XSF Discussion - 2019-02-05

How bad would it be to s/http:\/\/xmpp\.org/https:\/\/xmpp\.org/ in the xeps repo? Do I have any chance of breaking normative language

You'll find that one place where someone used http://xmpp.org in a xmlns

I feel http://jabber.org is used most often (when not urn:xmpp), but yeah I guess

...

xeps$ ack http://xmpp.org/protocol inbox/sxe.xml 262: <query xmlns='http://xmpp.org/protocol/disco#info'/> 268: <query xmlns='http://xmpp.org/protocol/disco#info'> xep-0284.xml 268: <query xmlns='http://xmpp.org/protocol/disco#info'/> 274: <query xmlns='http://xmpp.org/protocol/disco#info'>

grep

*great

pff

That's gotta be a typo for jabber.org/protocol tho?

Ok I'll be careful then

Probably in examples?

let's see

xep-0397.xml has a bunch of xmlns="https://xmpp.org/....

that's fine, I won't change this one :P

But yeah that should be fixed

https://xmpp.org/extensions/xep-0284.html#host this points to xep 30, so looks like a mistake

With a bit less effort we should probably add HSTS first

The 301 is already there

What about: "<p>The &REGISTRAR; shall add 'urn:xmpp:idle:1' to its registry at <link url='https://xmpp.org/registrar/namespaces.html'>https://xmpp.org/registrar/namespaces.html</link>.</p>" ? Is that normative? the registrar uri

PR'd

(I changed to https in this quote)

It is technically a completely different URL, that doesn't have to have the same content as the http: URL

err, I don't have access to the nginx that should do HSTS, do I

This docker thing only serves 80

<meta name='horrible-hack-no-372' http-equiv='strict-transport-security' value='max-age=infinity'>

:/

Technically I can already do that with the nginx inside, but that will send it on both http and https, and it's invalid on http (or just useless?)

wasn't 80 just serving a redirect?

I should be able to check for x-forwarded-proto maybe

Zash, 80, as seen from the outside, yes

Not from the container

so

the outside won't see hsts on port 80?

So?

HSTS SHOULD(MUST?) only be on https

so if the container spits it out on port 80, which is reverse-proxied to port 443 with tls

then it should all be in order

Browsers will ignore it over http

It will also spit it on http. But then maybe as I said I can check for X-Fowarded-For

moparisthebest, sure

So it's fine to send it either way

browsers are as compatible as one can be, because the web if full of crap

The point is they never visit http again anyhow?

I just like to do things properly :x

Don't forget to set up a redirect to https from http, and put it in the preload list :)

it's already there

pep.: if the public port 80 is configured to only s/http/https/ then I don't see how anyone can get hsts via http

Do things properly, the web, bahaha :)

So I might as well ask iteam directly

And not put this hack in the xeps repo

Zash, true.

Still that's probably not something for the xeps repo

Maybe the website, but I feel that should go in the host's conf

Is there a container that contains the conainer-container that holds the 7th proxy?

Anyway, I should stop procrastinating procrastination ✏

It will be nice to have same base for all without www. etc

I'm really tired of refusing OMEMO requests

I'm thinking of hardcoding the body detection and starting to reply automatically

edhelas: just blacklist the OMEMO node on your PEP

whats a OMEMO request?

> I sent you an OMEMO encrypted message but your client doesn’t seem to support that. Find more information on https://conversations.im/omemo

tell your contacts to disable omemo?!

yes, one by one :) if at one moment they move to conversations

lovetox: that's O(N*M)

with N=number of contacts and M=number of devices each contact has

I'm not there to do "remote xmpp client configuration over human written messages"

hm in Gajim you can deactivate publishing keys to the node

but yes this problem exists because you use a client that supports omemo ..

because you once used such a client

if you are on prosody < 0.11, you can restart your server to "fix" it

yes I do use Conversations sometimes, and Movim 95% of the rest of the time

pep.: there are some, mostly deferred/retracted, XEPs that use http://xmpp.org/ as the start of their namespace. I'd leave them be. If we ever resurrect them, they'd have to go with a urn one anyway.

fastest way would probably to set the devicelist node to whitelist

edhelas: write a Movim plugin that subscribes to the siacs namespace and removes all devices

or what lovetox said

Ge0rG that does not help

as devices are instantly republished

you can configure nodes in Gajim 🙂

pep.: and probably everything that still has shtml in a link should get looked at.

edhelas: does movim support Explicit Message Encryption?

vanitasvitae I as looking into that

You could use that to determine, if a message is omemo encrypted or not

Might be better than to check the body

conversations is adding the tag ?

