W3C Web Performance WG, 14 Feb 2019

Previous Meeting

⏯

youtube image

►

From YouTube: W3C WebPerformance WG design call February 14th 2019

Description

Agenda: https://docs.google.com/document/d/10dz_7QM5XCNsGeI63R864lF9gFqlqQD37B4q8Q46LMM/edit?pli=1#heading=h.a5xnjgd4ryvk
Minutes:
https://docs.google.com/document/d/e/2PACX-1vTO8f2vYWLHDwk57Zk4FYNMkSBaUkppAJelPT26uMYu_aAsF03XXonmVLHm_ULOohPqr_tyhZNIHGxf/pub

A

A

Hey Marcos, yes, are you I'm good new, pretty good yeah I lost I recovered from I? Could yeah I hear it's also been, did yeah a lot of still going on so.

A

B

A

B

Thank you, hey Marcus should.

C

Be are you good.

B

You know and have you been coming to my birth.

C

Not that much I skipped the last two ones, the previous one, that was sick, other one before I differently. Oh.

B

Yeah I'm glad you came by what it is topics. Yes, I, don't know what else is on the agenda, but we have scheduling yeah.

D

Good morning morning, Steven were you coming from.

E

I'm calling from San Francisco I work at Salesforce and actually I met you a couple of times also met a couple of sugar is at blink on I think why, by who and working a front-end performance for Salesforce, that is great great, that you King John. Thank you.

C

A

So should be, do you know if the fun join us.

B

Did you want to start with some of the other admin stuff.

A

Sure I don't know if we're not gonna wait one or two more minutes for Todd to join I think you should be doing, but.

D

But then we can start with the.

D

A

Okay, so let's get started, he's Nikolas.

A

His Nikolas has Nikolas drawing or not as well sorry room troubles. Okay get in here, yeah I'm, just wondering if I.

F

A

Start with the usual user timing, l3 and performance timeline or yeah or.

A

Yeah, it should be. Do you notice the fun is.

B

Very technical difficulty.

A

B

But yeah I mean okay.

A

So, let's just okay, so he's joined cool okay, so just the administrative stuff. We have meeting minutes documents on the like linked from the agenda. Can someone volunteer to scribe.

A

I'm also recording the call and will hopefully upload it to YouTube, assuming there will be no technical difficulties related to that, but I still need someone subscribe in order for us to. You know, communicate in writing and then yo.

G

A

It okay and regarding the next call two weeks from now, I will be theoretically available, but just getting off a plane, so it I prefer, if that's possible, to have the next call three weeks from now, but then we'll have the next call after that a week after that, in order to avoid opening up a gap from our desired schedule to our actual schedule does that make sense to everyone.

G

So I heard March 7th, so any object. First, seven I.

G

Don't care new objection so put that in the notes? Oh cool I just.

A

Okay, and so shall we get started with the actual schedule, which is a scheduling, API.

F

A

uh Yeah I'm! Sorry about that.

H

But I just checked: I am available. My 7c is it possible.

H

So next one's triage meeting right, yes.

H

You may say: I can change my appointment.

A

Otherwise, we can move it around during that week. It's also like it would work for me. I, don't know if other people have.

A

If yeah, if we can move it either earlier or to Tuesday or Monday, that can work for me, risk.

I

It with a different time on the.

G

Same day, one more 7 to work better if we move it earlier.

G

Already right so then March 6 Wednesday.

A

um Wednesday is oh, no I can't I washed. It I can't.

I

March 4 we're ruling out options quickly, north works for me, or is it going once going twice sold.

I

8 right yeah, yeah yeah is 8. Yeah, freeze, metal, ok, I'm selling it for the third time 8.

A

A

So ok, so let's get back to the agenda Stefan. Do you want to start with the frame budget proposal.

A

B

Let me see if he's.

A

He's on the call and I think.

J

He's trying to write.

B

A

B

He's saying he can hear so maybehe we got here new like any wage. While these this is uh bidding uh hey, I guess I can provide a little bit of context.

J

B

I guess feel free to dive in I was just gonna, provide a tiny bit of context, which is I've talked about scheduling on a previous wall as well as a tea bag, and well we've been working with a bunch of partners who are have scheduling, systems and JavaScript, and so the proposals that chiffons gonna talk about today are basically incrementally help them with their problems. So you have to take it away.

