Internet Engineering Task Force Interim Meetings, 20 Oct 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF-TAPS-20211020-1500

Description

TAPS meeting session at IETF
2021/10/20 1500

https://datatracker.ietf.org/meeting//proceedings/

A

Hello, I'm here it is my record. Yes, it's using the red device. Okay,.

B

How are you oh.

A

Doing all right not going.

B

On but I got coffee now so I had intended to make myself coffee and then I got distracted and ended up with water, probably better. For me.

C

uh Oh, I see sean is on his way in yep.

C

Don't know who these jabber users are.

A

I don't know either, but my account used to be basically stuck in the tub's driver room because they, I think they uh hi they shot. They shut down the matrix instance that they had set up, but then I couldn't log out or something but brought it back on.

A

C

D

A

In you can never.

C

A

Yeah something like that, and that was annoying because I tried to join again with the same name. It resulted in a name coalition and.

E

A

I couldn't join so that's why I was annoyed, but it's also weird to see your own name there and be like no, but I'm not joined with a different client.

C

Yeah I had I had uh several weeks of problems with webex teams, along that where, for certain other users, uh it was behaving very unpredictably, but not for others, and um everybody was willing to point to somebody else to say that it was their problem to fix.

B

Hi, michael yay, hello, hello, anna, I think, is she on. I think she's on route uh looks like she's hi, everyone, hello, hi,.

B

Hello, hello, hello, hey martin,.

C

I um dropped a few things into the notes page just uh oh yeah,.

A

F

A

Links would be handy.

C

For folks who don't have the github github windows open.

B

Of course, not locked in.

C

So uh before we get into the taps agenda, I just want to say that the uh the response for the hot rc has been somewhat underwhelming. So, if anybody's got anything that they want to talk about, yeah there's room on the agenda for lightning talks.

C

Yes, brian has his hand up.

G

C

G

Yeah, I'm muted yeah. When is rx.

C

Is hot about f it's! uh It is uh after the plenary the wednesday before itf week that the the secretariat's going to do like a virtual reception concurrently, we're doing a different format. We're going to do it in gather with videos and then have the presenter be like available.

C

Next, it's going to be like a poster session, with a sort of a chat space next to a a a video display thing, and so that people can asynchronously watch the video and then hang out and chat with other folks afterwards. So.

C

The secretary got a lot of gather technical expertise, and so they had a bunch of ideas on how to things to try. So it's kind of an experiment.

A

That'll be interesting.

A

Yeah, it sounds like a lot of fun. Unfortunately, I'm traveling that day, because I'm.

E

A

To the better times.

C

Oh, I see foreign.

C

Yeah, it starts at like 6 a.m. Us west right.

G

C

H

C

I'm sorry my commitment just isn't how.

G

uh Wow, how far east are you coming uh eastern time? Okay, yeah that'll be like there's a big difference between four and seven.

A

G

A

That wasn't obvious a month ago, right.

G

Iceland, we should hold all meetings in iceland all meeting in iceland.

I

I

So what we're actually going to try to do is make it. So if we ever have another online meeting again, which is which is not clear um that we would um make it have only one painful thing for time zones. So there would be one like kind of breakfasty one and one dinnery one and then one in the middle of the night for the three big regions rather than having two.

C

A

C

Over the course of the day, so you basically have a session like no.

I

No, no! No! No! If there are three remote meetings in a year yeah, um only one of them will be in the wee hours, uh and that means.

A

I

For any, given I mean obviously like with in some time zone it's awful, but for the regions um there would be one into the wee hours and then one that starts like around breakfast-ish and then one that starts about dinnerish uh over those over the rotation.

G

That seems reasonable.

C

Yeah, I kind of thought that's what we were doing by following the three continents.

I

Kind of in our like no we're starting we're starting in the afternoon local time, which is kind.

G

Of in between so um yeah, so it adds to pain. So, like you have one really good one and two like one really good one. One two.

I

Three one and one hell: no.

G

Right like so that's there's.

I

One where you wake up super early and one where you go to bed super late um right and then one that's right in the sweet spot.

G

Or like alternately, there's one that you basically just say, I will check the mailing list in the morning um and hope that my fellow um participants are. um You know in.

C

The right time zone fight, yes exactly and.

G

In the right time zone, yes, yeah.

I

Not just initial.

G

G

We have a quorum. I should point out speaking of time zone problems. This um fits, I mean like yep. This is this is here because it's usable for europe and the west coast, and I had something scheduled over five to five thirty, so I will know 5 30 to six so at x, 30. So in in 22 minutes. I need to leave to go. Do a manager thing um and then I will come back so.

C

Okay, so you think you think it's not going to take the rest of the slot.

G

No, no! It's it's like it's google, so everything happens in 30 minute increments.

C

Okay, uh I see gory on the list, but I'm not sure I've heard him uh gory are you online and able to gary was having.

J

B

Cameras but he's okay, all right, yeah.

C

Well, I think the first order of business is a banjo performance. Yes, yeah.

G

J

um That really will stall the process. So I'm not doing that one, I'm more interested in progressing this drafts, our important work, we're doing.

C

uh Where are your priorities? Corey? um uh Okay, I don't think I've got it together to kick things off. Maybe would you want to um get things going.

A

Sure but I uh yeah welcome people to the tabs interim, so our agenda says we have issues and prs and uh we also recently got a review from sean and I was wondering if maybe we want to start off with that one, because I'm not sure if we created issues or prs for that.

G

We have not created issues or pr's for that, but I noticed that sean is here. Yeah.

K

G

Yeah, thank you sean. Yes, thank you. Sean for showing up um are there any like. I, I would actually propose that we start off and and put sean on the spot and say. Are there any any things that came up in your review that you believe need discussion like I've, I've read it and I think that yeah I mean yeah yeah yeah, yeah, yeah yeah right, like I there's there's nothing that I would want to come to you as a reviewer and say you're wrong right um right.

L

So I mean I I so part of I I gotta remember to delay myself before I start speaking and meet echo, but um the way I reproached this review was two parts. The preamble was basically like to tell a lot of other people to relax. This is an api document. We've done these before um this is just like. Other ones know this review.

L

Isn't that different and then kind of lay out my methodology of what I think I was looking for in the context of robert asking for a rewrite or moving the deck chairs around right in the various documents, and in reading the document I was like yeah okay, I mean I'll, be honest right when I first start reading a document. The first thing I do is I look at the authors and I'm like. Are these reasonable people right?

L

The answer is yes, I've been to a couple of these working group interims and been like well they're diligently going through all these things. So, basically the answer is no all these are. They were mostly all knits and mostly just kind of waiting for the rewrite to be like yep. All the places ended up in the right spot right. It's an early review, there's going to be some moving around we'll do another review later. I didn't see anything that was like.

L

Oh my gosh, you guys need to you, know right the ship and make it go this way. It all seemed good to me.

C

M

C

Role that we had with uh by wanting the early review was to you know, catch any show stoppers and it sounds like there aren't any. So that's great to hear so.

L

So do you want me to submit you know like prs for like the nit noid things, or do you want to go through them first and say like no silly man, you didn't read this right, that's right, you know or and tell me, like hey, you know, nine nine eleven.

G

And twelve.

L

G

Offering to submit pr's for a direct review. The answer is hell. Yes, sir. um Well.

L

Do you wanna go through them first to tell me which ones I shouldn't waste my time on or.

G

So I've taken a I've taken, a quick pass over this and- and I did not like none of them tripped my you're, holding it wrong detector. So, okay, yeah, like you, could even just give us one pr with all of the nits and all.

L

Right, I can do that. I need to get my github foo up.

G

L

G

Yeah, um thank you for setting a bar for all future directorate. Reviewers um uh miri is not here, so I can resign from the transport directory.

J

I don't think we accept resignations. Brian.

G

G

I thought it was out.

N

But I would be curious to hear what your feelings were about the structural questions when reading the documents, because I guess.

O

That's something that will come up for discussion.

L

I mean so, to be honest, I read all three of the drafts that are, you know currently outstanding trying to like look for things. So I don't know it's just my thinking of what would I be looking for? Well, it's not in here. Maybe it's over there and they're all kind of interrelated and it says, go read the other documents anyways.

L

So I'm maybe a little more lazy, fair than robert is about where the stuff goes. I don't want to get put in the middle of you know, trying to fight the battle for the working group that you know. We shouldn't make the change, but I kind of thought it was kind of all right, but um I'm willing to see where robert maybe thinks some things need to be moved around, but I think for if I was an author, I would be pretty pretty um asking some pretty pointy questions about.

L

What do you think needs to move where? Because otherwise, you know you want to avoid playing bring me a rock right where you guys are like guessing like what do you want me to change? And then it's like? Well, that's not enough or you should move that back and that's just a waste of time. um So.

