Internet Engineering Task Force Interim Meetings, 1 Jun 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF-CBOR-20220601-1400

Description

CBOR meeting session at IETF
2022/06/01 1400

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

A

A

So francesca, did you hear us.

B

A

Wow, something is using a lot of cpu right now.

A

That's not normal.

A

Yeah, that seems to be the medical tab.

B

C

Hi baby hi, everyone.

B

I have apparently still been unsuccessful at convincing the medico people to sort all the jabber users at the end.

A

Do you have a message that I can plus one.

B

uh Let me uh so: where do we? Where do we post suggestions for meat echo? What is the discus we'll discuss? Okay, I will, I will post a message there and you can plus one.

B

Now I think I had suggested it on the working group chairs list ages ago. As the first time I suggested it.

B

B

I don't know if we're gonna, I know ian is not coming and christian is not coming. I have not heard anything from michael either way.

B

A

One two: three testing.

A

I heard you yeah but did transition francesca. Yes,.

A

And she's going offline again.

B

So unfortunately, it's just you me and marco and francesca if she can get her audio working.

B

Let's wait a few more minutes and see if we get anybody else.

B

I always hate to disrespect the time of the people who are prompt, but.

A

What is that interesting, sound in the background.

B

uh They are today is lawn mowing day in my neighborhood.

A

A

I had a guy with a chainsaw standing on my balcony yesterday, so I think that's an improvement is that monthly, the chainsaw.

C

Yeah not mowing.

B

Sorry is it what marco.

C

Is that a monthly event in your neighborhood or yearly well.

B

No, it's weekly. um Now it's uh it's supposed to be every tuesday, but I guess because monday was a holiday in the u.s. Everything got pushed out a day.

C

B

Yeah they um they mow every week, starting sometime in may and until um sometime in october,.

A

Okay, it seems I'm not successful.

A

Summoning other people, okay, everybody seems to be somewhere, but not online.

B

Well, I mean, I guess that we can, just if you and marco have things to discuss um in this, we can start it.

A

um Like if you care about zebra, picked.

C

I care I, I don't have any particular point to discuss other than what I raised in the review right.

A

Yeah, so maybe I should actually write up that design. That is on the slides. I have a couple of administrative points near the end, so maybe we can could at least go through those quickly sure.

A

A

Everybody's favorite subject, um so we have one errata that was reported a year ago on 8610 and we had a little bit of discussion on the mailing list. But we didn't take that anywhere. So who actually has the the ball in the court right now.

B

Yeah, the um the ad is the one who does the uh handling of the errata. They look for advice from the working group. uh We often for trivial things. We just do it, but um this one is not trivial, so um you posted yeah a year ago to the mailing list. I see that post.

B

I mean, I guess, uh if nobody commented on that, your suggestion of hold for document update is, can be considered. The working group's advice to the ad.

A

Yeah, I think, the the the text, the notes for the hold for document update, would uh mention that there is a missing pointer from 223 to 36, where that that additional effect that is actually supplied so that that solves lawrence's uh problem right here. All the information is there it's just not properly internally cross-linked.

B

B

B

Right, I guess the thing to do with if francesca can't hear us, uh the thing to do is just send email to uh seabor ads, um asking that they close this um withhold for document update. So I guess I can do that. Thank you.

A

So we now have another participant right.

B

So francesco says she can hear now but missed the beginning of the conversation francesca. The conversation is about uh errata report um 6575, which is on rfc 8610, and this the working group suggestion is to close that hold for document update so I'll. Send you that by email, okay, go on carsten.

A

Yeah, so just to to remind people what what the the underlying problem uh here is. uh The we have this this weird syntax we use in cgdl for uh basing for essentially grounding the the semantical uh definitions uh in the syntax of sibor.

A

So hash 7.25 is, is a major type 7 with an additional information of 25, and there is one weirdness in in that we, when we actually add parenthesis to that, we we are not using the additional information but the whole argument.

A

So the additional information just tells you how many bytes there are in the argument and then the the argument supplies uh the actual information so that that's um yeah, um maybe not that greater design, and um this actually came up repeatedly um in in various drafts, where we tried to use some some generics and and couldn't make this work. So you, you cannot say dash six dot x uh somewhere in a generic or you.

A

You cannot compute tag numbers and if you could, the syntax would be uh really bad, um probably not not possible uh the the way it would look like.

A

So it seems we want to do something about that in the next version of cdl, because it has come up more than once in in rfc under under discussion. I received to be completed, and maybe it's.

A

This is it's worth discussing this together with other formats, so, for instance, in the http with working group, uh they found out that they don't have a way to talk about structured field, values and cdl is approximately 90 percent uh useful, for that it just doesn't have an obvious way how to base it on the very different generic data model of structured field values. So this this is almost but but not entirely, unlike uh jason, so we would have to.