D

F

So um window dot, so I should first say that all this sort of naming in a lot of the shape of this stuff is not set in stone. We do want to ship these behind a flag and chromium soon as an experiment and possibly an origin trial, to to see how people feel about them, but um were probably some distance from a formal standards process for any of this stuff, so the first one is window dot time until next frame.

F

So it's method on the global object, I think the best reference point for this is the quest. I will call back which not sure how commonly known this is, but when you register a callback with request, I would call back you receive an argument to your callback function, which is an idle deadline. Object. The idle deadline.

F

Object has a method time remaining and every time you call it, it will give you a number in milliseconds, which is the implementations best guess of how much time you have left to run code before something urgent needs to be serviced without going into the details too much. Basically, if there's no, there are no. Pending events, input events or any things with that urgency, the time you get back will be the estimated time until the next rendering update is scheduled. If there are no rendering updates scheduled you'll get a maximum time of fifty.

F

So what we've been hearing from developers and I should say that people we've been talking to most our react. People at Facebook, as well as the Maps team here at Google, and then I think she's talked to some other external developers, but those are the big ones that have been it's warming, a lot of what we've been doing just.

K

Real quick, I'll interject because communicate this to Schumi I've got about 20. Microsoft teams lined up over the next two or three weeks to gather their feedback and whether or not they can immediately gain values from these eight guys. So in about a month, so I'm not gonna have I'll go to show them. Yeah do.

B

You have a sense of whether they have attempted like scheduling strategies. Yes,.

K

They have, in some piece of their systems, attempted to segregate some work in it, but not like a react scheduler or the map schedule they're, not that sophisticated, so I'll have a better to breakdown. Like I said, once I've gathered all this information and cold great thanks by the way they all think they want the api's I. As you know, I'm I'm always been questioning. So we'll see. How many will you benefit sure yeah. So.

F

The implement going back to the idle deadline object that you get in requests. I will call back the time remaining method on it.

F

What we heard from developers, even those who use or request I, will call that quite a bit is that they are. They would like a signal about the rendering schedule.

F

Apart from whether they're appending input events, so they really want to be able to to intelligently schedule, work around frame deadlines, which is why they don't so that again the idle deadline object that will return you. You know if there's a pending input. Events in the time reigning remaining method will return zero on deadline. They they, the developers we talked to wanted to not have that behavior. They wanted it strictly based on frame three budgets, so this would be a method that could be called any time, not just in request.

F

I will call back any time you kinda wanna, you can imagine if you are a sophisticated developer and you're writing sort of an event loop within JavaScript and every time you get to the top of the event loop. You want to check and see. If you have enough time left to run another unit of work or whether it's time to yield back to the browser's in the browser can update them, I mean that's kind of what this is geared towards.

H

So I think this API is a bit problematic for us, because I don't think we will know this time necessary because we'll rely on cz or seer to let us know when the next frame is but like to do that. We have to raise the call back to do that so like in order to know the next time. Rendering opportunity will happen. We have to basically render sort of.

F

So I'm not familiar with with the west nation, but in I will say in chromium. When we you know we, our system is aware of the refresh rate of the hardware, so we just whatever.

H

Refresh rate would dynamically change the basin's by the system so I we like it, for example, if like if the OS detects that application is not catching up with the frame rate, we would automatically filter back to 60 Hertz or in the 30 Hort. So the system like itself is dynamically. Changing the freshmen of hardware to save the battery life in that case, so does the so.

F

Theoretically, chromium can work the same way, but whenever we get the frame signal, we get an extra field, which is the current refreshment that the system is working at now, the hardware. So if we, if we were to apply some Intel, there's actually a discussion of this applying some acknowledges so for all go back the frame we would still be communicating to the rendering stuff. This is the rate that will run out right now.

H

Okay, that's that's assuming we did render like worth something we right, ender right in that case we would know, but in EXO like this with this API, because we are taking about the future. It's a bit for me because if we'd never tell to the system, oh I'm I need to render then system will never tell us the name. So just as.

