Internet Engineering Task Force Interim Meetings, 26 Feb 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: CBOR WG Interim Meeting, 2020-02-26

Description

CBOR WG Interim Meeting, 2020-02-26

A

Darling: okay: today's agenda see boobies again status, update then progress.

A

Both the etherpad.

A

And also opened.

A

And bit slow today, oh.

B

That's very slow.

A

A

Testing my patience.

A

Try to yeah so.

C

Hard about either bath.

A

Why is it slow, though,.

A

Does he work for you no.

C

A

Okay, let's, let's use this one may be done to.

A

B

A

B

That work I.

A

Hope so does it not.

B

Doesn't seem to behave more, thank you for that, and.

A

I see the gym is hot, so.

D

I'm not giving any content you're.

A

A

Well, I guess: I get.

B

Loaded, tell me about this I.

D

Have kind it now.

A

Yeah either bad this really. So let's try to see.

A

How do you do design element.

C

I'm getting packet loss on pings to that address.

A

A

B

That was, that was really bad.

A

A

Copy-Paste the agenda.

A

So February 13 I know it's been a.

A

Copy this the whole thing.

A

Okay, captain you can tell that that's the status.

B

Yes, that I think I broke.

B

Number 63, so I proposed some text and the text essentially states.

B

It's not really making any recommendation. It just talks about what's true and we probably need to discuss whether that's enough or whether we actually need to be more prescriptive. Yeah mm-hmm, so don't already sent a video I'm, not entirely sure that I can take a direction from this review, because it has two things that that don't really agree with each other.

B

So you seem to say that all decoders should always.

B

Detect and reject to wrinkles, and then you say: do we want to really exclude the chorus that don't do that.

C

So I was what I was thinking is what I understood from that text is: there's a difference between.

C

And one behavior is: is that as soon as you detect a duplicate you, you error, you report the fact that you detected a duplicate and.

C

Refused to refuse to go on yes,.

B

C

That's that is a we all agree. That's good reasonable behavior and some generic decoders will do this and should do this and protocols shouldn't rely on this and that that's in some ways it's an ideal case, but another behavior is when you detect a duplicate. You don't error, you just take first take or take last or or take random or some other. You know rule so, there's no error, you're still doing duplicate detection, but you say you're, just passing from the mound implicates yeah.

B

That's the interesting not so interesting case, because if you are actually detecting a duplicate, you might as well act on that I think you are now talking about the fermentations that do not detect duplicates but just handle them and then by losing the other entries with the same key yeah.

C

B

There's no duplicate detection going on. All that happens is those other entries are lost. I.

C

Would call that I am calling that duplicate detection, because you do have to detect that it's a duplicate.

B

An implementation that that well does search from the beginning and returns the first match. This is inefficient, but it might be appropriate for constraint device that would process of map with a duplicate entry by returning the first entry, but would never detect that there are ones behind it.

B

Similarly, implementation section.

B

B

Entries into a map, and since the map doesn't tell me where the previously was something in their entry, it just gets us there's no detection going on.

C

C

Yeah I understand what you're saying. um Let me, let me say it this way: um the problem with any kind of things, any kind of implementation where you lose or don't don't pass on. You know you try to implement, take first or take last. Is that I don't think we can specify.

C

It should always be take first or always take last. We can't specify that so it's some implementations are going to do only and others are going to do it another, and if we, if we say that you know one of these kinds of behaviors is okay, then um how do protocol designers know what to rely on I mean? Will they rely on take first or take last?

C

That seems a really bad situation to put protocol.

A

Designers in on.

C

And implementers in right.

B

That's why there's takes the terrifies that you cannot do that.

C

The diplom okay.

D

B

Tells you that generic encoders that lose stuff are not applicable in all situations.

B

So whether you consider that okay or not okay, depends on drawing that situation or not yeah.

C

B

So all this property can, we send clearer, can be said clearer than the text areas, but I think that the basic summary of the discussion is.

B

Pretty much what we discussed on the mailing list, yeah.

C

Yeah no I see what I see I understand the intent of your text. Now that that third option is the one where.

E

You don't know.

C

What the encoder is going to or what the decoder is going to return? Sometimes it might return, some decoders might do first. Some might do last. Some might do random. Some might change. So you you can't in that third option. You have no I mean I I, suppose, even in that third option you have you have you have no way of relying on what the knowing what the decoder is going to do.

C

I mean there is a class of behavior of a decoder that is not covered by those three options that it's very unlikely to occur to occur but might occur, which is that sometimes it passes duplicates on and sometimes it eliminates duplicates.

