For Ubuntu 16, I follow the `README` guide, thought there is `apt install virtualenv`, I still have to run `pip install virtualenv` manually to finish the environment setup. This is probably an issue.

cybai: I am now

ferjm: ha, thanks for pinging me again but I fixed my issue :D

I'd like to ask a question about the gstreamer error from mach yesterday but then I realized it's about `PKG_CONFIG_PATH`

so I sent #23175 yesterday :)

cybai: cool, thanks. I wonder if that could be done automatically by mach

ferjm: sounds like doable 👀 but I'm not sure if other platforms also needs the same env var 🤔?

yes, they do. We need the env var in linux, osx and windows

btw, somehow I confused why I was able to compile with my previous setting. I set the PKG_CONFIG_PATH with "/usr/local/opt/expat/lib/pkgconfig"

then I got into the issue after I did "brew cleanup" 🤔

okay! let me file an issue for it first :D

ferjm: ^ here it is :)

cybai: thanks!

🙌

I think we already do that for linux and android, but not for the other platforms

maybe this line 🤔?

yes

I found one thing interesting is that the env setting is run after gstreamer checking 😂

maybe we should check the env and set it before gstreamer checking?

let me add this to the issue description

ferjm: updated the issue description 🙇

cybai: well, we only need to set the PKG_CONFIG_PATH env var if the `pkg-config` check fails. So I guess we could first check with `pkg-config` and, if that fails, then try to set the PKG_CONFIG_PATH env var to the common known locations based on the platform until `pkg-config` is happy. If we are unable to make it happy, then there is no way we can set the

other env vars (LD_LIBRARY_PATH), so that's when we should throw the error

oh, I see, so we can just set it in `catch` and try to do `pkg-config` check again

if it fails, then fail

is that right? 👀

yes. IIUC the `pkg-config` check will throw an exception only if pkg-config is not available on the host. If it is available and the check is not ok, it will simply return False. So I wouldn't retry on the `except` handler. I would instead create a short loop to iterate over all the possible known pkg-config paths and do a set and try for each iteration. If

this loop does not succeed, then abort and throw an error

I see, let me update the description again! this way looks good to me!

Updated!

looks good, thanks

thanks for checking :D

standups: github.com/nox/moite_moite

lol

nox: exceptionnellement, peut-on avoir le super_moite_moite ?

Ah ah.

lqd: Btw I actually used thin spaces inside the quotes and before the question mark. :D

<3

(a non-breaking thin space one'd hope :p)

(even though it might be a full-width nbsp in guillemets I can't remember so ¯\_(ツ)_/¯ :)

I applaud the professionalism in both naming _and_ typesetting

SimonSapin r? servo/euclid #333

nical: Ouch.

nox: indeed: feel free to steal the review I'd rather push the fix asap :)

nical: r+'d already.

thanks

nox: I hadn't rebased properly and euclid was already at 0.19.7, r? servo/euclid #334

lqd: Wtf crates.io crates.io/crates/moite_moite

TIL thin spaces in Helvetica Neue suck.

:)

only people on mac should see this (and I'm not on a mac rn lol) right ?

lqd: Maybe, don't know.

lqd: irccloud.mozilla.com/file/HqLztqbG/…2019-04-08%20a%CC%80%2013.17.43.png

Works nicely for me :)

No thanks.

nox: one could argue this is rediscovering why all those rules for print are not so easily portable to the digital typesetting / web age :/

lqd: Or these thin spaces are just too thin.

for this specific cut weight/size/etc

TBH I'm not even sure macOS did let me input the right one anyway. Shrug.

standups: Did some fancy stuff around rayon, will spend the afternoon adding abspos tests to Victor.

paul: what's the need for shared textures on android now?

jdm: for immersive mode, fxr needs a SurfaceTexture

hmm, I see

jdm: and ideally, WebGL would use that same SurfaceTexture

the team at szeged believes they can start working on the texture sharing support next week

jdm: that would be great

jdm: does that work include updating WR to use external textures?

paul: yes

paul: I mean, it does already, right?