F

A bit of analogy: if you did a request, I will call back and your callback was invoked and you check the idle deadline. Object and you check time remaining and it's a 50 milliseconds. Then you would conclude it there. No there's no rendering scheduled. If the next thing you did was update style, the Dom node. Obviously the 50 milliseconds is no longer the right number. So don't.

H

Probably from that sure I mean, but I saw so, I have suggested this multiple times in the past feedback in various forums, but I think the better API is for alternative to requestanimationframe, where the function Kaaba itself will tell you the next time that the timing on which the frame will be rendered right like so so you don't objected to providing this information. It's just that. You don't like me. In fact, that's exactly what happens in like CCTV. They will tell you exactly when the next frame will be rendered right.

H

So, like they'll tell us the time, so we you know in fact, for most of native a guys like structure this way that you you, you know you ask me in GP or whatever, and then they'll tell you like. Okay, this is the frame like this is a time all the time them is to specify the frame you know the timing of which next time should be rendered to the system. Those are usually to be negative, I'm, not sure.

H

If all of your objection, though, if that information is freely available, like a native, it's it's not funny, I mean that's what I'm saying they will tell you when you about the rent, like you tell them to okay, I have things to render, and then they will call you back with okay render for this time.

F

Okay, so that well says so, is that basically, the equivalent of requestanimationframe, basically yeah, listen, um and so that would be after the rendering system has already received a signal saying we would like you to produce his reign of contract right for this time. Okay, so is it not I mean presumably the system?

F

How so the system never there's no way to get the system's the refresh rate that the system is using at any given time not available. No.

H

Really the system's dynamic I just think as we speak right all right, but you could presumably.

K

Is available but at any moment the system may trouble effectively to slower than vsync I. Think is your statement rectory also I.

H

Think I'm sure you know looking observe what the system is doing. You know and sort of guessed what the system's frame rate is right. Anything that is no so that's my question is it doesn't have.

F

To be guesswork, or is there called the system to say what is your current.

H

Check with my bollocks I, don't I, don't think, there's any API right because yeah in general, this is a little bit problematic, because even if you ought to guess right, like we don't know when a system decide to change that time, feature agreed. So it.

F

Would be, you would certainly be an estimate, but it's so the if you think about the the problems that developers have now. They may have a pretty good idea that the system is running at 60 Hertz, but that doesn't tell them that much about how much time they have left, they might have Q milliseconds left and they might have 10 dollah seconds left. They just don't know so. They're they're flying blind now this would be sort of making them fly a little less blind.

H

Honey act like use kissies, you know this thing that I mean the way requestanimationframe in what works is even is, as currently doesn't want to fix it, but the fires are right. After the bee sting cards right, I, think after we get the call back, we call them back. So the reason we did that is that way, the requestanimationframe happens. You know, but a fixed, more free. You know uniform minute about what the VC happens with that, like you should be able to prepare the frame for the next frame, but the like.

F

But the thing is that requestanimationframe can be. You know if you have other cat long running tasks, the requestanimationframe could be delayed. So if you're checking time stamps your.

H

You know the right yes in exactly well. The current state we have to do is you, you have to observe the how frequent you know. Requestanimationframe is happening and you have to collecting a basic packet of Frank prime frame right.

H

What am I think we are changing that to match what the you know, all other browsers doing right now, there's work getting done now, but it would mean, though, that I mean one of the reasons.

H

We have an objection, this the model that all other browsers implementing is precisely due to the lack of this timing equation right, if we're doing it after the brows had already done the work, then the timing on which to do an animation frame is called, is almost irregular and you're almost up man right, even if the frame was 60 mirror-like may seconds the.

H

If there's a browser work that to update the layout or style anything, then the time at which the best animation frame falls is no happening. Every 16 milliseconds, it could be. You know like after two millions or like in interval, could be like a team, a transitory, mere seconds right, depending on how all I can maybe less than that, depending how how much work the brows did or didn't do so. I totally agree that there we need to provide this information like that use case. I agree. What I'm saying is the API?

H

The I would much prefer if it was a callback like a recurrence relation friend, you know, in fact we could even other argument the existing request, animation frame and tell told you no developers how much time, what all the time at which that request animation frame Accola is full basically and.

F

So if I may try to paraphrase that slightly don't object to providing this information, but he only wanted to be available inside of a requestanimationframe callback. That's right, hidden, be able to dynamically request it at any time. That's right! Yeah,.