C

B

Be covered by the third option.

C

Okay, it wouldn't it would be I.

B

Think so, I'm not sure I phrased this correctly, but that was the idea. Okay,.

D

B

Check the phrasing.

B

Yeah, the third option said: lose some entries with duplicate keys. It doesn't say that you have to lose everything to be in that third option.

C

B

B

That's an example: yeah, okay,.

B

But we have to make sure this is clear, so you have any suggestion now to make this less easy to misunderstand. That would be useful.

B

C

So I was thinking. Those decoders can only be used in situations where protocol is not sensitive to duplicates.

C

That would be the condition not the reliance on the encoder being correct or the encoded see where I'm being not tampered with.

B

Well, yeah I'm, not so sure that that's actually true, because there are many applications for force Ebola that happened with India security domain and there you actually.

B

Usually, or often are in a position where you can rely on that and I wouldn't want to exclude those implementations just because they cannot be used on the open, Internet.

B

So forth, humans actually plant or put in non-strict flag, for which I need to find any that says the existing behaviors okay. And if you don't set that, then you for checking and a performance sitter.

C

B

So I think the last sentence of the bullet list is quite explicit in pointing out that these that the decoders that don't check have a limited area of application and I think that's pretty much. What what we took out of the Singapore meeting.

C

Well, there's there's protocols that don't have issue with duplicates then there's.

C

Environments or situations where.

C

You at least you think you can rely on on correct encoding, I, don't know I'm actually, so I think that the protocol part might be missing and.

C

C

Any text about situations where the encoder can be relied on should be I would put a lot of big warnings on that, because that's a classic area where people get you know implementers make assumptions that are not true. Like okay, it's behind the firewall. Okay, the firewall includes a company of a hundred thousand people yeah.

C

So therefore it's okay, yeah I, don't that doesn't seem, like you know, okay scenario to.

C

To trust the encoder, so I mean I, would I mean I, guess the words I would pick would be things like good practice.

C

You know what assumes input is always hostile, but some some scenarios in some scenarios. You might consider the input not to be hostile, but you should be really cautious about that.

D

Kirstin I pasted some text in the chat window. I.

D

Think we may want to have text along this lines, included.

B

That's absolutely something that that needs to be explained.

D

I think that's the only real problem I have with the text. Is it's not sufficient about telling you what what can go wrong.

B

Am I temp with this against these generic decoders can only be used, and so on was to be very specific on what what can be done and cannot be done.

B

B

Well, I, like your texture, so I'm pasting that in.

C

Yeah I think that that text is a good addition.

C

um What about text that.

C

Requests decoders to declare which of these three options that they have implemented. Yeah.

B

That was kind of implicit, but we probably need to make this explicit.

C

If you're going to keep the thing about scenarios, I don't know about situations where the encoder can be relied on upon. I think you should expand that to where the encoder end entire.

C

Transport infrastructure- or it's not just the encoder here, it's the the man in the middle. You know.

B

So I'll just replace about data source on transpose yeah.

C

And eg no men in the middle.

D

The fun case, of course, is when the man in the middle is yourself as you load it up. You write it back out and you change the order.

B

The others lot of problem of things invalid before they will be invented afterwards and if they were valid before they will be valid afterwards.

D

But if things were invalid before they may be valid afterwards, yes,.

A

My question is:.

A

This is a formulation problem we are having now, or is it even that, so are we disagreeing on the content? I'm not.

F

A

At this point, I thought we had some agreement at ATF 106.

B

And that was refined in the Middle East discussion we had afterwards I.

C

Can't remember exactly what was said in 90th 106, but I in general agree with the content here. Yes,.

D

C

The the some of the improvements we're discussing here with I think that.

A

Are mostly most.

C

Mostly aligned up here, okay.

D

I'm I'm basically happy with what is here.

A

C

I this lines up with the comments, many of the comments I've made on the list.

C

Yeah yeah the other another.

C

Maybe that might be helpful to add here is this is I, don't know you know whether there's a port where the protocols, a protocol design, something about protocol design, you design a protocol to tolerate.

C

Duplicates or not.

C

B

Is no way in people to really tolerate.

B

I'm not sure a hole that would work.

B

It just version with gyms to extend a few more changes from the discussion. A.

C

C

Every map entry is a addition or deduction to your bank. Account is but not tolerate, duplicates a protocol to report the current air temperature.

C

We would would tolerate duplicates.

D

Would give half the people the wrong answer.

D

We duplicates the tip in the temperature one yeah I mean if, if a pitcher had different values than half people get their own answer,.