G

We talked a little bit about this last at the last interim uh when we were sort of like filing some, some some bugs on things and the.

G

Let me put this in a way so as not to cast aspersions on anyone.

L

um You're not recording this right, no.

G

It's not recorded, um but you know I mean, like everybody, has memories, um so we're recording this. A lot of robert's comments were basically document set usability right like so, if I'm coming to this, like never having seen it, which robert was coming to this, never having seen it, you know, is this set of documents, usable and I think here's the part where we have aspersions.

G

I think that the documents in their current arrangement have a great deal of history of how we got where we're coming embedded into them right like arch. First, you know we're going to talk about what it is we want to do then interface, okay, we're going to talk is about how we want to do it and then implementation now we're going to actually see if the thing that we designed makes any sense right.

G

I mean it's like that was sort of like the three stages of this project and you can see um you can see which ones we did first based on sort of like the the uh the um maturity of each of the documents.

G

Is this a really super usable model for people who are to the documents and trying to understand what we're talking about? Probably um it's, okay. Could we do better yeah? um Is this set of documents more usable than the average set of three rfcs of this, um of sort of this level of complexity?.

G

If I can't say anything nice, then I shouldn't say anything at all, but I think yes right.

L

I mean so here's the deal right. Those are directorate comments, the ad may or may not take them up. The only reason why roberts might get carry a little more weight was that he was an a.d for a while. Yes, um the other idea is just to go back to him and be like all right. We talked about this we're trying to figure out what it is. Specifically, you think needs to be moved. Can you help us?

L

Otherwise we are going to spend you know months, trying to figure that out and just get them on, get them on the horn and get them to one of these meetings or have an author and him schedule.

G

L

And go through it.

G

Well, sean did it so come on over.

L

Well, I mean you know I want to get the. I want to get the documents published in a reasonable set of time, because it looks like you're about done so, like let's figure out how to do that.

G

And we have been doing a really good job of delaying ourselves, so we don't need like external delays, so.

A

J

Yeah yeah, okay, I mean I was going to say that everything brian said was true, but also robert's thing was good. There's no such thing as a useless review and to me uh what this really meant was.

J

um We haven't really read arch again, having read the other things and when I did that- and I saw glaring horrible sentences stuck in the middle of this- that kind of really detracted away and could easily cause robert's comments, because we we did stupid things in the middle of the document by adding one sentence here in one sentence there I think right.

E

J

The best document from his review already.

L

Well, no, so my theory is that you should, you know, take his review and go through it and take it seriously. I'm definitely not saying that we should you should discard it. I just think that whether you have to like move section, two, four and seven to this other document and two four and seven from this other thing over there like it's it. You know some of the review was, you know, move stuff, okay, great!

L

Well, what um I I think, you're right, that if you go through in a re-review with a fresh with a fresh set of eyes, that's one thing.

J

I think that's my.

L

J

As well, I I'd rather not move stuff over the document unless there's a a real, tangible benefit from it.

G

I think that's a good principle and I think we should adopt that principle right like so. We should make moves that make it way more readable.

K

Yeah I mean to some extent, I think this is just a a side effect of the the very issue-focused way we've developed the documents and that people have been focusing on particular issues rather than on readability of the thing as a whole, and you know we just need to take a step back and look at the documents as a whole.

G

We're trying to build an architecture like a standard, and this is what we get.

K

Yeah, but I mean I think things like quick, which did this very very github issue. Focus thing also then took you know, had people doing passes over the whole document, and maybe we we've just missed that step out a bit that much.

G

Well, I noticed.

K

That michael has already done this for.

G

L

So I was going to say that quick did that actually multiple times so I mean I think it. It wouldn't hurt.

G

Great and michael just did that for arch today, so.

J

Yeah- and I tried um earlier this week in a pr to to remove the silly stuff, so I I think we're probably getting there. um I suggest we apply michael's thing and somebody reads afterwards.

G

Yep so michael, do you just want to like everything I think it's sort of the same I've taken a quick look at your review and, like everything, that's clearly editorial, just pr it and and we'll look at the pr and then we'll probably approve it um yeah there isn't anything major at all, yeah that I found, but I'm of course,.

P

Biased there isn't, I mean some things: small stuff.

J

We probably should do the last issue because I picked that out in github, because it's slightly different to the other ones.

J

Yes, one of them's a little bit more of a discussion. Do you want to talk about that? Michael.

P

The last issue- I I don't know- I don't think this is so I think we have more important stuff. I mean if you look at these art art reviews, we didn't really get through all of them. Last time I remember dealing with many that were assigned to me, but I don't think we we ended up discussing each and every one. There are a few that are still undesigned and these are probably more useful to discuss.

L

I mean so I just want to check like so did I win sorry, did I win the bitcoin or not.

Q

Saw that one that was.

R

L

Okay, good! That's what I just wanted to check.

C

Sean you can have all the awarded bitcoins.

L

Okay, so what I'll do is I'll go through mine and put pr's that actually had suggested changes if there are ones that I had that were questions, I can throw them as issues, and you can answer away and knock them down right.

L

C

So since we're going to lose brian uh for a while is, um should we maybe start with uh something that we know uh brian's got opinions on, um so we can move on to things that you.

G

Don't have opinions well, I mean I'm only gonna be gone for like half an hour, um so I would say during that half an hour go ahead and all of the prs that I might disagree with just go ahead and merge them. um I think most of what we have as an assigned issue is either has a pr associated with it or is assigned to me, um because I only got like about an hour and a half today to um uh to put uh pr's to things. So I got like three done.

G

um I actually had a question when I was trying to do a pr for a previous.

G

I um a previous issue from devon uh who's. Also here, hi devin.

G

I am confused about a thing and I would like to figure out where my confusion comes from, and I would like to discuss that. So this is on uh the endpoint. Is this issue? 865? Should endpoint object, identifiers be discoverable? uh Last time we discussed this, we decided. Well, we really don't philosophically we're not um big fans of the idea of uh enumerating the entire list of local endpoints, and then we pointed out that we can enumerate the entire list of local endpoints by looking at um a pre-connection and resolving the pre-connection right.

G

um Let me how does this work? I can ask to share a screen right.

G

Yes and we're going to do a window.

G

So that would be sort of like we'd. Add it. You know down here when we get sort of like the array of local endpoints, the array of remote endpoints from preconnection.resolve.

G

But up here we say a preconnection object using a running view must have both the local endpoint candidates and the remote endpoint candidate specified. So you handed a bunch of candidates and then you ask it for the candidates you already handed it, which seems to be different than enumerating all of the possible local endpoint candidates.

G

So I'm confused about, like we decided that we wanted to resolve this issue by saying: um hey, don't do it, but if you need to do it, you can go and resolve your pre-connection, but be careful right and I was trying to write the text for but be careful and then I realized.

G

As I read it, this is contradictory and I'm trying to figure out how to resolve that contradiction.

M

Between resolve and rendezvous.

K

Right, I mean you, you need to have reason um you need to resolve them before you can rendezvous, but you you start with just the um just a subset of the candidates. Then you resolve to find the rest of them and you you do the exchange with the pier and then once you've done, that you've got rendezvous.

G

So if I've got a if I've got like just a pre-connection that just has a few um properties and I call rendezvous and I haven't given it any local endpoint information to start with, will it just give me all of the things that it could listen on? I I don't. I mean it's about like interface, enumeration right.

K

So so the the the issue is that you need to get the remote. So in order to rendezvous, you need to get the remote endpoints.

K

um So you need to find your local endpoints share them via the server with the peer that which then plugs them in as the remote endpoints to talk to, and then you call rendezvous.

K

G

I have enough to write text here. I probably won't wrote text here during the meeting because I got to go to the other meeting, but okay good.

M

I am unconscious, thank you to confirm you would. If you need to resolve, you would call resolve first and then you would call.

K

Yeah you need to resolve you, you need, you need.

E

K

Local and you need in order to rendezvous, you need both your set of local endpoints and the remotes uh end points, uh and in order to get the remote end points, the remote has to enumerate its local endpoints and then send them to you via the outer band channel, which then become your remote endpoints.

P

S

P

About this kind of thing, anyway, I think for this art art review that says how does an application doing eyes not about candidates? This is uh issue in 9-1-1, but this is time to call him.

K

uh Okay yeah. I thought we had that already, but I will have a look.

K

I guess what we have is not very clear.

G

Good, so next next issue um I'm actually going to go ahead and pop over to my other meeting I'll, be back in about half an hour.

A

Okay, so can we get somebody else to drive our github today.

F

M

It be possible for someone else to at this time. I'm actually also going to have to leave like about an hour, but my screen is not in a state where I want to share it. Yeah I'll, do it. Thank you, michael. Thank you.