B

I mean I, think you saying that it can. It should either be involved. It should be either in requestanimationframe as a callback I'm, sorry as an argument, or it should be accessible.

F

Okay, so a couple of things, one is so. It sounds like what you're suggesting is that the developer the web developers should have should we should require them to do their own bookkeeping like we shouldn't provide any assistance in this. They should be. Ticket am checking timestamps and looking at these time, signals from request, animation, I'm.

L

F

Saying- and the second thing I would say, is I can't understand why this would only be available from requestanimationframe if nothing else you cash it inside of the rendering engine and then just return it every time we ask for it, but what is so special about requestanimationframe that the information should only be available, then, because that's not what it's useful when it's useful is when you are in regular timers inner set interval set timeout whatever, because that's when you're you're not tied to the rendering pipeline at all, you have no bearings right.

F

You know no frame of reference, um so I it's hard to understand why this information couldn't be it I would other times of why it shouldn't be denied to develop other times, because we kind of wanted this information this. But if you could deliver any requestanimationframe.

H

Why can't you? Because that would mean that you know I'm gonna have to paint in our main thread every single frame. That's the latest information we move and I want to do that for power reasons, I don't follow you maybe eat. Let's say the only time to see you know the the system will tell us one on the next frame is when we have like, when we requested to paint right when we race or call back to paint okay.

M

B

Do follow up offline misery, so Kato got a better understand their limitation and the order it is proposed, and meanwhile, I was warming up for the other browsers what they were not like. The thoughts.

K

I had one quick question, which is this is until the beginning of the rendering steps and I actually liked this as an API, but there is also an amount of time from when the rendering sub start to when we would like to snap the UI thread to actually be able to render in time for the ex vsync, and that time is currently not surfaced. Through this time have we considered whether that can be surfaced? And yes, Rios ki, understand that that may not be possible certain architectures I'm just asking.

K

Actually they said because you repeat what what precise, which time you're talking about so right now. This time is the time until the rendering steps begin. There is also a planned vsync time, which is sometime in the future. After the rendering steps begin and I guess my core question: is there also be a budget exposed for that, so that the render is ups themselves can run fast?.

F

Enough or be right, my understanding is that it's not actually possible to observe the time of these. Think that the part the display Hardware samples, the buffers, but we don't actually know what happens that we we. Basically we run the GPU process at you know we can detect what the refresh rate is, but we don't exactly know what it gets samples. So we simply try to match the stride of the hardware's refresh rate.

F

In other words, I. Don't think that information is available, we can so there's been discussion about getting more, basically giving it how long we spend in the rendering pipeline and try to provide a timestamp of when we finished all of the hard work in in redmond rasterizing, so that we have a an estimate of how long it took to for rendering to run and like when we actually set up the pixels to be sampled by the by the display.

F

However, there is some concern about fingerprinting and other privacy concerns like if you can detect how long it takes to run the rendering steps you may be able to figure out what kind of hardware you like I.

B

Guess the short answer, dart is that it's a separate use case and we are like nuns. We've done some preliminary investigation and we want to have a definitive answer for you know what are like whether we, some even better something, can be exposed. Yeah.

K

I know they're not saying this is bad I'm, just asking the question if we thought about it yet I'm, okay, with this being only a step along the path to you know where we end up. There's no yeah.

F

Well so, yes, the answer is yes, we thought about this, so maybe we so really kind of find the conversation about a requestanimationframe. So if there's no objection, maybe we want to march right into the next time. Yeah I think is.

A

Good Joey do that: timing, wise I'm, okay, yeah, sorry I was on mute timing. Wise I think that we want to reserve Nicolas for the user timing and performance timeline. Discussions would twenty-minute be enough.

A

L

A

So let's have ten more minutes on scheduling so.

B

Before we move on from anyone else who hasn't spoken.

B

Ok, let's move on the.

F

Next one is window, dot, request, closed animation frame, so, as Ryu skated was saying what kid currently is.

F

It hits differently from the other browsers right now, in that requestanimationframe is scheduled to run as soon as possible after a vsync, meaning that it gives you the maximum possible amount of runway to prepare for the next frame.