Pretty sure it does

ok

lovetox: how do you configure PEP in Gajim? I've opened the "Configure services.." menu on my server, but there are no items listed and the buttons are greyed out, except for "Close"

because your server does not support configuring nodes

lovetox: would be awesome to have that displayed in that box

yes once i rework the dialog, its still from ancient times

Ah, the Good Old Times!

What happened about the LC on 345 BTW, did I miss any decision made about it?

pep.: I suppose the Board needs to vote on advancing it next.

Or maybe somebody can make a PR incorporating LC feedback

dwd: I would kindly ask you to put https://github.com/xsf/xeps/pull/752 and XEP-0410 advancement on the next Council agenda. cc jonas’

Ge0rG: with the Summit happening, we haven't had a meeting to discuss the LC feedback on XEP-0345 yet. Thursday is the first opportunity.

pep.: ^

There was also a Board PR that Council was asked to look into.

hey there. I was thinking this morning, it would be nice to have a QR code with my jid in it on my website, so people could get it easily from phone, and it would be nice to have a way to make it easy to recognise. It's possible to put an image inside QR codes, so would not it be nice to agree on something (e.g. putting XMPP logo inside) so we could know immediatly "OK this is a XMPP identifier"? Would it make sense to have an informational XEP on this?

goffi: yes, you can increase the FEC on QR codes and then just bump some picture over the middle

sounds like a fun little shareable service.

Ge0rG: FEC?

(forward error correction)

Wasn't it "the image isn't part of the standard, it's just ignored with error correction"

goffi: forward error correction. Also related: https://github.com/ge0rg/easy-xmpp-invitation

hah, right

although I’m not entirely sure that’s the correct term for QR codes ✏

I know it's possible, that why I'm proposing it, we don't need much characters for a jid.

jonas’: but it still is FEC.

ok

just because people invent new fancy names... 🤷

goffi, what do you want to do with "a JID"?

Add as contact?

pep.: yes

not only on website, on visit card too

goffi: what if the person scanning the code doesn't have a Jabber client yet?

I just want a way to recognise immediately "oh this is a jid"

goffi, have a look at what Ge0rG pasted

I'd put a Jabber lightbulb in it.

We can probably just put a QR code there

Ge0rG: it's the same issue with a xmpp:myjid@example.net link.

https://yax.im/i/#vladimir@xmpp.example

my point is just to have a way to recognise the QR code as XMPP related.

goffi: I've heard you are good at webdev.

using http or xmpp is an other thing. There are issues with http (you have to keep a service, and it won't be catched by registered client). But that's not the point of my idea, I just wonder if it would make sense to agree on a way to recognise XMPP QR Code (image inside is one thing, we could also play with colors).

> it won't be catched by registered client yaxim will catch both yax.im and conversations.im landing pages

Ge0rG: what do you mean ? I'm working with web sometimes yes, and on a XMPP web framework.

I'd have a QR code with the XMPP logo (to add to the bikeshed discussions)

goffi: the easy-xmpp-invitation project is looking for volunteers 😁

Ge0rG: and what I have gajim ?

goffi: how do you scan images in Gajim?

Ge0rG: I would love to help on that, but I'm already close to burn out with my paid job + work on SàT + all side stuff (like preparing FOSDEM)

goffi: alright, thanks anyway :)

Ge0rG: I was mentioning gajim for the contact link, a xmpp:toto@example.net will work there, https://yax.im/i/#vladimir@xmpp.example wont

goffi: on my OS, Gajim doesn't handle xmpp: URIs

is it possible with QR code to have 2 links ? A http: one and a xmpp: one ?

nope. But the landing page contains an xmpp: link ;)

what if the size is down ?

site*

Then it's down I guess

xmpp: will still work. I mean I get the point with https: link, and I agree on most of them, but the xmpp: is the standard way of doing, it's not linked to an entity, in short it has many advantages too.

it's not linked to a client or service*

goffi, what if your xmpp server is down

goffi: it's got many advantages, but it doesn't work for normal people.