C

Yeah, maybe, depending on what the point of the protocol was and how is how those duplicates you know, I'm getting insulted.

D

And what I can and how reliant I was on getting the right answer for the temperature I'm, sorry I said it again and how reliant I am I, giving the right aperture answer for the temperature if I was going to turn on the air-conditioning, your turn off the air-conditioning based on that we're all set now in a situation where you may not be taller.

D

If you have an application which is tolerant, you're gonna want to be in a second bucket, because you're gonna want to know all of them.

B

That's why I didn't even address that, because I'm not sure what kinds of tolerant we actually would have and I wouldn't want to design a protocol in in such a way that the elaborate mechanisms to cope with something that that is invalid in the first place.

C

So generic decoders that are not in class 1 or class 2 or though they're the first two bullet points, have really limited applicability. Yes,.

C

Yeah, no man in the middle.

B

And I didn't put in the shaman in the middle because I take her I. Think it's a good enough.

C

It has to say: I: I think it has to say, is not possible if the transport or the encoder is under control of the attacker. It's the transport that we really care about here, the most likely the encoder is going to be. Okay and the the you know, the the sender originator of the Seaboard is going to be. Okay, I think it's it's the it's! The man in the middle.

D

In my case is I'm more worried about the originator and the man in the middle, but to me the encoder and kind of implies. The man in the middle case is whoever did the last encoding.

B

Now maybe so, the the Federalist line already talks about details and transfer, and maybe the second to last line should just say data instead of encoder.

C

D

Encoding might also work.

B

B

B

Okay, so my plan would be to wait a little bit until we. We have more more thought a little bit more about this and and have reviews and then to ask Paul to merge this, and we also have another pull request.

B

165 that has been.

B

Around for a while and I think I managed to put in the issues from number 160 Eva from issue number 160 in there as well, using some of the text that Peter oxide had proposed. I had written.

B

So that's, maybe one one other thing that I would like people to look at then we could merge those two poor requests and essentially stuck with the clerical pull requests like checking my registrations and checking the list of changes and so on, and when that is done, we can put all the next version and to awaken bla scholar.

A

Sounds good yeah.

A

No, no, no answer from Jeffrey.

B

He had 45 minutes before this. Oh no.

A

No but I mean I thought about the.

A

Before the last changes you had already written so.

B

That the situation is that that those points were kind of peripheral to his points. So maybe he simply doesn't have that much of an opinion.

B

So I I, don't blame him. I.

A

Kind of remember you last interim, you said that you had reached out to him. That's why I was.

A

B

I hope you will respond to the ring Osgood yeah.

A

So yeah well inform the next update done. I think the discussion was useful for progressing the for improving the text. Yes,.

A

And if there is anything else, any other comments that people who are here here I think that should be solved before working group in school.

B

Is to raise an issue.

A

A

Anything else, let me just go through the points from last meeting.

A

Still waiting waiting on Jeffrey's approval and this done.

A

Floating point: really it did so that's good.

A

Request form a 3d tea, okay, Lawrence to check PR, 159 and notify personal mailing list was another 59 I.

C

Haven't done it yet, but I will keep promising. Okay.

A

If I keep it in the minute, so we remember oh.

B

Hi Michael I.

F

I've joined about three minutes ago, so I have no idea what you're doing I.

B

Think we are ripping up, I know.

F

Figured that out so.

A

B

Main discussion point was about the this hot issue of met uniqueness.

A

The link to the minutes working after next mission, I, am.

F

Dean, are we using hack MD instead of etherpad wonderful, well,.

B

Having problems, okay,.

F

Anyway, hack MD school.

B

Have essentially pushed me into a can using heckum.

B

A

And we talk to implementers geeky, okay, that was a new issue, a hundred sixty seven.

A

System was closed.

B

A

The first encoding tagger runtime I.

B

Think that the in 168, let's tell us to close that.

F

I read the text in the pull request, I think and I. Think I'm. Ok with it.

A

F

I said it the last IETF that I thought should go on in a specific way, but I was in the rough. So that's ok.

A

And we got new new ish 170.

A

Let's take a look at hundred sixty seven hundred seventy.

F

These consider allowing Sieber null as a valid contact tag, content for tag number zero and one with our tag. Number zero and one.

D

The different ways of specifying dates times, I, don't know why they just don't want to put in at all it's just the type value.

D

Because this would basically say you want to do it, you always want to have Noah as an option on the tag. What happens if I don't have that mime value there well I have the tag there and then no write.

F

Database is like it when they can specify no, instead of a other value.

F

What does it mean, however, it still fails to satisfy the protocol requirements for always encoding tag, number zero and one which can lead to compatibility problems.