F

The other browsers do it differently. The other processors will do the requestanimationframe signal immediately before we're going to update the rendering pipeline. This is pretty observable there. Some wpte tests that can expose the difference. Behavior here, I think that there are.

F

There are good arguments on both sides of schedule, request, summation frame, one way or the other, so this proposal for request, post animation frame is basically to say why can't we have both so requestanimationframe expected to run sort of synchronously and contiguously immediately before the rendering pipeline, without any opportunity for any other JavaScript to run between requestanimationframe callbacks and in the rendering pipeline? And what we're saying is we would like to also have a slot to run code immediately after the rendering pipeline is finished, without the opportunity for any other JavaScript to run beforehand.

F

Use cases first of all, the one that your ears could described, which is have the maximum possible time to prepare for the next frame and then there's also another concern that we've heard from developers, which is they they really want to avoid forced layouts. They really want to avoid call in doing Reeboks. I may have information of a long time, and they don't know how they don't have the way to wait to guarantee that they won't do that, and this is a way to guarantee that you won't force a layout.

F

If you are the first, if you have a full position in there, post animation frame, slot you're, guaranteed layout is clean and you can read back without forcing them out, probably as much as I need to say about that.

H

So this is basically adding like current versions. The requestanimationframe asleep like Rick, was also animation played. That's right, I.

H

Guess it's a bit confusing I guess the one possible way we could do is also I can make it as an option right instead of like I, think a Cabrini. You.

F

Go boy, nice, so another way, another proposal was to change the signature requestanimationframe to take two callbacks, one that happens just before rendering in one that happens just after so I would avoid any a new method, enemy that actually may be more intuitive. It wasn't clear to me not.

H

Even two missiles, but rather like just have an option. You know option dictionary where you just say. Okay, give me call me at the end kind of.

F

Economically I'm not sure that that's better than to call back slots, but either way would probably be okay. My one reservation is a requestanimationframe has high usage rate. It's not clear to me. Currently, it doesn't accept a second parameter. It's not clear to me whether there'll be some fallout anticipated fallout from changing the signature. Yeah.

A

And other than your color of the shed does the does that make sense from like, from an implementation point of view, for the various browser folks in the room.

C

I'm not sure if you can hear me but I think the Vixens, so there's also a bug in the w3c but Zilla from 2015 we're all you find a very similar bug. He also suggested I think a second argument, which is a often stationary and proposed names, were in sync or whats rendering but I can get to that. But in the minutes yeah I think it make sense. Oh yeah, so a brief description of what we do with the Firefox. We get to be synched as soon as we get the decent.

C

We start running requestanimationframe callbacks. Then we do the rendering set and then the new thing will fire.

F

It may actually be the case. It's have to check, maybe actually be the case that we are doing the same thing and in the sense that we we trigger requestanimationframe callbacks immediately when we get the vsync signal. The difference between us wicked is that wicked that will then run other JavaScript callbacks after requestanimationframe and before running the rendering pipeline. um You know, I should go back and second-guess a lot of things.

F

I said, because, because of the fact that I, don't believe in actual vsync is observable, this is more the like against a well that we get back to when the GPU process, so even in calling it vsync is a bit of misnomer, but it's it's some signal that we get from. You know the GPU.

C

Yeah for for Sonic, it's the simulating thing if I was at certain times that that's what we call it is. They say.

A

F

I would say this is like if you go back, and you know: tech, blogs and fog, trackers and everywhere, you'll find a large litany of people requesting this feature long long, desired generators.

A

Todd any opinions on urine I.

K

Definitely see the need for this API I constantly see scenarios where frame rates could be improved with this API just.

H

So I look at his Bohemians this right.

B

Sorry, can you say that again.

H

Like to to sort of echo what the polls say, that's precisely why we requestanimationframe interpretation will get is does this it, which is you know, call after day, because that's that way you have the biggest. You know amount of regret from the x-ray. That's.

B

Right but I think there's also obviously a separate use case for request, animation and the other time into rights of light. Yeah yeah right.

H

Probably I think I'll, probably the option dictionary makes Muslims I. Think the to respond to your audio concern. I think usually illegal I experiment to function doesn't result in any issues, and the nice thing about that is. You can also check feature check very easily. You just called the customization frame documents and it will tell you whether function and supported mean global you're, not adding a new function.