It uses the GL_EXTERNAL_TEXTURE_2D extension?

paul: oh, I think we're talking about different things

tell me more

jdm: in servo, we create a GL_TEXTURE_2D, right? We will need a GL_EXTERNAL_TEXTURE_2D to work with SurfaceTexture.

google does not find anything for GL_EXTERNAL_TEXTURE_2D

jdm: I think it's GL_TEXTURE_EXTERNAL

do you mean khronos.org/registry/OpenGL/extensions/OES/OES_EGL_image_external.txt ?

jdm: let me look

jdm: to render a surfacetexture in a shader, we need that special extension. I haven't checked in details why and how, but it's necessary in FxR

jdm: and yes, your khronos link is what we need.

jdm: I'm being told "a SurfaceTexture can't be rendered in a shader using a normal sampler"

paul: ok, that's good to know. I suspect they'll just be focusing on desktop linux to start, and hopefully they can build something that can easily integrate into android

jdm: this is for the immersive mode only

jdm: I'll ask Imanol to add some more info about that extension to the issue

great

pcwalton: how stable is pf3-pipelining? Should I see what the magicleap demo is like based on it?

I'm going to do the unicode functions for harfbuzz today, unless somebody wants to jump on that, as it's on the critical path

raph: Nice.

Does anyone have a version of the Rust logo without the R in the middle?

ajeffrey: hi! If you have a chance to read this, I'd appreciate some feedback: #23179 - i'd like to change the life time of the compositor

nox: users.rust-lang.org/t/how-to-remove…symbol-from-the-rust-symbol/25897/8

jdm: twitter.com/nokusu/status/1115286003357507584 :D

standups: Tidying up magicleap pathfinder demo codebase.

jdm: regarding servo/servo #23115#discussion_r270872628 if we store it in InProgressLoad, I need to satisfy this in ScriptThread::handle_new_layout() as well. Would it be ok to pass it as a member of NewLayoutInfo?

pylbrecht: yes, that would make sense.

jdm: alright, then I have another question: the compiler tells me that the trait `serde::Serialize` is not implemented for `std::sync::atomic::AtomicBool`. What should I do about this?

oh, hmm. that is indeed a problem.

pylbrecht: which structure is that coming from?

NewLayoutInfo

script_traits

jdm: falling back to storing the flag in ScriptThread itself? Although it's just passing through to Window.

hmm

does not feel right, tbh

pylbrecht: can we either create it in handle_new_layout, or accept as an argument, rather than storing it in NewLayoutInfo?

jdm: creating it in there could work, I suppose.

jdm: it would still be shared across layout and script, right?

jdm: Though I'm not sure, if it makes a difference to the first approach of UnpriviledgedPipelineContent::start_all().

pylbrecht: as long as the new value is passed in LayoutThreadInit, then yes

jdm: well, I just start hacking and see what issues come across.

if any.

:)

standups: reviews

sigh

our weird timeouts were caused by rogue servo processes

jdm: I ran into an issue: the other location, where InProgressLoad::new() is called, is in ScriptThreadFactory::create(), so we can't create it in handle_new_layout() :(

jdm: so we're left with passing it as a function argument?

pylbrecht: I'm not sure I understand why that prevents us from creating it as well?

Why are these MediaList2 tests failing? servo/servo #23096#issuecomment-480556323

jdm: I was thinking STF and handle_new_layout() are called sync

sbansal: it looks like they're failing because the stylesheet is missing

jdm: like after each other, in whatever order

pylbrecht: previously they tried to use ownerNode and threw an exception, so the test did not complete

er, sbansal ^

pylbrecht: I believe the handle_new_layout code path is distinct from STF::create

jdm: ok, I don't understand the interaction between the various components good enough, obviously.

jdm: Initally subtest 1 was failing, which is now passing but al other subtests started failing. stylesheetbis udes in subtest 1, so it should not be missing

pylbrecht: what if STF::create returned the new atomicboolean, and LTF::create accepted it as an argument?

those are only called in components/constellation/pipeline.rs

pylbrecht: or even easier, pipeline.rs creates the atomic boolean and passes it as an argument to both stf::create and ltf::create

jdm: the latter was the initial approach

pylbrecht: indeed. that's probably the right way to do it, then.

sorry for getting confused.

jdm: nevermind, git to the rescue.

pylbrecht: to be clear - we should still do the handle_new_layout changes I was talking about

but I now understand why passing in an atomicboolean in InitialScriptState makes sense

jdm: seems we're on the same page now.