Will the other server resend the subscription request(?)

pep.: it won't prevent you to add the contact, after the subscription will happen when it will be up again.

goffi, ok. I wouldn't mind having the xmpp: uri in there, but then you only target people already present on xmpp

best way would be to find a way to have both in a easy way, but doesn't looks like possible.

(I agree we're mixing two issues, but I don't know how to entangle them)

have two qr codes...

just to make it more complicated.

mtavares, I think that's be even more confusing

using a http link works well enough, me thinks.

"So which one is it??" "It's the same" "Why do you have two then?" -- my mom

https://qr-creator.com/multiple-url.php

it seems possible

ahh.. interesting.

goffi, it's just text right, you put whatever format you want in there

pep., QR codes have multiple subformats for different payloads

I see

partly because they use different codes for different payloads to save bits

And of course this website only accepts URLs

not XMPP URIs

ah it's just redirecting and shows the 2 links

pep.: yes I know, but the QR code readed need to know how to handle both.

it's just a https link to qr-creator.

mtavares: yes, so not good :(

https://op-co.de/tmp/georg@yax.im-bulb.png

https://op-co.de/tmp/georg@yax.im-xmpp.png

I like more the second one

it's neat :)

yeah me too

This is the moment to complain about our well-known marketing name being absolutely burned, and our official protocol name being too technical for regular people.

Maybe just remove the "XMPP" from there

So you don't have to pronounce it.

yeah the logo only is fine I think

also I think the text could be used for something else.

for instance, would it make sense to distinguish QR Code depending on their use: for instance a contact to add, or an ad-hoc command to execute, or a pubsub service ?

so you would but the XMPP logo, and "contact" under it instead of "xmpp"

did I mention that normal people in our region consider QR codes as something awkward?

why ?

dunno

nearly everybody is getting use to it, it's a nice techno

http://picturesofpeoplescanningqrcodes.tumblr.com

I have it on my applications as vCard. I still hope some technical employer finds it nerdy.

ta: do you have a utm tag in it?

well, i don't have a link in it, so no...

I mean application as in paper/pdf. QR-code being just contact information as vcard.

But i should consider tracking when applying at some online shops. they like everything i try to block ;-)

woever wrote me a pm here, i cant answer because converse does not support pm 😃

<-- is just asking about the gajim muc not beeing reachable from some servers.

ta, would be interested what the error is

im joined and im on jabber.at currently

if you guide me. gajim and xml console didn't give me much feedback.

you can guess my JID, so we don't spam here. I am using trashserver.net (no comments on that name ;-)

but it should, restart gajim open xml console, join the muc and maybe wait for 1 minute, whatever the servers timeout is, at one point your server should inform you why its not possible to connect

otherwise ask the admin

lovetox: Looks like a IPv6 problem, please try to fix it.

can you be more specific

im not the server admin, but i can tell him

If only XMPP servers could gracefully support both IPv4 and IPv6

lovetox: traceroute to conference.gajim.org (2001:bc8:342f:101::1), 30 hops max, 80 byte packets ... 5 * * * 6 2001:bc8:400:1::8e (2001:bc8:400:1::8e) 27.964 ms 28.068 ms 28.263 ms 7 2001:bc8:400💯:26 (2001:bc8:400💯:26) 25.912 ms 26.035 ms 26.204 ms 8 * * * ...

Ha. Blame Link Mauve

Zash: is he behind gajim.org?

No I’m not.

No, why I bumped a timeout back up to the point where it gives up completely before trying the next IP

Zash: you bumped Link Mauve?

can't you unbump him?

Link Mauves toaster was too slow for the low timeout

Well is this the cause already? I restarted gajim, have a very spammy xml console...

Zash: can't you also increase the *other* timeout then?

Zash: it's a bit shortsighted to break connectivity for all users of prosody just because of Link Mauve

ta i reported it to asterix but will not be fixed immediatly

Actually, this specific configuration still works because the write timeout here is 60s and the "give up completely" timeout is 90s

But ... this varies

Hi all. Do we have an incremental update for disco#items and things? I've a feeling that Sem Whited documented something that HipChat did a while back but I can't find it.

what do you mean by incremental update?

lovetox, thanks. No hurry. I was just surprised that nobody "cared" in all the other XMPP-centric MUCs ;-)

So if I have cached, for example, all the disco#items, and I now want to find what's changed instead of downloading the entire list again.