H

It is okay, but I think if we added that you know just 42-mile, namespace and I think the onion to call box my mixes in a case where you want to be both, but then I think there are cases why you want to just use one on a data right and it's just a bit awkward having personal one, fine or whatever to data callback. So I think mining option is probably the cleanest API. For my mac shading standpoint.

B

So yeah it sounds like there are some senses, at least on that first use case um Stephan. Did you want to talk a bit about the the second use case of clean reads and what people think about that? Oh.

F

It doesn't seem to be controversial, but certainly we've we've heard this again, like a state matter of people. You know we've, we are always telling developers, don't force layout, don't force layout and a lot of them complain that they try not to, and they can't help it sometimes, especially when you have third-party libraries that are doing scheduling their own work. He just is just impossible to know when layouts go be clean, I mean.

H

Not to mention different browsers trigger they on a different API, so like it and money yeah, that's totally fine one browser which I got totally in sync: they are inaudible, I mean yeah.

A

Yeah, so, regarding the the third bullet point that touches on scheduling it, will it be possible to cover it in the remaining two minutes or I.

B

Mean I guess we can find on that main thing made up date. I wanted to get there is that these are the three specific low-level ideas here for seeing we already talked about is input pending in a vast role, and then these are the other two concrete ones that are in the pipeline right now. The second thing we are exploring as a proof of concept is the high level scheduling, ABI and thought who's on. The call isn't actively working on this, and maybe in a few weeks or in the next design, called we loved.

B

You have a chance to present kind of an early proposal for that. What.

F

Another thing I: we didn't get it ready to talk to you about today, but I'd like to wrap my really at least, which is that we're considering doing dynamic frame rates are linking chromium, as you described that you're doing in Safari.

F

There's also thought up to our API, where developer can choose a maximum refresh rate that they want to run out under the assumption that, for example, a game developer might decide that a reliable 30 Hertz is a better experience than a janky 60 Hertz. So we should talk about this now, but I just want to see the country.

C

C

So I just thought: I showed you the explainer for the first time: accessibility, testing and 25, two bucks and the issue is that don't place to have follow-up discussions?

C

F

Think it was a sort of personal repository, so I'm not sure it's gonna survive long, but it's such a preliminary conversation that I think it's okay, okay, well.

H

I should tell me: can we open a new you like well for.

A

Summer, so ideally this would be incubated somewhere rather than in the working group. But that's.

H

Fine I, just don't want like conversations to be happening and like I mean because they, you know these things come into play and you know like later in a discussion, it'll be if we could Disney some official WCC location to you know that Macy's is fine as you: okay, yeah, either WIC or.

F

K

K

A

H

Yeah- and it please said, send out notifications, amazing, so something to the whole know what why not.

A

Okay, so with that Nicolas, shall we move on to user timing, Pelphrey sure.

H

Can I ask about the main threats? Could anything a bit sure well intervene disappear? One thing we a bit confused about is like what is the API that's getting like polls here? Is it the new task? Queue thing is that the one yeah.

B

Right, so that's the one that's done up in progress. We should hopefully have a actual proposal for you to look at soon, maybe in the next design, before the next design call. Okay,.

H

So then I guess they intend to implement is just the process I and see how it goes. Is that the idea to.

B

Implement was just for them rather effort, if you don't know which things will like a lot of. These are going to go in order to trial and then change drastically. So it was just sort of the you know, larger effort, and if it includes the these ideas, but also includes the high level API, but it will also have its own intent.

H

Okay, uh is it also possible to this to move this till I know some ycz group or exempt some like The Oprah Show WPC place, so that we can start targeting issues and stuff like questions by sure.

B

That make sense you are yeah.

H

L

I, just posted a link to the current death spec.

L

Mr. time, al3 was I think presented a while back, that's not the prettiest effect or the one before than one but I just wanted to add. Everyone know that the spec changes have landed. I can give a brief recap of the changes for both mark and measure. We have a new field which is of type any it's basically to append metadata. It's called a detail in addition to that, we can now mark or measure using arbitrary timestamps, so we can now set the start time to any done.