sbansal: oh, MediaList2.xhtml used ownerNode and threw an exception previously

sbansal: so we never got a chance to run the tests that are now reported as failing

jdm: Oh, so should I mark them as failing now and report them in new issue?

sbansal: yeah, that owuld be great

sbansal: the trick is we don't support github.com/servo/servo/blob/2304f02…t/dom/webidls/StyleSheet.webidl#L15 yet

jdm: ./mach check did not complain, but unfortunately I'm running out of time today.

jdm: I pushed my changes and will start working on the communication with Profiler the next time.

good night o/

pylbrecht: good night!

jdm: how much do we care about 6DOF support rn?

because to properly support standing to sitting transforms I'll need to implement matrix decomposition

which we have code for but it's for css-like matrices, not euclid matrices, so porting that is going to be ... fun

Manishearth: at this precise moment I don't mind ignoring it, but I expect we would care about it in the next month?

cool

jdm: then again i have almost nothing on my plate if I don't work on that rn

I can:

- work on this

- work on moving euclid to column vectors (involves updating gecko)

- work on getting the transforms stuff into the spec

everything else is blocked on ferjm or spec folks :|

Manishearth: would making some kind of plan around webxr testing make sense right now?

.... somewhat

i was hoping we'd be able to webxrify rust-webvr itself and invert everything, but that may be more trouble than it's worth

(it's a perf win since we don't double-compute things like view matrices)

jdm: making a plan around this is doable i guess, the problem is that i don't want to modify rust-webvr while the ML port is ongoing

yeah, my fear is that we keep kicking the can down the road and nobody actually knows what it will take to get automated tests working in servo

Okay, I can go through the chromium code and see what they do

I have a rough plan for servo -- for everything but the final rendering bit

but I can verify against chromium

and against the wpt tests

Manishearth: what am I blocking?

ferjm: the scrcobject dom side

if you're not working on that i can pick it up, i'm just not 100% sure how to weave into the fetch code :)

i can implement it as a hack though :)

no rush, looking at the webrtc stuff was low on my list here, it's just that the webxr stuff is also blocked

i can also attempt to implement the XR stuff for input, it's not specced but i have a rough idea of what it's going to look like

Manishearth: there's an open PR for that already, but it has tons of WPTs failures that I need to check tomorrow

oh i missed that

ferjm: anyway, no rush

I'm not here for wed/thu/some of fri

yep, got it

Oh joy, more dynamic link errors with no hint about what caused them...

standups: docs.rs/moite_moite/0.2.0/moite_moite/sync/fn.split_with.html#example

ajeffrey: I just did a crate which is kinda as single-purpose as your swapper thing hah.

nox: it has one job to do.

ajeffrey: YES :D

ajeffrey: big yikes

jdm: I think the most likely cause of this is I gave in and updated the OS on my ML1.

[INFO ] Executing command ['/usr/bin/python', u'-m', u'pip', u'--version'] in directory '/root'

[ERROR ] An importable Python 2 pip module is required but could not be found on your system. This usually means that the system's pip package is not installed properly.

so far so reasonable

but if I ssh into the VM, `sudo su -` and put myself in /root, and run `/usr/bin/python -m pip --version`, I get totally valid output

Sigh, and doing the matching SDK updated produces...

OK, I get the message, you deprecated an API, how about a little tiny hint on what we're meant to use instead?

no use, only deprecate

amazing

the stock xenial VM comes with a python installation that has incompatible versions of requests and urllib3

and the salt code that decides whether pip is usable or not runs before salt can upgrade anything, and pip relies on requests

standups: got salt provisioning to complete successfully in vagrant in xenial VM.

ajeffrey: over my vacation I came up with a plan to integrate multithreaded rendering with the pipelining work. I'm working on that and then I'll start integrating your PRs

the reason why I'm waiting for this pipelining stuff to land is that it reworks quite a bit (and the multithreaded rendering reworks even more) so it's likely to break some of your stuff

but don't worry about rebasing if you don't want to, I can do that

standups: Working on getting pipelining working with multithreaded rendering again in Pathfinder 3.

pcwalton: sounds good. I am fighting annoying ml1 OS/SDK upgrades at the moment.

pcwalton: I'm going to be on vacation next week, if we can land stuff before I go that would be great

that looks likely

pcwalton: ta

standups: Rust-mozilla IRC coordination

standups: start reading chromium xr test coding

*code

eh whatever