Sorta like XEP-0237 for Disco.

lovetox: The prosody module mod_bidi might help a bit, since it reduces the number of connections that can fail

dwd: Isn't that caps?

Or you want caps on ... things?

Or disco-sub or whatsitcalled?

Not caps. Though caps on ... things would be nice, actually - good plan.

But caps doesn't do disco#items.

Right

Did caps2 do that?

I'm sure we discussed some semi-recursive caps thingy recently

But yeah, if I have 1.2 quadzillion MUCs, I'd like to just fetch the three new ones and not the 1.2 quadzillion over again.

caps isn’t really incremental, either

Yeah. Figuring that out would be nice.

jonas’, Yes, indeed.

disco#items versioning essentially

I can't recall if we decided that Sam's proposal should be a XEP or even if he submitted it.

I was leaning towards not returning complete results in disco#items at all and defering to some search interface

I don’t recall anything in that direction

RSM!

I only recall that there was "disco-(pub)sub"

But otherwise, I'll do a ProtoXEP based around '237's design if there's interest.

Ge0rG, RSM is good for paging, less good for incremental update.

RSM is also weird if items can be removed

dwd hipchat has a couple of features documented here https://docs.atlassian.com/hipchat.xmpp/latest/xmpp/rooms.html

the room push comes closest to what you describe

Or... I could do it based on pubsub, and then list disco#items things in a pubsub node... somewhere.

(it also does something with service discovery - but that does not seem to match your description)

dwd: But if you have 1.2 quadzillion MUCs, having a cache of them locally is probably not fun anyways

dwd: did we solve the multiple-items-per-node pubsub problem yet?

Ge0rG: What problem is that?

Zash, True, but I have a way of having a set of MUCs I'm interested in anyway. The problem is that I have a lot of them.

Zash, Also, caching 1.2 quadzillion items is not fun, but fetching them anew every time is even less fun.

dwd, wouldn’t some type of search interface make more sense then?

Perhaps, but I do have a few cases where I want all the disco#items.

dwd: I am not sure, but you are maybe looking for https://xmpp.org/extensions/xep-0366.html

But I guess for a delta sync of disco#items you probably want something like disco#items versioning…

How many implement the incemental roster versioning thing?

dwd: having a pubsub node to represent available channels would definitely be interesting.

dwd: with RSM, it would be much easier to keep lists on the client side in sync.

We kind of already have a specification for this with `pubsub#subscription_type` with the value of `nodes`, and using that in subscription options towards the root node.

dwd: did you receive my message further above regarding Council Agenda additions for tomorrow?

https://xmpp.org/extensions/xep-0248.html#notify-items suggests there's a `<create/>` element in pubsub#event to be used for notifications, but the schema in XEP-0060 doesn't current have that.

I have no time to check it now, but is not this RELOAD protoxep doing something similar to jingle?

Based on what dwd wrote, I wonder if it's rather something for the IETF?

Too bad the date on the inbox entries are all from yesterday.

goffi, to me XOR looks more like 0174 (Serverless), but with NAT traversal

XOR depends on '174 AFAICT

OK, I've just overlooked and it looked like a way to establish a stream.

interesting then, I'll check that later.

Ge0rG, I've not noticed it, and would welcome an email. Travelling tonight at short notice, and I'll try to do the agenda stuff on the train.

dwd: sure thing

So, XEP-0366 is more or less what I was after. I think including caps, though, is probably interesting too.

"I have stuff with caps hash xxxx, what's changed since?"

https://tools.ietf.org/id/draft-faltstrom-unicode11-07.html

oh no

> Appendix A. Changes from Unicode 6.3.0 to Unicode 7.0.0 > Changes from derived property value UNASSIGNED to either PVALID or DISALLOWED.

In the context of RELOAD, I'd suggest defining an XMPP wire-format to exchange X.509 CSRs and CRTs, which could be immediately plugged into a XMPP JID-identity CA

Zash, No, caps can be an all-or-nothing for disco#info. But it'd save me a huge amount in this particular instance. I think an entityver thing would help with disco#items, though.