L

High res timestamp, now, based on feedback, we added a method measure with options, instead of just putting all the new aki capabilities inside the old measure method, because they would become way too complex, because, right now a measure already receives three parameters: the measure name, the start mark the end mark, but with the new API, we want to allow the start and end to be done. Iris, timestamps, instead of strings, so if it can be one or the other, and we also want all out the detail or we can have a duration.

L

So, for example, you can have a start time and you can measure of it from the start time to a certain end time, but you don't have to give the end time. You just specify the duration of your measure.

L

So, instead of adding all that to the current measure method, we went with a separate measure with options method which receives the measure name separately and then all the other parameters go inside a dictionary which makes it simpler to call, and also to add, more information in the future if needed, the final change of user timing.

L

All three is that now these met, these methods are no longer void, so, instead of being void, and then you have to query the performance marker performance measure for from performance, docket entries, you can now just use the return value of the method, so those are the main changes out. One additional change is that we allow performance mark to be constructed. This is to be able to support a use case from I. Think some service worker I don't know if.

L

Yes, when Kelly's service worker proposal- and so basically this would allow you to create the object without it being added to the timeline. Thanks for that use case, you don't really add once to add it to the timeline if they want to use a mark as a input to the method, they they just want a mark instructor instead of performance that mark, because it basically with that unnecessary memory.

L

So that is a brief summary of the changes, so the the spec I think is pretty much all the changes are already in the draft. So I would like to hear if there's any feedback from other browser vendors about any of what I just said.

H

So do we really need a new method with a measure of adoption? Why can't we just override over all the arguments.

L

Wait I mean it's not a necessity is it's. We got feedback that it would be clear to have a new method, because if we do not, then basically the method would become measure which receives measure name. Then a dumb string or performance measure options, which is the name of the dictionary start, start mic or measure options. I guess that would would have to rename the family tree and then so. We.

H

Should put no concerns as to how completely was audible of like right, like that, is of no concern what we should concern about, how the developer we use these functions. Yeah.

L

We actually I I, asked in one of the issues and I think was Nick or Charles who were actually saying that they thought it would be better for developers if it was this way. So it's not like we're concerned about the idea of complexities, it's more about what is more economic for developers. um If we want to.

N

Separate Oh, what so I was gonna say that I tend to agree with Rio skate. I actually saw that too, a few days after it had been discussed. Those will delay looking at it, but my reaction was was measured with options seems a bit awkward and presumably future it'll be the more common, the more common to use that then the old way with start and end name, and so it feels like we should be optimizing for that being the standard going forward and make it a simple name.

L

Once I mean I agree with that, we can't just change the name of the older one, because then it wouldn't be backwards, compatible right, sure.

N

I I agree: I guess. My point is I think if even if it is a complex idea, if it is possible to disambiguate the cases in the idea, then my preference would be that just keep the name mark. Orcs are key to name measure and if the second option is a dictionary, then we assume it's in the new mode. If the second option is a string, we fall back to the old blood Nicholas.

A

As far as compatibility goes, there aren't any compact concerns with the we D overriding right.

O

We did this research a while back and did determine that there were no concerns. Okay, yeah.

N

I think the only concern was I do and then, and then there and I think Nick or Charlie also voiced their opinion that they thought it was a more developer-friendly option. So I guess that could be a matter of preference. But maybe one day.

A

I made you can try to gather more preferences on the issue so and one.

H

Idea- which is that, because everything can become a string, it would be slightly backwards incompatible in that, if there was a somebody who was passing dictionary, you would be currently competitive, yeah.

N

Researching Paladin didn't happen. Nobody was doing that yeah. Ok,.

H

Ok, great so yeah, why don't we just go with a signal method.

L

Then um honey can't we shouldn't just ignore the feedback from the other people. Would you like been out of the car right, yeah Charlie.

A

P

Yeah, it might have been me who, who had a comment and a github thread. I thought it would be easier to get right from implementation and from a developer's perspective might be my comment: I stand by that, but I, don't if it's possible in aid IDL that we can get it right using the existing method.

Q

I wouldn't object to that at all, I mean from it from user perspective. If I, you have to remember the Jerry's measure address always option. I can see myself being conscious all the time by the way.

L

Well, you, you will get an error when you try to measure with and I will.

Q

Remember your name.

O

By using a single method, yeah.