Q

You really want to do it. I changed my mind. Oh yes, it says.

A

Yeah that changed my mind, things kind of didn't change. My life. We take concept.

J

Wow, that was great. You just multiplied everyone by an infinite amount; yeah, so that's cool, isn't it yeah.

H

That'll get it finished quickly.

P

Yeah beautiful yeah: what do you want to go through these issues or go through prs? Maybe.

L

M

Yeah it's first.

P

I would say yeah.

M

Q

M

P

Learn, oh yeah, there is one mark discussed that I would love to discuss. Can we.

A

Can you can you make it a little bigger? Please.

P

Yeah, let's discuss the connection to a group, and I mean: do you see this by the way.

Q

Yeah, but it's small, if you could, it is small.

M

If I look at what.

Q

You see, then I see that again or can you can you command a plus.

M

P

Make it bigger is that okay, cool.

A

Yeah, maybe a little bit more, I don't know it would be good for me.

B

How is the full screen.

A

No, I don't think you can see it that way, but maybe two more control plus steps.

Q

A

Yeah one more yeah.

Q

Okay, huge now for me: okay,.

A

Q

Thanks, that's okay! Is it readable now.

P

Yeah, okay, so this vr I'll defend it. It adds some text that lets the passive sides decide about grouping, because now we cannot do that.

P

So you get a couple of connections inbound, uh you listen, you exactly have the connections and um you didn't issue clone, but you want to group them in some way, because maybe you know that they're in the tunnel in the common tunnel- or maybe you know that you know something about the network where they're gonna be across the same shed fossil neck and then we could use all the interesting research that I have done at rfcs that I have written and things.

F

So I I I mean some of them should be grouped right. If you get an incoming and quick stream, it should also be grouped on the passive side.

S

If you got a stream, this is so annoying when I don't want to look at something.

F

I don't even have, I think.

S

That's grouped automatically.

P

Right, yeah and.

R

That's what the original text is.

S

P

And that is already written in the text. So if we look at the next, um it says something about this can happen, but there is no. There is no functionality there to make.

E

P

If I want to make it happen, it hasn't happened before so if it's not a stream, but it's a connection that comes in and I want, I want to do something on it, see it's a web server and it gets separate connections being opened.

P

Then maybe you may want to do something on them. It sends data back.

F

So I'm very vaguely remembering, but we did discuss at some point. You know this is doesn't have to be symmetrical and you can just assume these are single streams on the receiver's side, but you know I probably didn't discuss if it would be valuable to group. The manually is.

P

Yeah I haven't found that discussion.

F

S

We just thought it not necessary.

L

S

Are there any concerns.

F

About enabling that function.

M

M

You know what would the semantic meaning be for cases that wouldn't make sense. um You know when you're doing an outbound connection yeah, we always know what is the right thing to do for cloning or putting you in a group like if you're, a multi-streaming protocol, you're opening up a new stream if you're not you're, just opening up another connection along the same path which could have the chance to be coupled um for an inbound stream?

M

Yes, we can passively add it to the group that makes sense, but I I can see how okay I, as the application know, these could have coupled congestion control, so that's cool. But how do I? How do I know that and what happens if I, for example, say?

M

Oh, I'm going to add these to a group and they're over completely different paths and they have nothing to do with each other or, alternatively, I have two different streams of different quick connections that are each in their own groups passively or actively- and I say oh no add these to the same group and pretend these two things are actually part of the same multi-streaming connection. Even though they're not what does that do to everything else in the group? um What if the connection's already part of another group.

P

The second one is easy: it's just gonna fail. I wrote that here uh adding attempting to add a connection to a group with mismatching transfer properties or when the newly added connection already belongs to a group. Okay, so that says that that doesn't happen. You shouldn't be doing that. If you try to do it, you know you can query and find out that they're in a group already, so this shouldn't be possible.

N

P

N

Could add two quick connections to each other right that would not violate this constraint.

P

N

M

What would it mean.

P

Yeah, that's a quick.

E

P

Everything that's written about grouping is not written as a promise, so you apply your.

E

P

But they may not work out.

I

So well so so, quick, quick connections are a connection group in in tap speak right. So there's no doesn't mean anything to add a quick connection to a group because it already is a group.

N

No, it only becomes a group when you have two streams: two connections over two streams right before that is just one connection. The first connection you set up that go very quick is just the connection. It's not the connection group. Yet.

M

Yes, so so translating martin, I think, it'd be. The question is like what, if you try to add two quick streams that are part of two different quick connections to the same group.

J

And what happens if they have different transport characteristics like they have different rtt's or different path, mtus or something.

E

About it so you're going to check.

J

E

Whatever doesn't match, let's just let it fail, I don't have a problem with that. Failing a lot.

P

Because you know I mean if this is happening possibly later, all kinds of weird mismatches should really be something that I'm getting these connections, and I want to clone them before or group them in some way before, even starting to use them. That's really how it should be.

N

But there's nothing that prevents you. If you have this to do it lately, right.

P

No, no and if it doesn't create a problem, but all these issues of possibly transport not matching, possibly just not being possible to do it, because it's in a different group, because you may have one being reliable one being unreliable.

E

P

Know things that just don't match right, then it should just fail.

I

I'm sorry: what's the use case for adding a adding a connect connecting joining two connections into a group after initiation? Okay, so well, the.

P

Use cases is, I mean, the benefit is, is the same as with grouping in general that you can apply priorities and the way it's done is is using mechanisms that couple the congestion control. So that would be done when you have some reason to assume that they traverse the same bottleneck.

P

For instance, you could have multiple tcp connections being tunneled over the same five top five tuple you could have. uh You could have a situation where you know just by configuration that this path is really the same.

P

um You could have something like in uh like in this chatbot and architection, or see where you use heuristics to measure and identify that, probably they should say, then you can enable a coupling mechanism that puts the congestions all together. The benefit of that is less delay and well less delay, less loss and control over priorities.

F

P

F

Explained it nicely now, I think actually, I'm the question is: why should we expose this to the application and left the decision to the application because it sounds like if you want to have a conditional? Maybe that's like a system parameter where you can say. Please.

P

R

F

The stack is making the decision and just telling the application yeah here, because.

P

Then you cannot use properties but.

F

If you, if the stack, if the stack makes.

Q

F

Just that it exposes priorities within.

P

F

The stack makes a decision to group something and tells the application I grouped those, and now you can assign priorities right.

M

Right it could do you could do it as a passive group. It could say I received two connections. They share the same path. I.

Q

Yeah, that's right.

M

I like that direction a bit better and I'm also concerned about putting this in like the base api spec, because.

M

Instead of extension, because, like I'm not sure how common it is to be able to support this yeah and like, for example like if we just say, hey, I'm going to add the whole api like, we would have this function, which we would say like sorry. We can't actually do anything for you right now. Here's a function that does nothing.

K

Sorry I can see how this works at this, the sender side, I can't see how the receiver can influence these things.

P

It is about influencing how data is sent. It's just.

N

But allowing the system to do it may be a better choice here right, but I guess I guess you can already check if two connections are in the same group, because this you need to be able to do anyway, so yeah.

P

I I I think mirror is completely right. I think that that could just be automatically done when it's possible, and then you apply the priorities in query.

I

P

So then you don't need this.

I

A

I

In any widely used apis today,.

I

Does this capability exist in any widely used apis today.

B

Gory, you have a question.

J

Yeah, I I I I was going to try and press michael to to write texts that allowed him to do his research, but didn't actually put it in the spec.

P

I just found that this is useful functionality and we cannot do it, but I think the way mirror is corrected. We can I mean it can be done. You can just use it whenever the system finds out that it can be done.

F

We have this function already that uh you can get as a passive on the passive side coupled groups anyway, because that's how screen works.

P

I think, then, we need no text at all actually and we can just kill this pr.

P

Yeah I'll just close, that makes sense. I agree.

F

I guess you can check if you find the right piece of text that says you know on the passive side. um Connections can.

L

F

Cloned because they are streams or because the system decided to do so, I I.

P

Exactly that that text exists, it says, for example, when there are streams- and it says questions so that's all right.

F

You can, I mean.

P

Like you can add another example, if you want to make it more explicit.

P

That's a toggle okay next, uh which one.

F

It could be something you actually discuss in the implementation draft.

Q

Yeah, that's true, I could add some text there.

P

Yeah, it's all fine. Thank you, but.

N

I think it is kind of a niche use case for, like the general purpose stack, so in that sense I think as long as it's doable, I'm not sure we need a lot of text on it. I think the the functionality is there. If you want to do it right, so tiny.

P

Bit of text in the implementation graph, but yes, I agree it's a niche thing uh anyway. Let's, let's go on to the next pr that we want to discuss.

J

How about 9 9 40, michael? Because if you get those words right, you might lower what you want.