MattJ: so RFC 7622 will need an update?

ralphm, thinking about <gone/> again for <moved/>, and MUC affiliations/PubSub subscriptions. I can't just return <gone/> for everything that comes pinging my account for privacy reasons. And my server doesn't know (could know, but allegedly doesn't track) these subscriptions, so..

Well, if you want to leave some kind of tombstone, but only to selected people, you kinda have to keep the former relationships, whichever way.

It will work for contacts, because we have the roster

But that's about the extent of what the server keeps track of, in the general case

I suppose for pubsub subscriptions it might be harder indeed.

and MUC affiliations

Not even talking about server shutting down etc. yet

But this is one of the reasons why dwd suggested PAM, so that your server keeps track for you. When you move, you could take that with you.

(I'm thinking about not considering this use-case for a first rewrite of <moved/>)

PAM?

https://xmpp.org/extensions/xep-0376.html

pep.: the hard question with roster is how to tell multiple clients of your friend, if the first client will replace unsubscribe your old JID.

Ge0rG, yes I got that, you don't have to repeat it over and over, that wasn't the question

I know we need a tombstone of some sort, and that was discussed

pep.: This came up at summit, about needing to keep the roster, or maybe some magic hash of it

Zash, "this"?

you could create a PEP node with a whitelist populated with your roster when moving.

pep.: what's the question then? Who is allowed to see your tombstone? Either everyone or old-roster-only, I suppose

Ge0rG, no that wasn't the question

Ah well, yes sorry

But not for the roster

I'm asking for muc affiliations and pubsub subscriptions (read the question)

ah!

so couldn't you whitelist the muc services on your bookmark list (private storage or PEP) and the pubsub nodes you have received notifications from?

there is no way to enumerate all your MUCs and PubSub subscriptions, unless your PAM keeps track of them all, from day 1

Using bookmarks for MUC seems interesting

But yeah for PubSub we'd need PAM or sth

bookmarks probably contain the relevant subset (what do you care if you are affiliated with a MUC you never joined)

*also*, besides leaving a tombstone, there's no reason why you couldn't send a presence with forwarding address

presence is ephemeral

not for muc rooms

it might suffice to let PubSub and MUCs know.

i.e. you only need to have it received once

but you don't know where to send, in the first place

Technically the server can know

But yes that has to be done from day1

wouldn't you likely cover this by the list of muc servers that you frequent rooms of?

unless 100% is your goal, in which case good luck

gateways & transports are another topic

I would hope services GC after some time being unresponsive

a tricky one, actually

I wouldn’t want to loose my biboumi MAM because I moved accounts

hmm.

We're really trying to sort all the problems in the world at the same time :/

This thing is never going to progress

I’m calling it "enumerating"

I don’t expect your proposal to solve all of them

jonas’, I guess it's time for a client-side MAM? :/

but it should enumerate them all, and clearly distinguish between those which are solvable and solved, those which are likely solvable and deferred and those which are not likely solvable

client side MAM?

So that you can just ditch the server archive once you've fetched it, and then have your mobile ask your desktop client for archives

yea, no

Why not

I don’t want to implement the server side of MAM things in a client

huge complexity, especially with Smart MAM ahead

True..

(and how about the clients which have the "I don’t want to store everything locally anyways" philosophy)

It's really not the first time I hear about it though

yeah, you hear from it mostly from crypto fashionistas

Sure

who want to solve the PFS problem with a huge pile of complexity

...and then end up talking with people who just dump everything into sqlite

Anyway, this is not a crypto party, the topic is different

but you can't just just expose your new JID to every entity that claims to be a pubsub node.

in any case, client side MAM won’t solve the biboumi case

you don’t want to have that all stored on your phone anyways

jonas’, it would yes? I think

I see this problem as moving house. You let your friends know. In the Netherlands you have a service that informs (selected) entities of your new address so they can update their records. Also, the occupant typically temporarily forwards stuff to you.

(new occupant)

You never reach 100%, and that's fine.

Right. So we're inevitably going to drop mail, but yeah it's fine

They'll recontact us, (or not)

And the rest of it is usually a thing you solve socially, out of band, not technically.

