XSF Discussion - 2020-09-27

MattJ, are the new MAM filter fields 'before-id' and 'after-id' exactly equivalent to RSM's before/after except that they're allowed to be combined?

MattJ, if so, couldn't the "server MUST return an item-not-found" thing be limited to those new fields, as clients who rely on that behavior could just use the new fields?

Holger, so allow non-existent ids in RSM to proceed without an error?

That would be a breaking change, no?

and what would the server return Holger ? a complete=true?

> so allow non-existent ids in RSM to proceed without an error? Yes, as permitted by XEP-0059. It would be nice if generic XEP-0059 code could avoid special-casing MAM. > That would be a breaking change, no? I guess. Unless that MUST was added after the :2 bump?

(I think you didn't have it in earlier 0313 revisions, but I'm not sure at what point it was added and can't find it in the revision history.)

Holger, but the item-not-found is not a special case for mam

https://xmpp.org/extensions/xep-0059.html#notfound

Anyway if it was added before the :2 bump I see how it would be a breaking change, so it might just be an item for the next bump.

If you'd be fine with that change in general.

lovetox, that 0059 section allows servers to return an actual result if one of "the following circumstances" doesn't apply.

sounds weird but yes it says "all circumstances must apply"

hell knows why

but what would you return?

say you delete the database, i request the last mam-id which is invalid

how would i get my latest messages?

if you dont return an error

> The UID itself cannot be used to derive directly the next item within the set (e.g. the alphabetical or numerical order of the UIDs do not specify the order of the items). ejabberd has numerically ordered UIDs. If you request <after>42</after> it'll return the messages with UIDs >42.

yeah you have, but thats just your implementation and ordering is not required

Sure.

still if you reset your database

you start with 1

i request 1000

what do you return?

nothing? then i just lost messages

You ask me about my implementation and then you complain that I'm just talking my implementation 🙂

ejabberd's UIDs are timestamps. So nothing bad happens in your "reset DB" example.

But I don't think this is relevant.

so what happens if i request mam-id=asd

or are you saying you dont want to send an error in the specific case that you can determine a correct result?

XEP-0059 says "servers can do $foo if the conditions $bar apply". XEP-0313 says "servers must do XEP-0059 except that they can't do $foo even if the conditions $bar apply". That's the part I don't like.

but in others where you not, you still would send an error

> or are you saying you dont want to send an error in the specific case that you can determine a correct result? Right. XEP-0059 says that.

> but in others where you not, you still would send an error Right.

lovetox, actually I seem to remember you arguing the other way round and liking my behavior because it avoids another MAM request 😉

im not arguing for anything currently, i just wanted to understand your proposal

at first i thought it was, send never an error

Holger, can you take it to the list? Partly because I don't have the bandwidth to think about it today, partly because I don't want to forget you asked, and partly because I think other peoples' input would be good

now i understand it, sounds good to me

MattJ, ok!

I think if it weren't for backwards compatibility, this restriction would indeed not need to apply to RSM

so if we can find some loophole or way to remove it, we probably can

MattJ, apart from this question; my understanding that `before-id` and `after-id` are equivalent to the RSM elements except for being combinable is correct?

I.e. I guess being able to combine them was the motivation to add them to 0313?

i think equivalent is the wrong word

before and after set in a rsm combined makes no sense, so no its not equivalent

yeah if you think RSM is a filter and not a paging mechanism then it would be equivalent

but its not

Right.

Yes yes I agree 🙂

:D

Holger: yes, being able to combine them was the reason

Ok, thanks!

Since many people requested that and abusing RSM even further seemed like an undesirable path

Yup, makes sense.

puh implementing vcard4 spec is .. challenging

i would say its so complex that probably no one right now implements it in full

and we are probably in a situation where one impl overwrites data that it does not understand

lovetox: what problems do you face?

i don’t yet, i just think its very much work to implement vcard4

i dont think you can support a subset

you have to implement the whole spec

otherwise when you receive a vcard with properties and values you dont know, its very hard to "keep" them and push them again out, and only change some fields you support

in vcard temp it was more like a simple key:value store

with lets say 20-25 keys, so it was not hard to support all of them

now you have various "parameters" for all the values, there are "groups" which group values

im working on implementing this now, i keep you posted :)

lovetox: you'll need to support something like this for bookmarks2 anyway :p

no pep. i dont think you can compare that

in bookmarks2 there is a dedicated element which is an extension

you save it away and you are finished

in vcard this would be insanely more hard

lovetox: same issue when you start implementing extensions

very theoretical discussion as there is not even one extension

but yes 10 years, and 20 different extensions and you are halfway in vcard :)

What extensions to vcard-temp have there been?

Zash, JABBER-ID was an extension compared to vCard 1 IIRC.

I’m not aware of anything else.

Not sure that's the kind of extension lovetox was talking about?

JABBERID* sorry.

Ah, I only read the very last message before replying. :-°

Probably not then.

im not talking about extensions, pep. talks about extensions

JABBERID and DESC and whatever were additions compared to the never finished vcard xml variant of vcard3

Right.

im saying the vcard4 spec is big, supports many things, and i consider it very hard to support a subset

without losing half the data on every update of course

well I already know of 1 non-standard one :x