P

J

Yeah, which is anna's pr.

Q

N

Yeah this book, just fixing two things that I thought was mistakes in gory's update, so.

J

Yeah and one of them comes to be um what uh whether connection groups are at the sender at the data center at the local or what is the context of a connection group? Just give us the right words, and I think we michael is happier.

P

The concept is that it's it's influencing sending behavior right. So at the sender it's fine, uh the the issue is just. It doesn't need to be the active open side.

N

But I I think the sender is not well defined right. This is also kind of a strange notion that this group exists and goes away. When I receive the packet and comes back when I send, I mean either the connections are grouped or not, because you.

E

N

Fact that if you update something on a connection, it takes effect on all the other connections right and you don't necessarily do that when you send so to me, the original text was fine, but it's relating.

S

To the transmission.

P

Of data bi-directional anyway, unless we specifically don't have that I mean we have unidirectional ones, but normally they are bi-directional. So it's strange to distinguish like that.

M

Should it instead say you know, a connection group is a set of connections that shares properties and caches, as seen by kind of like one endpoint right so like it's saying like the connection group is not necessarily a protocol concept that gets sent across the way is a way to manage connections and their relationships from the perspective of one application, not necessarily of application. That's too many words yeah.

N

I think that is correct. I was actually when I did the pull request. I was looking for another word, but I couldn't come up with a good one, because endpoint has a different definition in the context of the draft right. This was actually also the first word that I was thinking just to replace with. So if we have another good word, we can put it but um yeah.

N

It's true that we should maybe point out that it applies to one end, because the connection group doesn't exist at both ends because you group it so so that I that is important, but.

K

Yeah can the two ends have differing views of what the connection groups are. It doesn't.

O

Work they can't in.

K

Some cases are there, restrictions on what you can do.

M

K

M

Are certain cases like a quick connection? If that's a group where you would have the same notion of a group, because it is a multi-streaming protocol and gives you that notion?

M

But if you're just saying I have a group of things that I'm doing coupled congestion control and priority on, but I don't have a way to communicate that to you on the other side, then you may have different notions of what the group is.

K

Yeah, there are not cases where one side decides to couple them in one way, and the other side decides to couple them in a different way and things break yeah.

J

Because it always it's only to do with how your size sees the network path.

Q

K

Yeah and I'm just thinking cases where you've got multi streaming, where they have to agree.

J

Yeah within the same transport, they are going to be the same. Sorry yeah.

N

J

N

Other end may not view them as a connection group because it may not even be a taps connection, but they are still connected.

F

Yeah, but so for different streams. There you have uh on the wire or like in inventing landing. So it's clear for both what the state is right and for everything else. It's really just a local sending decision how you prioritize sending out your packets locally. So I don't see a problem exactly.

J

So honestly can go ahead if we can work out if we want to change up the sender, which is three words into something more concrete or simply remove it. I added it because of a review comment somewhere in the art art review, which indicated that that caused confusion about whether this was signaled end to end.

M

So I think we should just add a sentence and I'll I'm I'm trying to propose some text right now. That will.

J

Tell me, as a proposed text, yeah.

N

Yeah, so if we we should say something like it influences the sending behavior.

J

N

Sending behavior.

J

Not the res anyway, I like tommy compost,.

N

Yeah I I would say that it's maybe not yeah, it doesn't influence the received behavior, but we have to be careful with if it's only descending behavior, because maybe it's also like timeout values and stuff. I'm not sure if that's ascending behavior, that's maybe a connection timeout right. So.

K

Yeah, are we accidentally prohibiting coupled receiver driven congestion control? Do we care.

M

I don't think we should talk about sending or receiving. I think we should talk about. It. Doesn't.

N

Yeah a word that characterizes the endpoint is what we need. Yeah.

J

I think we let tommy compose because I've tried it and it was harder than it looked. Even though I understood what it was yeah.

N

A

N

Got confused by your text because it didn't seem right to me.

P

Do we really, I don't know, do we want to go through them all? Do we take the most important ones here.

J

Because many of these are proposals against which I think we just need to say yes to.

P

Yeah, I can make a general question like if anybody has any issue with any of these editorials small ones, you should see it and if that hasn't been said, then I'll just merge them in.

N

So I had questions on this example right.

P

Oh yeah, that one yeah, then I'm not gonna, mess that one in yes, okay, we can discuss that one a bit, that's a good one.

P

So the issue, I think the main issue yeah, I mean a couple of things right so they've been this place. I wasn't really sure I mean I've written tabs like cold and it was very clear in my code and still I couldn't really understand that example fully, because some things are just.

R

N

Yeah, maybe we need to see the example, it's very hard to look in the pr here and have an opinion. I I mean to even understand this. I had to look you know in in the full yeah.

P

I can make it a tiny bit smaller. It fits fully almost.

P

Yeah, that is the example right local specifier local empire, but then it has the local specifying that this is where it all ends. I think, after this yeah that's already the end of it.

R

P

Yeah I mean so, for instance, you complained about message: data response coming out of the blue and I completely agree like I read it and I tried to make this message context, but where does that come from? But I think there were a few more things where I thought.

N

Well, at least I complained about the inconsistency now right, because somehow you ex now create the message context and I think actually for the received you don't need to create it. Maybe but yeah you, don't you don't so that that's wrong inside the receive event handler right.

N

M

On the received call should be it's a different context. It should be called like received message context and then there's a send message context, and so I think you know if you.

Q

Just move this down object.

P

Q

You fill it and you create a new one. I just put it early okay. I get that, though, just move it down.

N

H

N

Also strange right to create now the the message context, but not the the message: data response.

P

Yeah also, if I move it so.

N

The question is: do we need to create them at the wall or should we just leave it out, or I mean it's not so easy to read about the example actually.

P

Yeah I mean I agree also if I, if I move it down, I mean this is actually by a part of why I kept it up there, because I thought, if I move it down, then it seems that I need to allocate this before I use it for sending, but it makes me wonder what about the one on receiving do I need to create it. Is it like a pointer that I'm getting it's just something? That's handed to me, yeah.

M

That that's a pointer that you're getting.

P

Okay, that's all right! Well, I mean it's okay, but it made me it made me wonder about that. You know having this here and having it allocated below. We can't try that we can just make it below when we send. When we change something about it. I think it becomes a bit clearer now that we have these properties on the message context that yes here, we need to create it and do something about it. So we can do that, but the message data response, so we have to do something with it.

N

Yeah, I think the assumption somehow is that this is not a complete example right. So the message data response is created somewhere else, but it becomes a little bit inconsistent now when we create the message context, but not the message, data response, so that may be confusing for the reader.

A

I mean you could just fill it with some example string. That makes it clear that it's just just a variable I mean we have.

F

We have an example.

A

And the interface draft, where it says message: data equals hello and that is sort of okay. Here's the thing you put in a message.

N

A

It seems like either.

N

We should create both of them or we should not create either of them and just use both of them right, because otherwise it's a bit inconsistent. Why do I create the message contact but not estimated message, data response so either both or none? I would say I agree.

P

I agree the thing is: if we do it here, we have to do it in odd examples as well. It's very strange if, for example, gets a message response or something some data object out of the group from somewhere and the other one doesn't right.

N

Yeah, I don't know how many examples we have that would be affected if we created both of them, I mean otherwise. We just don't create either of them.

P

Here also there's a messenger.

N

But there it again it was for a receive right, so that one should not be there. I think, or what's it or wait.

P

What yeah? Right, because it's just a pointer that we're getting or like an object, that's handed to me yeah! I guess that's! Okay,.

N

I think the event handlers is also a little bit confusing, because sometimes we have the begin and the end and sometimes not. I think.

M

I think we should probably just iterate this on the issue on.

E

M

More um yeah, I don't think it's.

P

Yeah I can, I can try and update and create it. I don't know what create a message. Data response, I don't I don't know how that's supposed to look.

M

I mean what fundamentally, what is the issue really asking for.

Q

About this was about the message context being confusing when it when it is implicit- and I guess.

P

Implicit meant that it's not being allocated, I I had trouble even understanding this. This is why I wrote this. Is this enough? Yeah.

N

I think the confusion was that you have a receive and you get the message context, but then you don't send one. So maybe the question was to make it more.

N

P

I mean we could also not do it because it only says consider doing it and also saying we considered it, but it made all the examples much more complicated and longer.

N

But I think I think the simplest fix would be to just you know, add one setting on the message context and send it or even just send it without allocating it.

N

M

I mean there's also like you know, I'm thinking of this coming from how we give our coding examples of how to use the api at our conferences right and when you have like this is very early in the document, and you have your first example saying: here's how to use the api.

M

That's what people are going to copy or say this is the canonical way of doing it and, having you know like absolutely, we need to be able to set the priority, and if this is a final message, but that's not usually what people should do and if people don't have specific properties for the message, they don't need to send an explicit message context.