Ge0rG, would you be ready to compromise on that <moved/> thing, or do you want to get all the use-cases into the one XEP

If so, I'll split efforts because I'd like to have something ready by 2025 not 2030

(I'm being generous)

pep.: you know, I was like you, back then in 2018. I thought "how hard can it be". But then I started work. And I realized...

So, what exactly do you want to compromise?

pep.: privacy problem of gone, needing to keep the roster

Also persistent PEP with a configurable access model.

I want to drop the "My server is offline" use-case. I am not sure I'll be able to support "My server is scheduled for shutdown" entirely either, because we need tombstones

But I guess if your server shutdown is planned in 6 months time, that's already quite enough time to migrate most of your contacts

pep.: do you want to drop muc affiliation and pubsub?

I don't know. At the moment I see it as something the server will have to keep track of for us, unless you have a client-side solutions

Do you want to support users migrating from prosody < 0.11?

non-persistent PEP?

No I don't care

People please update.

Well, there is MAM, I can do with that if you really want

A cryptographic identity overlay network that works over XMPP as the routing layer

yeah no thanks

I wouldn't mind trying to make it future proof, as in, compatible with a possible crypto thingy that would come on top, if that's possible. But I'd like to have something out soon

Actually PEP/MAM are not necessary if we do <gone/>, on the old server.

The old server must support it then.

Also access permissions

The old server must support RFC 6120?

I hope it does

Ah ok I see

hmm, we can prompt for support, and if not fallback(?) how is that

> Also access permissions

Is there a protocol to *configure* gone on your account?

Yes, we can prompt for support on the old server, and if not PEP/MAM?

We talked about adding that to IBR, so that'd be a new thing indeed

The problem with PEP is that it needs to be explicitly checked or subscribed to, and works on a client level

So you'd say the default access permissions for <gone/> will be the user's last roster?

<gone> errors in response to probes are much nicer

But do require support from the migrating server

MattJ, sure, they all require support on different level..

*Somebody* has to support *something* :(

pep.: sending messages to all your contacts doesn't.

Ge0rG, yes, clients need to understand MAM, and the moved payload

pep.: I wanted to make a protocol that gracefully degrades if it's only supported by clients on both sides.

How far did you get

pep.: messages with a moved payload from the old and from the new account. MAM / Carbons on the receiving side as faux persistence.

(all these french words)

pep.: maybe a private PEP node on the receiving side where the first client can store the Moved from-to pair

yeah I remember these discussions, and I'm fine with that

I thought that was dealt with already

But then there is a nice race condition on populating that PEP node.

pep.: not written down yet.

Ok, let's do it then

pep.: you want to continue the work? I can commit and push my last sad state.

A compliant server could process a Moved PEP node on your account, and send a gone error to entities that are allowed to read that pep node

Or maybe always send the gone, but only fill the JID if permitted.

yeah that's details to me at this stage

hmm,

pep.: https://github.com/ge0rg/xeps/tree/xep-0283

that's the code behind the rendered version I forgot to tell you before Summit.

k

I wouldn't actually mind only including compliant servers or clients in my assumptions, tbh

Deployment is an issue, and we all know that. My goal is not to solve this. (/me eyes non-rolling release distributions)

but a feature that only requires clients to upgrade will be adopted faster, especially if users want it, because it's fully under their control

But their contacts also need to support it

So maybe not that fast

not if they can fall back

like 'oh you don't support automatic roster migration? ok send message "hi I was x@x now I'm y@y"

with the main reason for this being "abandonden servers" I'd pull the line at modern client support

Ge0rG, it's not my main reason fwiw

My main reason is services like quicksy, or any other public services, which are great for adoption, but then users might want to move away from that

I've heard there are ~250 users on Quicksy.

Sure

"or any other public services"

then it has to be a client-only feature

moparisthebest, why does it have to

public services that don't want people to move away would just never enable it?

especially if they get $ from users, directly or data mining (google)

Yeah I'm not considering evil services tbh, that's another can of worms

I'm not talking about evil

Why would they block then

client-only doesn't work with actively evil either, they can block stanzas etc

Indeed

Any server can just block the PEP node

I'm talking about "should I download and install/configure this plugin that makes it easier for users to leave?"

why would they do extra work for that?

moparisthebest, if it's a PEP node it's no extra work

so that's client-only?