A

We cannot just rely on the analogy to to jason um here so that that's something that uh we should look at if we think it's a good thing that other um documents can use cdf for defining their data models.

A

Of course you also could use yang, but then you need an even heavier adaptation mechanism to to transfer the yang data model to the data model that that is actually uh being used here.

A

So that's one point: I wanted uh people to start thinking about. uh We probably won't have the the solution in in this meeting, but it's definitely on the agenda for for an update to cdl.

A

Okay, I'm wondering if anybody has anything to add them.

A

Okay, I make this point to to other people again and again and and see whether there is a point in trying to make this uh happen or whether we should uh frugally limit ourselves to sibo and json applications of cdl.

A

This is a bit difficult to narrate.

A

Yeah so the the other discussion, the the sibo pack discussion.

A

um We had a good discussion two weeks ago and I tried to uh flesh this out and uh I I now have a design that that I think can be made to work. I haven't completely implemented it yet, but what we took away from the previous meeting was that we won't have separate prefix and suffix tables and that we probably want to make the merge table available to additional processing that we hadn't um defined uh two weeks ago.

A

So we we already have the the processing that is making use of the information for prefixes and suffixes. But we haven't looked at other ways of using this and I have a little bit of slide. We are doing that and let me just show how it looked like uh two weeks um ago: that's the the one with the tank 51, and uh so what the the change would mean that we wouldn't have separate, prefix and suffix table. They would all be in one table and that just means we have to use different numbers here.

A

So we cannot start at zero again for the suffixes, but then, on the other hand, we can use the same string for suffixes and prefixes, so this is probably the the right thing uh to do.

A

um So. What what we now have is we, the package consists of the table setup tag, uh the um sharing reference, so a reference that just points into the sharing uh table and extracts the the item there unchanged- and we have this argument, reference thing where we have a tag that has both their their own data, the argument and the reference into the the uh merged table and a bit that says whether this is a prefix or a suffix reference.

A

So that would continue to work on on pairs of strings, arrays and maps as before.

A

So within those domains, string area and map, it would just work as prefix and suffix, but it also could work with tags and by consuming such a tag, you would execute a function, for instance like this midfix thing, so the processing model.

A

Is not really defined for packing. That is not a change and for unpacking.

A

We would have the two processing models: the accessor in place model, which we have also called pointer, tracing and the the unpacker model, but the unpacker model, where you unpack the whole packed silver data item that that can be described in terms of the accessor in place model, so would just use the accessor in place model for defining the the meaning and um the the meaning is essentially and when you have an when you do an accessor in place uh unpacking uh if you are looking at a reference tag, so one of the tags that we define in this document.

A

If it's a sharing reference, then it has an implied number n somewhere in its in the tag which is 0 to 15. For for a simple data item and 16 or more for take six, then you index the shared table, uh apply the accessor in place, uh processing to that and and return the whole thing.

A

And if it's an argument, reference where which has an additional parameter, the content of the tag that is doing the referencing, then again, you you do a lookup in the argument table and you run the accessor in place on both the argument and the tag content.

A

And then you run the argument processing which is dependent on the the flag. This a prefix or suffix uh on the reference that that was implied by the tag number and on the uh unpacked uh content, which is referred to as the rump here.

A

So this this argument process uh would look at both the reference and the rump, and if both are strings, it would do a string card and if both are arrays, it will do an array cad and if both are maps, then it would do a map merge and otherwise it would do a function process. And that's the new uh thing that that's coming in here and the the function process uh works by using a tag on one of the two uh parameters here on the reference value or the ramp value.

A

So there must be a tag there. Otherwise, it's not uh not a valid uh accessor in place reference, and we can use this v flag, the prefix versus suffix flag. Otherwise, we can use that flag uh to decide whether we are looking for the tag in the reference or we are looking for the tag in the ramp.

A

So it is completely determined by by the referencing mechanism and not by what you find in the referenced uh data which of the two is is true, so essentially you find a tag there and then it's the job of the tag to tell you what to do. So. That's that's, essentially the extension uh model here and let's uh look at that in an example.

A

Let's imagine we define a midfix tag and uh we could uh put this midfix tag into an argument, reference tag uh with the the index zero. So we would extract the pack.example and use it with the mid fix semantics. That, of course, have to be redefined somewhere uh like for every tag and um then um the the function processing would find the the midfix tag because v equals one. It's it's the one in the argument and not in the reference and uh the matrix tag says.

A

uh Take my take the array inside me and put the the first element. First then, the referenced data and then the last element.