M

So it seems like we would just be encouraging kind of a bad habit here.

N

Yeah, I'm fine with leaving it out. This was trying to. I think michaels was trying to resolve the feedback from the review by inserting it right, and I think the insertion did not work. So that was my complaint.

P

All the examples get longer, you know we have to do it in the server we have to do it. In the theory. Example, we have to allocate the method. I.

B

Don't think we.

P

B

P

We can also by saying that we found things made on the examples.

I

So should we have like a later example that that does this maybe somewhere in the message section- I I think maybe we do right.

P

uh Yeah we have, I mean the lines that do it. We have like for adding properties. We have that we have these uh add property thing.

N

P

N

Comment was to have it earlier in the document right, not not saying that it didn't come because, at least as you said, how to set the properties is explained later.

P

Well, it talks about this specific example.

M

Yeah it's talking about this specific example, but I think I just disagree that it should be that early. It's not it's not the best practice. Yeah.

P

Yeah, I think we can just disagree on having that example. I mean he only says consider, so it's not a very strict request. I would say maybe something that would help, but it's a trade-off right because it does make things longer. Oh here we go actually what.

T

About my own, video can.

P

I I can stop my video leave space for others.

T

um So we can't discuss the next, but probably.

J

T

P

J

737 because it should be trivial and I think tommy can just do it because I changed the text. The pr.

J

Yeah, I'm asking tommy to look at that one because, okay, you might be able to do it just just and say yeah that looks fine or maybe I did something stupid in github.

J

Okay, so the context.

N

N

Is this the one that was in the intro to start with.

O

J

So I I I took my pr removed it from the intro, because that really looked stupid. When we came when I came back to it and I put it, I think what tommy suggested.

M

Okay cool, let me check it. Yeah, okay,.

J

M

If it looks like okay, we can just merge it.

J

Yeah right right, okay and probably doesn't require lots of people, probably just you and anna to see. That sounds good. Thank you. Thank you for moving. It.

T

Okay, okay, uh the one that we just discussed. I should probably just close it right.

P

To get things cleaned up a little bit.

M

P

Close this example, and then.

N

Yeah, so maybe you could just see if we could clarify these event handlers while you're editing the example, because now, when I looked at your code that looked inconsistent to me or maybe we should make another issue for that.

G

A

Yeah and maybe briefly explain why you're closing it.

P

Yeah and uh or you mean here, I thought in the issue.

N

Okay, yeah, I guess in the issue.

S

K

You're, increasingly quiet and very hard to hear.

Q

Okay, uh all right here c issue, I'll comment: I'll write in the issue that we, where is it.

Q

C

Yeah, just playing mommy.

P

We decided against this because it made the examples lengthy.

C

There was another point that tommy made that it introduces a level of detail too early in the document by including it.

P

Too much detail.

N

Well, I think it's that it's not the typical way of doing it right. I think that's the main thing that if people copy this code, it's the wrong code to copy.

I

Like a rich example later in the document which which twists all the knobs, um maybe an appendix or something so, people could see examples of using all these mechanisms, but I mean I agree with tommy that, like that that up in at the front of the document, it should be a very bare-boned server, etc.

I

But I mean I don't know what should be in there exactly, but just one that, where you have like everything turned on, were you setting like security properties and you're, using messages and your framers and and all the other good stuff.

P

Abstract and language independent, I don't know.

B

P

Anyway, this one I'm closing.

P

So we can go back to the prs.

P

P

We want to discuss multicast, I saw that colin has reopened it or reopened an issue or something.

P

K

We can discuss it.

P

If you like, I don't know, I don't care, I just I. I won't do some efficiency, it's probably never efficient to begin with multicast.

R

I think there's some anything that somebody is keen on discussing like I was keen on discussing my uh other side.

M

There are some simple ones: can we just go, maybe from the top of some of these ones that brian had and just try to merge them? It's probably here yeah.

P

Okay, the ipv6.

P

Seems, okay, doesn't it go.

R

P

This question about scoping so much.

T

J

Problem with cleanup cleaning up for me is, I still don't know what it means.

P

Clearer than cleaning, then without cleaning up state.

M

Essentially, the the fact that it's cleaning up is an implementation detail. The main point for the application is that you're dropping the connection in an abrupt manner and it's not going to be graceful all.

J

Right yeah, so we just don't need to say anything about cleanup. We just ignore the whole thing after we spent like um 15 minutes one time talking about it. We we all know how it works, but really it's behind the scenes.

P

Yeah, I I think it I mean I I I.

J

P

Prefer the second I mean seeing this without cleaning up state as as a reader. I might think that this is uh not freeing any memory, but it might right so I mean I will. I think it's good as it is every merge.

P

P

Next caching and raising an implementation, intro, eight, eight, three and eight ninety.

P

Why are we one? Do I want to isolate sessions? Okay, but this talking about erasing before you've described what it is.

P

T

N

So this was longer.

P

M

So we're turning into a list well.

M

Q

M

Just breaking this out into a list.

J

Was it the same content.

M

It's adding to the list the fact that you can support candidate protocol racing to select the most optimal stack in each situation and to provide state caching and application control.

M

I think this looks good to.

T

P

Should be capital s.

J

J

I think someone should read it carefully and then just do it, but maybe we don't have to do that. One now.

Q

J

Yeah, because this one just.

N

Came, I think today, right yeah.

M

Yeah, but in in principle yeah we think it's the right direction. Yeah.

J

Just needs some review and it'd be fine.

P

Okay, so then, next small fixtures, together latest arts, update we had that what was what was the conclusion that we agree on this. We should merge it or show oh yeah.

M

P

T

Some text, you did it already a group. I.

P

T

Stay for managing connections within the single application.

R

J

And it's if anna thinks it's ready. I think it's ready.

N

Yeah yeah, that's fine.

M

So if you want to click, commit suggestion, yeah.

J

P

Oh dad yeah me, I should call me that is true. Thank you indeed,.

R

J

Yeah, thank you for helping that little things.

Q

S

J

P

Is it okay? Now, no I'll, do it later small fixes to go address.

T

How to interfaces there was a bit of discussion on that one.

N

N

I think that one needs some more.

P

Right there different ways of doing it with interface or the other way.

I

Well, philip, isn't here to defend himself. So um should we just table this one.

B

I

P

um Yeah this one, I think people agree right and then I'm now brought up that uh there should be a reason.

P

N

Kind of strange right, because the documents are actually not consistent now, even which should they are not consistent, because I I would expect that we followed or should unless we say why not.

P

Yes- and I I mean- I agree that the thing is that I always wanted there to be a reason if we, if we don't have any of these things- and I remember that as we made that decision, I couldn't find a reason, but I think corey puts it exactly right.

J

J

I think that was a good idea. Yeah. My rationale was the thing we missed was actually a soft arrow. It is, and there could be, there could be other soft arrows, so we used to have a soft arrow handler which deals with all these soft errors. Rather than saying here's a new one.

P

We have that I mean everything is really solved. It's just about fixing the mindset appendix.

N

uh Well, do we have a soft error where it fits already or we need to modify slightly how we.

P

We have that already.

J

We have it, but it's not called quite that. I believe.

P

M

Let me see: there's a notification of icmp software yeah.

J

Yeah, so so, if this wasn't icmp software, it simply was a soft error. Then it would. It would cover tcp soft errors, as in.

M

There is a generic soft error event.

R

Currently, uh stairs.

S

Soft error message: arrival yeah.

P

Okay, so that should maybe be you renamed to be called notification of soft errors.

N

And then you can have icmp as the example.

Q

N

And then also I mean I.

P

Support that, but that but hold on, because I think this is just about enabling or disabling, isn't it here, isn't that a selection property.

P

Because we do have the actual just soft error, yeah specifying transfer properties. So this is just a selection property.

J

P

The wrong one did I have these these errors or whether or not so where is.

J

P

Application considers it useful to be informed of it.

N

So, where is the actual.

P

Error is an 831, that's just a soft error. There you go.

T

All right, yeah, yeah, okay, oh well, that's good, because we already did what I wanted exactly so we we have it. It really fits that's what I thought.

P

T

Really does fit, it's everything is covered.

N

So then we just need to take away this. This bigger, I suppose, stinging appendix c, then.

P

Yeah exactly I just need to put the explanation and appendix see that this is covered by this.

N

P

N

And that just says just removes that we don't support it right. Yeah.

P

So I'll keep this open as a reminder to myself.

P

Michael will close this. After finishing the updates.

E

Interface mindset appendix.

N

Right but we still want it to be a a shoot. I guess or.

P

Michael will merge this.

S

I think we do, I think we agree on the subtlety.