PEP is not client-only no

Neither is MAM. They all require server support

yea but it doesn't require specific server support is all I meant

Yes it does

not for this specific thing

just PEP in general that's used for lots of other things?

Yes yes

So here's your reason why they would set it up

So the feature can require server support

yea that all sounds fine then

I just meant if it required all servers to upgrade before it was useable, it'd be worthless and never adopted

think MIX

Well there are still prosody < 0.11 out there, and it's going to be the case for a while

so you are, vaguely, publishing a pep node as old@old.com saying you've moved to new@new.com ?

Yeah. And that tombstone should stay up for your contacts' devices to see.

I tended to like what Ge0rG called a 'cryptographic overlay' better, it's maybe a bit more complicated, but not too much

I don't mind having that as a goal for later, in a different XEPs, with different goals (including moar uses-cases)

vaguely, you publish a public key, your contacts save this (public key for old@old.com), then if a new account sends a subscription request with a message "I was old@old.com but now am new@new.com" signed by old@old.com's key, you know to update it

not that much more complicated, supports everything that does plus the (I would think) common usecase of a server disappearing

You don't actually need crypto fwiw

Except if your old server is _already_ offline

But then you're probably not in a good position anyway

that seems like a valuable usecase to support though

Yeah, I'm not worried about it for now, but you're welcome to

so is yours, uh, lazy

And you should have preemptively declared your key if that's the case

it relies on all your contacts checking your old PEP node every time?

They can check it once and be done

once every time the client connects or?

Once they've acked it

it doesn't react to being sent a message though?

I don't understand

it checks pre-emptively for all roster contacts?

"it"?

your client

Yeah, you +notify on the node

so when a client logs in it has to check them all too I guess

them?

xep-27 already announces a key even, 'moved' could just define a signed payload for "I've moved"

seems far simpler

> so when a client logs in it has to check all pep nodes for all roster contacts all too I guess

A client doesn't "Check", notifications gets pushed to it, but sure

even notifications sent before the client logs in?

Sure, you add +notify in your disco stuff and the server will resend it (I think?). You'll receive the last event ✏

if I understand it's just adding a bunch more login traffic, at least one extra outgoing stanza and an ack per contact?

vs the key thing, which supports all the same use-cases, plus missing server, with no extra login traffic?

There's no extra login traffic if there's no event. And once you've seen the event on contactX, you can unsubscribe from it.

what key thing

crypto signed "i've moved" roster subscription messages from the contact that moved

But then in my use-case you'd need an extra PEP node on your own account mapping old-jid -> new-jid of your contacts, so that all your devices can ACK.

Then you do that from MAM?

You can also use MAM without crypto. And that's how Ge0rG implements it for now in his draft

isn't the roster stored on the server? can't only 1 client change it then?

technically the server could change it and not involve clients

So what's your use-case then, do you want the server to be involved or not

that server involvement would be optional, on the non-moving client's end?

wouldn't need to be mentioned in the XEP, the server could just optionally do it instead of the client

That can be mentioned in the XEP yes, and we already had something like that in mind.

I tend to think moved that doesn't support a server going away in the future wouldn't be very useful

I think it is

And that'd be my main use-case

my point is that if you do it a slightly different way, it supports both use cases

"slightly" meaning crypto?

:/

sure

you aren't inventing primitives or anything scary, just signing a message and verifying a signature?

You don't need crypto for most of what you just said

If your use-case if "server going away in the future", you don't need crypto

As your identity provider is still alive

ok let me clarify, I mean "I'm gonna go ahead and publish this now, and still want to be able to use it to move if my server goes away in the future"

publish where

either signed presence like xep-27 or a PEP node?

Ok so what about non-supporting contact clients (which is a concerned that's been raised), at the time you published it, until you server went down

Also presence won't work, it's ephemeral

but you could save the key from the first presence you see or something, in your client

If you support it, sure

non-supporting clients I think is the same regardless of method, if they don't advertise support, your client sends them a manual message?

And if you see it, in the first place

(after it moves I mean)

Which is why I just don't want to worry about this use-case. With the current changes we have in moved, nothing prevents you to also publish your signed payload