F

What does that mean.

B

Well, it's you can create problems and by making requirements in your protocol, then turn out to be bad and one such requirement may be to always have a tag number one somewhere and then, if it turns out, you have to be able to encode any find reuse. Then you cannot fulfill this protocol requirement to always have a tag number one. Unless we change take number one and I, don't think you want to do this at this point.

B

So what my solution would fix the protocol, the.

F

Protocol, meaning the user of C, bore the.

B

The text that says there must be a tag number one. There.

D

Which is the easier C board.

B

Which is one of the users, but some.

F

User says there must be a time in 0 or 1 in in my thing somewhere and- and they can't accept the that that if there is no time that there should be no tag mmm and therefore they want to put a none initial in a null. Instead.

B

This is very known from certificates which have exactly the same problem in certificates. You must have expiry times. The convention is to put in an expiry time in 9999 a year.

F

Or not specified right.

C

The problem in certificates- you know where the field is, you don't rely on the data type to know which data item is which certificates just don't have to don't have the ability to put null in. But this this problem is different. This is where you're, relying on the type of the field to figure out which field is which it's some sort of any implicit.

C

Thank you like I, think so so, and I I think there is definitely something that you can't do in see more here, because because null and undefined are excluded from tags, 0, &, 1, I, don't know.

D

C

This is a critical thing to be able to do in C, where you might be able to work around it. Maybe your protocol is a little bit bigger because of that it doesn't like.

B

C

Something you can't do because of this and I.

D

Mean CR else, the expiry time is an optional field. You don't it's I, don't see any problems with making it an optional. You know that the tag.

F

D

Was supposed to be so.

F

Yeah, so in in the certificate issue, the issue is that the expires on was a mandatory field, and so we had to put a stupid value and if it had been an optional field, then that could be okay. So let's say someone had a map and they had expires on. Could they know what, instead of having a zero or one with a null value? Couldn't they put a null just the Seaboard null in? So it's not tagged, zero or one? It's just a null I think.

E

That'll work for for lots of protocols, but I think there are some protocols where it might not work. I.

F

Agree, there's some particles that might not work. Do we have exemplars right now? In other words, are there people we can't fix, except by doing this, that not.

C

This probably affects other than this I have to sit and come up with the examples, but I.

C

Mean this would be an incompatible change to zero and one wouldn't.

F

It so who is who's the requester, it's Montgomery Edwards, yes,.

B

We're set for four.

F

Eight logging is X 4, 4 8, because it's it's definitely a pseudonym and one wonders if he's dealing with a certificate issue itself.

B

So I think I will I will write a response who this the thing we need to decide is whether we we add some explanatory text to the definitions of tech, 0 and 1, but I, don't think that we actually want to change the definition of technical 0. Well, though, we have other things like 1001 that what, with all that I mean this would be true for like every tag right.

B

A

Ok, Mike just commented: yes, I'd send.

D

B

F

Was the other issue that you wanted us to look at.

A

The one that we had last meeting was.

F

A

Which Montgomery Edwards closed I? Guess? Because.

D

A

So it just opened a new.

A

A

And altered one.

A

Okay, yeah there's 68 with us close because under 67 or.

A

Basically also related.

A

Now what Thurston was asking people to look at was the PR 165 right. Carson, yes, which is the older open request.

B

And it's pretty simple, so I hope that that we can reach agreement on that quickly. I.

F

Use the C compiler in the 1980s that didn't have a floating-point and after all those years, I find I almost never almost never use floating-point. It's like I'm a can't like wow.

D

We have floating-point.

F

F

Have to remind myself that they're not new.

B

Okay, the decision between using of cortex and 3-yard cortex m4.

C

That's that's the hardware, but the compiler can add lettings point support in or not I know what one of the issues that I mean when I was at Qualcomm, one of the in the TE. There was no floating point just because no one had got around to looking up the compiler and the libraries to support it. It just wasn't a priority until things like fingerprints came along.

B

Yeah I think that's one of the reasons why people this is a little bit more careful about throwing all kinds of numbers into the same bucket and maintaining a clear distinction between integers and floats.

E

Yeah, this seems like a good idea.

B

A

There is nothing else, I think we're done.

B

Yeah we could mention for those who have noticed that we have a new RFC. Yes,.

A

B

B

A

So 8742 see both sequences, yes, the new one, so good job very quickly and smoothly done and now we're waiting for seaboard array. Yes,.

B

Okay, that is not yet recorded in the earth 48 page, but probably just because pyramids.

A

Stopped recording now.