A

Yeah, maybe we can take an action item to try to gather more data or more feedback from actual developers as to what they would find more confusing, maybe through the scientific means of a Twitter poll. I, don't know, but.

Q

So we haven't repeat your first drafts for l3 for use at I mean one way we could get more attention. It could be referred raft as it is today and highlight these Croatian in the draft hotel.

A

That would give a good place to like I. Don't know if publishing is absolutely required, but having that as a question in the draft can help to point people who want to read more about it and get a better understanding. So.

H

You can have the issue number four that whether this was discussed 48.

N

Thanks yeah I, actually don't think a Twitter poll is about of an idea.

A

Yeah I, don't know just try to get, or you know, as a funnel, to get more opinionated people on the issue to try to understand. You know the opinion of multiple yep and Kristen's folks.

P

Yeah but my comment: if I can just have a second here, my comment was probably based on the nightmares that we've run into instrumenting, add eventlistener and how that's that used to take. You know different number of parameters, and now it's been normalized and you know there's footnotes and mdn articles where that's what I want to avoid is like Oh long ago. It could have been used this way and we do instrument edit baluster but like if, for example, we needed an instrument marker measure, we wouldn't in the future were instrumenting other people's code.

P

We wouldn't know because of the object coercion we wouldn't know which flavor they were using from a JavaScript perspective. Okay,.

A

So so you're saying that having that complex IDL is not an extensible web kind of thing like it's something that you would have a hard time, reimplemented in JavaScript. What I'm.

P

Saying is if a library were to one day want to instrument the mark method, he would not know which flavor of the development, which version the developer meant. The second argue. The second argument was an object. That's like incredibly easy right. You just called PI Paul will tell you no, but if you, if it was an object, you it gets call it. You get to strained.

H

You call to string on it. No, no! That's precisely. The ambiguity was talking about earlier. What was slight income, but I beat you that bug you have. If you get the object, you have to trade as an object and then process its object right. So.

P

Maybe it's moot if, if the Google team is proved that nobody does, that.

O

Girl that data I- don't remember us off my head, but it was very, very low usage of any.

A

Okay, so I suggest, let's take it to the ich like. Let's go back to that discussion, maybe add a note in the spec and then we can, you know, get more eyes on it and flush it out.

G

Can I interject so philippe suggested also publishing first draft: should we publish her first draft of level three and full of questions about is in that draft? Should we maintain the curve measure with options or should be reverted back to measure and then a denotes suggesting measure with options which which do we actually want to publish I.

O

Think publishing just measure given that there's a reasonable amount of consensus in this group is not full consensus, makes sense. I.

H

Mean even this I'm sure she seems like a single method since for approval.

G

Okay, so I'm writing down any I here, which is I, guess Nicolas, updates the spec to is measured the ideal and then to leave that on blocks us to publish the first working draft of l3. Yeah, yeah and I will note.

Q

That I will provision recommendation for l2 next week, so I can deposit assay time.

L

H

Option doesn't also takes start off in anymore. Was there any reason, not just the.

L

The start, the start can be a dumb string, so there would be the mug. Maybe it should be renamed to just start at start time is the term string or hat on iris time stamps that would be Mike? Oh, oh, oh, yes, sorry, I didn't see that.

H

Yeah, that might be a bit convenient, solid start time.

L

In that case spots yeah, it could be renamed to just start yeah.

H

Now the question is: what happens if you specify start and end duration in the consist inconsistent, and what do you mean by inconsistent like so? Presumably delation is n.

L

Minus stariha every time some specified. So if you specify a start, duration and end, you get a, we throw an error, you shouldn't be specifying all three of them.

H

L

Okay, I guess I was going to ask about the buffered flag and said we're running out of time yeah.

L

Could we take that in the next week sure.

A

Yeah and similarly for in progress, resource requests.

L

Design question but I guess just painted Tom to that site and really Sookie already replied something but Ben to talk to. Please take a look. We can chat in the issue. That's fine. I just wanted to paint that so.

K

L

Readers timing, one three: seven.

K

Okay, I'll take a look: okay.

A

Okay, so with that we're out of time, so I'll see y'all on March, 8 and yeah, and that the car after that will be just a week after that to shorten the gaps a bit okay. Thank you very much.

A

I