N

I mean the reason you changed. It was because of this inconsistency, but maybe it's still a good change.

P

Yeah, I think it probably is- I mean somehow I don't like it if implementations skip too many of these things, because it takes possible functionality away and most can be falling back to not doing anything anyway, but it is very restrictive right I mean, and the tap system is a bit heavy to implement. So I suppose should is probably better in general.

P

If we agree okay, then I'll merge this. After doing the other one.

T

um Remove special values: updating the oh yeah, oh, this is one to discuss. Actually it's just an editorial thing, but it's a bit heavy.

P

um So there was this well, let me show you there was this thing about the special value of an integer. An integer can can take a value like uh undefined or something like that, and um the suggestion was that I would do it with the type enumeration and it appears in multiple places.

P

The way we've defined the types I found that enumeration isn't the right way to do it, because an enumeration in our in our definition here has the same type only but tuples really fit, and all of these things were I well, they seem to be fitting with tuples I mean so. For instance, here I have a tuple that is integer boolean, I say the integer value of this tuple specifies the following.

P

This particular case really does have an additional special value of zero that I think sean complained about the special value and the special value this one now just has one special value uh and then the boolean value of the tuple. If it's false, the integer value will be ignored and the meaning is full coverage. So I have text of that style everywhere. Now.

P

Instead of defining a new special value, for instance, here the boolean value and distort is false: numeric value will be ignored, meaning is disabled.

P

So is that general approach? Does that look okay to people.

M

P

M

Think you would end up looking different in different languages.

M

It's fine for this.

K

Yeah, I I slightly worry: it's gonna confuse people um who are coming to this from languages which, where, where they have enum types and tuples.

K

Either an error or an integer: it's not returning a tuple of an error and an integer.

P

T

Oh, when you period.

F

It seems to constrain this interface more than needed, and I actually find it more confusing than the previous definition.

P

Well, it's very warmer and yeah.

P

But then I mean I see the point of saying that an integer with special value undefined or something like that is unlimited, that's also not beautiful. I mean I agree to that as well.

R

F

It's not beautiful, but I think it's easy enough to understand, or I mean sean is here so the concern.

G

I don't think this was sean's concern. Was it no? No.

P

That was just about the special value of zero on top of having another special value, great.

K

Yeah I mean, I think what we're saying here is that these things either return an integer or they return an error right.

G

So, like there's like newish languages, do this um exception-oriented languages do this through exceptions, um pretty much all over anything in google or that's leaked out of google. You use absile status or right like so. This is a very common problem in in uh well. Yeah control flow, mixing, control flow and um and return so.

K

We're not going to solve it.

G

In tabs, something.

K

Where, if you're, implementing it in rust, you'd return a result and if you're implementing anything hyphen you'd return, you'd return an integer or for an exception right.

P

Why do you talk about returning.

P

I mean this is not.

G

Oh yeah, this isn't a return right. This is a property yeah. This is for adjusting.

P

R

I

Value right, this is an.

G

Input value for everywhere.

P

R

P

Principle though, but it's not primarily about what it returns, it's about how you.

E

Adjust an integer value with.

P

A special case as well great.

G

So this would be yeah.

F

You know writing out. This is an abstract api. You don't everybody would imagine it's different in a different language, so writing out. In words, this is a special value. It seems easier to me.

K

Yeah I mean, I think it's the same principle that what goes in is either a an integer or it's a.

G

Some other thing something.

K

Can be represented.

G

As an integer right.

P

I really don't know, I don't have any strong opinion either way. We did that because, because we asked for it, I feel this is the cleanest way out I can I can make, but we can also say we don't do it.

G

I think this is the cleanest solution to the thing, but I also think we shouldn't do it.

P

Yeah I mean, maybe people can take a longer look at it and make wrap their heads around and think about what they about their opinion and then you can make a decision about this data.

P

I don't mind if it's closed, I don't mind if it's merged, I don't mind. If it's changed, I don't I mean I just did what was asked.

L

Yeah so just to clarify mine was about one particular paragraph just describing to make sure the words were there to talk about the special value.

P

Had a special value of zero.

L

P

Having a special value of something like unlimited or whatever,.

A

P

Particularly confusing.

A

So I just wanted to point out: we have appendix a of our api document, uh titled implementation mapping, where we already talked about sort of mapping, an abstract interface, to a concrete api in a given language, and it already talks about types, for example, for integer. It says it can be represented differently on different systems and in python a function might return none or raise an exception. So we kind of already have part of this, and maybe maybe we can refer to that appendix somewhere, but maybe oh yeah, yeah.

K

Yeah, I mean, I think it's also okay at this point to say you know that this either takes. uh You know one of these or one of these and not actually trying to write it as a type and just say. Well, if you pass in an integer it does this or you pass in the special value of whatever yeah and then let let the people in you know influence it in whatever language do do whatever it is they care, but.

G

Yeah yeah, because we already have.

C

Like the language.

P

That's in a1.

G

And types already kind of covers that as.

P

Do you want to write an or in the type line.

K

R

Or especially, yeah.

K

And yeah, maybe that appendix needs some extra words that says you know there are different ways of implementing these.

P

P

uh Okay! Okay I'll put it on my to do this to change it and do it that way.

T

P

This one that one should be easy. I hope I was just going through these cases of traditional and modern and at top and underneath.

R

K

Have a discussion with the rfc editor about why you should merge it later? I guess.

G

P

One but there's a proposal here on something.

P

Implementations that provide the interface.

N

I think gorilla didn't like the system providing the interface right.

J

Yeah, I didn't.

O

J

um The in yeah.

B

F

B

I just get rid.

N

K

What would you say, implementations of the interface okay.

P

Okay, implementations of the of the interface yeah easy.

J

And it's to disambiguate a little bit the whole system of taps. This is the interface.

N

Yeah, but it's actually the it's the implementation that provide the interface in some sense right, because it's not the interface itself, it's what you do with it afterwards, yeah right so, depending on yeah.

S

I don't know how this is a separate line.

P

Implementations.

P

Q

P

Interface to dynamically, so are you okay with that? If I, if I commit these two and then I then I.

J

Yeah, that's good.

P

Okay, I don't know I need to finish, probably so, let's say commit.

T

T

And then I merge.

P

Let me see if it if it works. Okay, that specifies the interface of the okay implementations.

L

That provide the interface to.

P

Dynamically choose the transport? Yes, okay,.

P

Doing it, oh checking.

R

But somebody's merging in the back.

G

R

G

I would have guessed.

R

G

I'll send another pr in a moment so like that the number will start going up again: okay,.

P

Priority values, so this one this one uh I blame I mean I, I look through history of things of what happened and I blame brian's memory on that one, because we had the discussions going back and forth about priorities, and then there was a conclusion that the higher value is a higher priority and we changed this everywhere. Then you changed it back on this particular line somewhere yeah. So this fixes.

G

It back yeah so fix it, yep! Okay, I don't know. I still think that I I'm still like a fan of sort of like the eunuch's nice way of doing things.

G

Totally agree with the fix there cool brian.

C

Your opinion is always priority. One for me.

G

H

Now we get to multicast.

P

um We're entering mod 6 to fix 876., just explain it be tiny. So you use this for thought that incorporations is probably in vpr.

P

Okay, not much look! Look at it! People.

J

Okay, so I I made two, I made two proportions of examples and tommy, and someone else said they were okay and then colin project. That warrenton was a bit um suspect, so I just removed the extra example to keep the one which wasn't the suspect in any way.

K

Yeah- and I think your fix here is fine- I think the bigger issue is the other. The multicast issue, yeah yeah, okay,.

J

Okay, we can merge them.

P

Five: four three: two.

P

Oh, this one cannot be merged.

P

Was it the one that we was there, something with it? No, we said we right. No, no, that was the system providing right got it merging.

P

um So there's only this one, which we don't discuss today and we can start looking at issues. Okay,.

P

Shall we take all the ones that are not designed to anyone yeah? Is there a way yeah to nobody? That's what we take plenty, but that's all the old stuff. That's.

G

P

One page: that's not too bad yeah, but is it time for quick? This is old stuff from top. Maybe so we have the newest, yes, art review. Oh, that is from me. That's.

J

Well, we just decide who I mean I'd I'll, happily do it, but brian won't want to do it and I'm not going to fight him over it. Go.

G

To take it, corey yeah all right decide that one to me, I just read the.

J

G

J

E

G

Know what my calendar looks like between now and the next interim meeting and like it's going to be another one of these things, where I find two hours right before the meeting to write stuff, and I still have like three open issues that I need to fix. So brian.

J

Do you want to review it.

G

R

P

That one's starting.

T

P

Cory you want to take this so who wants to take this there's a long one.

J

I think the long one needs it needs a discussion thread does it. I can have it, but somebody used to read it and think about it.