A

um So this can be done the other way around. um So if we know we will be using pec.example in the matrix situation, most of the time we can go to the v equals 0 reference, so the the prefix reference and uh leave out the matrix there, but put the matrix in in the argument table. So this is essentially the the same thing, but uh the diff differs in whether the referencing side says uh that we want to use the matrix processing or uh the argument table says that so this all works.

A

Even if it's a little bit difficult to to express in one slide, we have a couple of new terms: the argument table, which is the the merged, prefix and suffix table, and the argument reference, which was one of the ethics tags, and we have function tags that define some processing that goes beyond what is defined in in this document.

A

So the function tags obviously are the extension point here.

A

The a generic unpicker would not necessarily know about all application specific function tags the generic unpicker would be able to find out that it just hit on an unknown function, tag and then function tag unknown to it, because the argument reference semantics are telling it using the reflect where there needs to be a function tag um if we are using a tag at all.

A

Otherwise we have the three string array map uh mechanisms that that are implicit functions, but if it's none of these three, because we have a tag, then that must be a function tag and the generic unpicker can decide whether it can. It knows the tag or has to leave this to the application, and the only thing I couldn't really figure out is how would the generic unpacker present this to the application? So clearly, this is outside what you can express by head, just handing a zero structure uh to uh the application.

A

So we probably need to say something about that, so this relies on on only ever adding function tags. If we actually add reference tags, then it's much harder to do forward compatibility.

A

So if we ever find out that the set of tags we have defined here is not enough, that will be much harder to add to existing implementations.

A

Okay, so we said last time that we want to do this. This whole sequence thing and the unwrap splicing proposal outside of this document- and we probably don't want to do a cushion sync set of function tags in this document, but maybe we do it should do at least one. So we have a. We have something that people who are trying to implement sibo packed uh try to implement, so they they have an incentive to actually get function. Tags uh going so maybe midfix is a good candidate for that, but we can find that out.

A

um Yeah then, the the error handling the error handling of course provides additional opportunity for extension points.

A

I didn't find any particular reason to put extension points here, but that's maybe one of the things that one should think about so right now you can do string plus string array, plus array map, plus map or tag plus any or any plus tag, depending on on the v flag.

A

um Yeah there's uh one uh ugly thing uh remaining tag. Six is is uh uh expensive as edge. It's it's an efficient way, uh tagging things. uh So we are uh splitting tactics between shared references and argument, references and with the fact that we now allow any um in the um set of parameters to a function. Tag we no longer can rely on the fact that there will be no integer there, so this becomes ambiguous, or at least if we define it the way it's defined right now.

A

It means you cannot use tag 6 for supplying an integer argument to a function, tag and yeah. If and the function tag wants that. So maybe we should just bite the bullet and reserve another tag which would be 224, naturally um as a reference tag.

A

That is also using the first item in the argument table, but doesn't have this this shared duty between shared references and argument references, so this would be spending one more uh tag, but it would make uh would avoid having this weird watch where you suddenly cannot use the the argument.

A

The first entry in the argument table for for something that all the other argument table entries can do yeah so extension points. We can define new table setup tags, we can define new function tags, but we probably don't want to define new reference text. That's that's my proposal to handle that.

A

And then we probably need to think about complexity. Do we want to do some profiling uh here? I'm pretty sure that we will want to have a sharing only uh profile for for people who uh only want to do point out, chasing and no function tag processing and then the the number two is the one with function tag processing and then maybe maybe there are subsets there uh as well, and one question would be whether we we add additional tags that explicitly identify um this using a specific profile.

A

So that's the the questions to the working group I had prepared. I hope that some people will look at this in the uh recording that will be up soon and try answering this this off list. But if you have a reaction right now, I would be glad to hear about that.

B

I hear a roar of response from two people who are on.

B

B

So I guess we are done with the agenda.

B

um I'll note. uh You've probably seen the message to the list that the chairs requested a session in philadelphia. So we do plan to meet for an hour in philadelphia.

B

And we will work on an agenda for that over the next month.

A

Yeah, so the the assumption by assumption would be that we will have a completed uh reaction to the regular blast goal comments for the pac document before the internet draft deadline for that meeting, so we will either have a second working plus call or however, we we handle this um and discuss it in in philadelphia.

A

um I'm I'm not sure we have any big other agenda items, but we will find out.

B

Yep um yeah my inclination right now, depending on how extensive things need to be. Is um that we'll just let any other working group comments on pact come during regular last call? So probably uh when you post this revision, we'll just uh send it up to francesca at that point,.

A

B

Okay, then I guess we are done for today we will meet again in two weeks agenda given, and um I will not be here- christian will run it.

B

So um christian will see you guys in two weeks, and I will see you after that anything else from anybody before we close it.

B

Okay, it is a meeting enjoy your day.

A

Thank you, bye.

C