdm: at some point we need to have a planning conversation, perhaps in Chicago?

yeah, if we can carve out some time there it would be useful

jdm: I can add an agenda item under your name :)

perfect

jdm: done

jdm: i hear you have a good example of using rust as a lib for c code?

trying to write a quick test project for a miscompilation but can't get the linking right

Gankro: huh, do I?

who is spreading this information?

I guess not

Paul: regarding the constellation and embedder messages, I wonder if we couldn’t add a constellation message that takes an embedder message and forwards the message to the embedder. This would avoid adding new constellation messages for each new embedder message that needs to be sent from a script thread

cbrewster: that would work.

that's definitely a preferred way to duplicating messages

Sounds like a good way to move forward then

Gankro: Here's a bare-bones (non-Cargo) example github.com/rillian/rust-ffi

3 years old but it still works with current rustc!

that's impressive

mbrubeck: what's up with the awk and lib.a.out

Gankro: no idea

w-what

Gankro: It doesn't seem to be required, on my system...

the .out file is just empty so it ends up being a no-op

Aryx: Hi

What you already said in taskcluster, is it related to this ? treeherder.mozilla.org/#/jobs?repo=…=unclassified&selectedJob=176310290

cbrewster: \o/ re #20671

ajeffrey: woohoo!

standups: Tracking down variability of perf-testing archived amazon.com. servo/servo #20722

standups: Trying to reduce the size of the mozjs crate: generating inlined symbols produces a 3G binary!

jdm: re twitter.com/lastontheboat/status/991024050003640320 , I’m curious what motivates hiring specifically a new graduate

SimonSapin: I'm just amplifying an uncommon job posting; I don't know any details.

fair enough :)

jdm: welcome back!

:)

o/ jdm

wait sorry

dammit you're not in #servo-offtopic

either I'm on-topic or I'm offline

there is no in-between

pcwalton: could you take a look at servo/webrender #2705 when you have a moment?

gw: on it

thanks

larsberg: research.mozilla.org/servo-engines uuuuuh layout is part of WebRender, and WebRender is "to be written in Rust"?

SimonSapin: Thanks for the note! I'll follow up with Marketing to see if we can do an edit pass there. We don't own those pages (marketing owns everything on mozilla.org/mozilla.com, in general), but they're pretty friendly to changes :-)

SimonSapin, the last part "to be written in Rust" i understand, because it's looking at it from the perspective of Gecko

not so sure why layout is part of WR though

larsberg: don’t these kind of articles get reviewed by someone on the relevant team before publishing?

are we going to support WebRTC anytime soon btw?

standups: went on vacation

standups: helped with rust reach decisions

standups: making list of MR security papers