K

P

Are they connected or or I'm assigning it to? But if I sign it, then yeah, okay, I'm gonna, assign it to you, but.

G

P

We wanna come, do.

G

We wanna come back and do we wanna all like, maybe take a moment to read it after we've assigned all of these other things, and if we have time at the end, we can come back and talk about it. A little bit.

J

Yeah yeah, I think.

G

J

G

L

J

um This is the last of michael's review comments, but I thought it didn't fit with the next list that the previous one was. It actually has some substance to it.

N

Okay, okay, you split it into.

P

So this one uh okay, does it need this?

P

Okay, I think I mean I was a bit confused as I read through the document, but because of because I saw things like connection message, capitalized early, but I believe now that this is a long discussion for nothing, and it was really just one sentence that that got me stuck because indeed, everything is being defined in that document just later, so I think it's about avoiding capitalization before the definition comes and maybe in the end it's only one sentence that does that.

P

The last item of the list in section 12, which is, I think, a new sentence that corey added, probably which has this inline definition saying what the connection is with capital c. But then it has a capitalized message and that threw me up.

P

I think it's a tiny thing, probably in the end.

J

Yeah that was putting to fix a tiny thing as well, so michael, why don't you do a pr suggest it and I'll I'll review it.

P

I can do it yeah, okay,.

J

That sounds easier than the plan was that's good.

T

Sorry send events to limiting yeah that needs a bit of discussion. It's art art, not me! I do this.

P

Wherever I move my mouse something pops up at the discussion of send events, I worry that the api is closing the door and allowing the api to channel information about what's happening because of the scent that may be available for future transports.

P

Maybe something would be communicated as events from connection instead, I don't know what that is. huh It's a message.

T

From the future, it's a message from.

G

Like so, this, I think is one of those things. It's it's um the wages of us trying to build the most general possible api, even for um transport protocols that don't exist yet is that somebody will come up with, like I think we even have an rfc for like ipo over faster than light right like we can re, I I don't yeah, I'm not sure what to do with this.

J

I think it's a protocol, specific extension and therefore we don't have to document it.

G

Yeah like when it comes seven years of ascend that may be available for you, yeah.

K

I mean, unless someone has an example, yeah.

G

Yeah, it's like you, might care about it's hard to know.

K

G

I was being facetious when I was talking about faster than light, but like.

K

Yeah I mean if someone had an experimental protocol where they were playing around with different ways of doing it and needed some sort of extension. Then we could see if we could generalize that.

I

K

I

Are these like? Is this network feedback yeah? I think so.

K

This may be something more path aware that has more information, but from this we're sort of guessing what they mean.

G

Channel information about what's happening because of ascend.

A

U

A

In response to ascent action.

G

U

G

You send a packet and that causes something to happen.

G

Which causes connection states change in such a way that, even though you know which send.

G

Sent it like, so I think actually, no, this is fine. You can get connection. Events are completely um are completely asynchronous anyway. Right, and it's like the problem here is we're talking about. The only thing that can happen related to ascend is ascend event and, like I think the point here is well, you might actually get some other event. That happens because you tried to send something, but I would think that, like that's, that doesn't need to be bound to that. Send right. That's just a it's an asynchronous thing that we don't care about.

G

The causality of yeah.

P

Exactly I don't think it affects the sound. I think this is asking for something like a message context as a parameter to soft error, but I think we don't need to define it. We can maybe say something that.

I

I'm not, but it's not a connection software. It's a yeah software. Well,.

P

He's awesome yeah: could it be a part of it?

P

Maybe this would be a connection software or even if it's specific to ascend. I mean copy down like that.

N

But I mean you can also add more events right, if you wonder.

A

Wait, why does this need a connection? Soft error? There can be a send error there.

K

A

I mean we have sent expired or send error can be the result of a sent event a sent call, and then you can have a connection event. In addition, I guess.

N

You mean it could be one of the send errors that could be more specific.

A

I mean it just says: send error, I guess and arrows can have specific error codes. Yeah yeah.

P

Yeah, so what do we do? Anybody want to write some text about it that somehow addresses this.

J

I'm not sure this is great transport protocol designed to do this yeah. I.

G

Think this is, I think this is. I think this is closed with.

N

I think this comes in the mapping document for that new protocol.

B

N

I'm writing something.

K

um Restriction on exactly one event in response to each cult ascent is that what they're complaining about.

T

Might be, or it might be, specific errors.

T

I don't really know.

N

But it also complicates things if you can get multiple right.

A

We can actually also uh throw this back to our reviewer.

H

We order we already invited.

A

Him to look at issues or if we have prs yeah, we can reply to that email saying.

K

O

Mean I suppose there might be.

K

Cases where you send a packet and you get a trace of events from different devices doesn't mean coming back. It might be an extension.

P

Maybe that's what maybe he just saw that and you think okay, a message can be sent or expire or cause a specific error. But what if something else happens that people haven't thought about.

P

That must have been the logic.

N

But then gender- and I think it's hard to think about all future functionality that may be in the protocol right. Yes to cover that in the api, that's kind of not the procedure we have used, because we.

P

Started, I think corey is writing text, so we can look at his text data and go on discussing the next issue. Yeah.

N

You're writing text for the document or as a response.

J

N

J

Issue, I'm writing which text in the issue which says? Basically I don't like it, but it's nicer.

O

Yeah yeah, that's what I understand and we can look at that.

J

Yeah, you can add to it. If you don't like it or watch to get more clarification, because maybe the answer is we get. We understand what the issue was written already.

P

I think this is probably best solved for a new transport that needs this sort of api might further make the transport api more complex.

P

I can also answer saying this may fit in the future. The protocol yeah document.

J

Sure it would, I think it would yeah.

B

Okay, all right.

T

Okay, it's called versus framework. Is the scope used for message context ever anything but a framework? If not, could this section just save very much.

T

Message context.

P

J

Could somebody explain what the com? The question is.

P

The question is: if the scope is ever anything here just this parameter here, it's cool.

T

If you see it just maybe it should make it even bigger yeah.

P

T

Yeah, it goes is ever anything but a frame off, and if it isn't, then it should just be called.

K

P

Yeah is he still here.

J

No, but we can, we can still uh enforce that question onto tommy and see see whether he has any.

P

T

P

P

I can see you all multiple times.

S

We can assign it to tommy, so he notices it yep.

N

Sign it and put a comment on my kill that yeah yes, we think it is the same, but can you check.

P

We believe that it is the same. If so, it should probably be said explicitly.

J

Can you check.

N

Yeah, I think they would. He has wanted to use framer instead of scope right.

J

Yeah, that's one less thing to worry about. If it's true yeah.

P

Should probably indeed,.

S

Be named, framer be replaced yeah named.

P

Bravo tommy. Could you please check doing that.

S

T

P

Okay, be clearer with this reference.

P

Seven: five: five, six.

R

T

P

J

I will play my ask tommy card on this one. I have a comment and if tommy agrees with it I'll write text, can I go for that.

N

I think I think I think it's an example. I think you can write text and then ask tommy to review it.

J

Oh yeah, okay, all right, then.

P

J

We'll write text.

P

Additional properties, what is this anyway? The query connection properties will include different information.

P

This is when we query properties of a connection and then, when they establish, then we can learn about the path and use like, for instance, from the provisioning domain.

P

Aha yeah so I'll I'll sign it up wrong unless, okay, it is already assigned to you good next.

T

P

Okay, the nice of you to copy the text out here. Specific properties are defined in a transport and implementation specific way and must not apply across for follicles.

J

J

Okay, so so in the rfc series we generally use must not because it's um a requirement to interoperate this isn't such a thing. It's a beautiful part of our architecture. That really goes wrong when people don't do it, but it's not a must not for interoperability.

J

Should we write something like it should be recommended to not reuse or yeah, but we can prefix it by being really strong and say the taps architecture expects or um whatever it expects it to be used in this way enough applications yeah. We recommend that you don't do it, but we say beforehand that it's really important that you do.

J

Does that work.

N

J

So say this is important.

P

Because I have a feeling that you could write this safely and well, may I.

J

I will do it.

S

J

Sure you so much if we've all talked about it. That's fine.

P

Great. Thank you.

P

uh Improved text about local copies of objects, I think we're reaching the end of the underside things pretty much.

T

um There are several places where the documents has changed to the function. It's been called, do not have any connection existing as a as.

R

F

T

P

I think this is probably what we had in mind with I mean I remember that we thought about things that way quite often that this is like a local copy and it shouldn't be cross-influencing, and then that made us write some text.

J

This is, this is the deep copy.

P

Stuff isn't is.

J

Is this the thing we previously talked about deep copying when I needed to understand what that meant.

P

J

P

Deep copying yeah, I think I think so. I think that is how we were trying to avoid that problem by explaining the local copy and you have it and it's not gonna affect something.

P

It's gonna, be it's gonna, be that text, probably or maybe, but it says, multiple places.

P

Know you know what I'll just uh volunteer to go through the text and look for these things and see if I can say it clearer and then I'll make a pr and people can look at it and say no, no, that's not it or whatever.

P

Is that a good idea.

S

That is a good idea. Yeah.

P

Yeah make a start at least.

T

uh Okay, then,.

A

Can we maybe do a quick scheduling discussion now that we never.

U

B

G

Oh no, wait! No, no we're good! Actually, I think we're good on assignments. Right, wait like this event,.

J

Yeah, can we do 879 and just remind me that we're all happy with what 879 was.

J

In other words, we're not going to do that. We're going to loot carefully language is that correct.

P

T

P

I haven't really read it.

G

940, I think.

P

This one's actually fixed issue.

N

But this was some conclusion that came from last meeting right. That was then not agreed with. I don't know, I don't know the discussion from last time. Was there.

G

There were smaller fix it. There was.

C

G

System requires belonging this id the api, the separate rfc.

G

I think that we should we should roll this over and talk with talk with robert about it. If we can get him to the next interim.

J

Right I mean so so my suggestion is that we intend, on the current basis, to do the things we talked about earlier in this meeting for this one yeah. So we try and improve it in a way that we think it's no longer necessary to re-partition backwards sessions.

G

So do you want to keep this bug open to track the fact that we're gonna do that or do you want to say? We have agreed in principle that we're going to do that which we need to do anyway and close the bog as.

N

I think we can keep it open, maybe because maybe we need to make a few more passes over the whole document so now yeah to to see that we. So I think yours did a lot of good updates corey.

N

But when I was looking at the pr, I also saw some parts that still maybe were not so clear, but I was not reading the whole document at that point in time. So maybe more of us can take passes over the whole document and see if we can further improve.

J

On this okay, so my what I intend to do is put a comment just at the end now saying we discussed at this meeting and we intend to work on improving the document. Clarity see if this can address.

J

Does that seem fair.

P

To me, anyway, all right.

A

All right yeah- that was my understanding from earlier, and that is why I put in the notes that we are moving text around only if it significantly improves readability.

A

I guess, as opposed to moving a lot of text to interface.

N

Right got it okay, before some suggestion to not have normative language in language in arch right and to change the status and things, but this we are not aiming for. As I understand it now, corrector.

N

Who are you asking adam the rest of the people that were there last time and that are there now.

A

Oh yeah, last time we were basically saying: oh yeah, let's go with robot suggestion and make architecture informational to just be like a very short high level document for people trying to figure out if tabs is interesting to them and then have all the sort of meat of the specification in api, and that would imply moving normative language to api.

A

But I'm not sure if we're doing that anymore. I have not that's really never.

F

Been a major change.

A

Yeah, that's what we were discussing last time. We seemed in agreement, but then tommy tommy pushed back and I'm not sure if we're still.

F

Yeah, I have to agree with tommy that, like talking, we did discuss the normative language thing and we did decide that we have some high level architectural principles where it's worth using using normative language and that should stay in the architecture document.

N

Yeah to me that also sounds reasonable, but but I was also not there last time, so I don't know the discussion.

F

I wasn't also not there last time.

D

And I don't remember: okay.

A

We agree with robert's assessment that the architecture document by itself is sort of not very complete and people have to read api anyway. So why not put it all in api.

F

A

If we can make architecture more self-contained.

F

Yeah this is this is a good thought, because I thought architecture is quite self-contained, so we should figure out what's missing in architecture rather than moving some of the normative stuff from the architecture into the implementation.

F

J

That's why I wanted to talk about it, and I think I can write something now and we will stick with reviewing and improving architecture to make it. Okay.

N

And I think it's a good good idea to see if we can have a discussion with with robert for the review. Also.

C

Robert did.

N

O

U

Was going to try.

C

To attend today, so I'm not sure what happened but.

F

I think there are multiple ways to split um architecture and interface, but one important point is really that both of the documents have an understanding value.

E

F

Own by being in our own document, so.

P

Okay, reese wanted to get to scheduling.

A

If we have sort of an idea where we stand on 879 now, which I think we do.

F

A

Yeah, okay, cool yeah, I mean, I wasn't sure if scheduling would take 30 seconds or seven or the rest of our time, which is seven minutes. So I guess I wanted to do it now um so default.

A

Our default uh meeting date for next interim would be november 17, which is the third wednesday of the month, but which is also the week after ietf, and I'm not sure if that is sort of too soon and people might have other calls, and so what we could do is, I guess, maybe not the week of thanksgiving in the us, but uh we could just do december 15, which is the third wednesday in december, or we could move that one a little bit up to sort of have it earlier in december.

C

A

That's one approach.

C

The other approach I was thinking is that we keep the meeting on the on the 17th and if it turns out that there's a I star conflict which will hit a bunch of our active members, um ib or isg, then we would just cancel.

C

I guess it depends on how people feel about keeping.

C

I mean I'd like to see keep the momentum going, that I think that, if we, if we uh lose two months between meetings, that we end up having conversations like some of the ones we had today, where no one can remember exactly um where we left the discussion.

C

F

I just think that we should like add some extra time here, because nobody will be working on taps during the itf week. So that's the week. You can already add on top right.

C

I see some agitating for december 8th uh um so why.

B

C

Me just confirm.

C

I didn't hear that martin. What was that? Why not december.

I

G

I have a conflict on december 1st.

F

We have an ib workshop in that week, but that doesn't mean you cannot have me.

B

All right uh is it the ideal.

I

Workshop later in the day, I I would assume.

F

So it's anyway kind of a hackathon-like thing during the week. So, okay.

C

And the naive workshop would take uh myriad tommy and colin, so I think that's a that's not.

F

Necessarily good michael and I could probably also step out so this is not a big concern.

I

Well, I mean well, we can do the eighth, I mean I you know I it's not a big one week is not a big deal, but just in the principle of trying to move it up, not have such a big gap, as people asked.

A

And then we will have a gap between december and january anyways. I guess.

C

I mean in is there anybody who knows they have a conflict on the 15th just to see whether that is that even really an option.

J

Yeah, I'm not available on the 15th of december.

C

Okay, and is anybody blocked on the 8th.

N

Yeah, I have a problem on day.

C

I hate scheduling.

F

We should, I think we should also check with tommy, because we signed some issues, though.

B

A

Seems like more people have a conflict.

C

With uh with the 15th than the eighth, definitely the the 29th is out. um Sorry, uh I'm confusing my months here. uh The 24th is out because of the u.s holidays,.

C

And I think that the argument that no work was going to get done during itf is a is compelling. So I think our options are december, 8th or december 15th.

C

uh Well, it seemed like that hit a bunch of I mean I. I thought that you were the only iab member who would not be conflict who had the option of missing the workshop right now.

F

So I'm not sure if tommy's participating colin is participating for sure, but he might also be able to step up because we have the second hand back thing. So I'm not sure how that will be organized.

K

Yeah I mean I I could probably I prefer not to have it on the same day, but I could probably step outside.

G

Thing exactly from five to seven on the first, that is not not avoidable. Okay, uh but I mean you can also go without me right so.

C

Seems to me that of the three dates that the eighth is the least conflicted. But we should check with tommy.

C

Does anybody sort of disagree with that characterization that um and whoever can't.

G

Make the meeting that we choose can give all of their issues to the people who can't make the meeting.

C

I thought it was the other way around.

G

No, no and then they get all of the issues that come up then.

C

They all get back uh all right. um Okay,.

A

I can check with tommy.

C

Tommy awesome. Thank you thanks everybody, uh a lot of progress today. It keeps getting better we're sort of in zeno's uh draft editing, but uh you know I know that series converges all right have a great day.

F

Thanks to sean.

G

F

Up, thank you meet echo for this purpose.

G

F

Think the limited video is a real.

G

Problem, the nine people in the video is a problem.

A

G

I will point out.

A

That in webex we had four people and then we had to switch screens back and forth.

G

Right, um if we're trying other things out like I can set up a google meet thing yeah exactly my.

F

Computer reads:.

G

You know you know like so like the cool thing is, if you actually, if you tilt your computer, the right way and you put like eggs underneath it, you get like air, fried.

D

It keeps your coffee warm as well.

G

It does keep your coffee warm.

D

A

Does anybody know how to open it use me anybody know how to open a ticket with me deco.

F

Yes, just send something at support, itf.org or there's also meat. Echo email address that you can use directly media echo whatever I don't know.

A

Okay I'll see, if I can find it all right.

K

Or invite the filthy people to the next meeting.

G

So you're all around ietf in a couple of weeks, yeah.

G

G