jdev - 2020-01-25

I've never touched the functionality. It was put in by Jive, at the time. I think we used it with Nimbuzz, at the time. Must've been 2006 or something. I never get questions around it though.

https://bit.ly/DateMeDaddy

IIRC ejabberd also supports extended stanza addressing and MAXS makes use of it

Hello. Slightly off-topic, but I'm failing to join this room using my regular client (currently using the web interface on xmpp.org), apparently my server fails to connect to muc.xmpp.org (208.68.163.210:5269); is it just me, or a common issue? And I've only tried that after failing to subscribe to the jdev mailing list (HTTP 403 there), is it also a common issue?

defanor-tmp: hi

Not a common issue, pretty much everyone here is joined from their own server

Is 208.68.163.210:5269 the correct address? I failed to connect there directly from my computer (different network/ISP/country) too.

I can check jdev subscription is working later

Not sure of the IP (I'm currently on my phone so can't easily check DNS)

And netcat from the server also says that connection times out.

Oh, 5269? Pretty sure that's the wrong port at least

Thanks, I'll try to investigate why that happens then. It's Prosody that picks it.

"dig srv _xmpp-server._tcp.muc.xmpp.org" says "_xmpp-server._tcp.muc.xmpp.org. 900 IN SRV 1 1 9269 xmpp.xmpp.org."

Prosody would always look up the SRV first, and would only use 5269 by default if that fails to resolve

Debug logs from Prosody would help

Hrm, though not from the server. Apparently an issue with DNS there, Prosody is what I'm using.

It was a DNS issue indeed, the server tried to use ISP's DNS instead of a local one. Which for some reason didn't serve SRV records, and somehow Prosody didn't pick it up from alternative DNS servers (which do serve those).

And the on-topic question I had is about reusable XMPP client implementations. I keep looking for relatively complete (in the "advanced client" compliance suite sense) libraries providing C API, implementing XMPP correctly, and embeddable into a program's main loop (instead of using their own or, say, glib's), but not finding much. I also keep wondering whether it's viable to implement it in a way that would allow to define the core and

each XEP separately (in the hope to achieve a relatively clean/maintainable implementation): apparently it can be (and sometimes is) mostly achieved using some kinds of hooks on stanza (and other depth-1 element) sending and retrieval, but it still wouldn't suffice for the "direct TLS" XEP (and possibly some others). Are there known attempts to do that, and/or is it known to be futile?

i would not do that

there are XEPs that make sense as its own module pluginable etc

but others absolutely not

like 0368

nobody will want to disable that

Indeed, that's the "direct TLS" one I've mentioned. It made me to think of hooks on all the I/O actions instead of just XML element sending/retrieval, but then it'd probably be a mess of its own.

i just rewritten a library

and most clients would want the library to pass the domain, and the library should connect and do everything on its own

there are still many things to consider

I also agree with that (having worked on a number of libraries)

The way XEPs are split does not necessarily match the ideal split in a library

as a library author: don’t make the low level stuff plugins

it’s a terrible choice which only leads to heartache. instead, make your core modular so that people who want to experiment with low-level stuff (like '386 or compression or something) can easily do so in their own fork

99% of users won’t want to do this

they just need additional payloads for standard elements -> make *that* easy

- DNS resolution and connection priority (suprisingly complicated) - Happy eyeballs - Proxy Support - Other connection method support (e.g. Websocket) - DIRECT TLS, START TLS, Unencrypted - Support for Client certificates - Support for functioning with non- valid server certificates and thats just out of my head, and thats all *before* you send a single stanza

I see, thanks. I'm not writing a library though, since it seems rather wrong to add yet another one, given that there's quite a few already. Yet finding it puzzling that there's nothing in sight that could be reusable from arbitrary languages via C FFI, complete, and hookable to an external main loop (say, glib's/GTK's, libev's, or whatever the program using that library would want to use). Am I missing something, and/or does such

flexibility seem like a bad idea?

defanor, I guess the C libraries could be made to be usable that way

I forgot the name of the library used by profanity, but it’s in C

howevre, working with XML and text in C is a PITA, so nobody wants to do that

and people prefer to write an idiomatic library in whatever more high-level language

libstrophe, but IIRC profanity struggles to introduce even XEP-0198 for years with it

which makes sense to me -- a native library is more likely to be idiomatic and useful for the language ecosystem either way

oh and yeah, libraries created before the widespread use of '198 are likely to be in pain (looking at you, libpurple)

'198 is one of the things you absolutely don’t want in a plugin

Yeah, I used to use (and contribute to) libstrophe

i've added 0198 into libpurple more than an year ago, but it's still not released

it’s hard enough to get right even when having full access to the stack’s internals

jonas’: FWIW I've implemented it in a plugin on the client and server side, I don't think it's that bad

defanor, 0198 its not "hard" or "complicated"

MattJ, mod_smacks isn’t that bad? ;)

but a library should have that in mind when written

otherwise it gets almost impossible if you have bad luck

> lovetox> defanor, 0198 its not "hard" or "complicated" I am not sure i'd put it that way given that most implementations of xep198 ran into the same set of issues at first

Yes, that's what also seems awkward about relying on core modifications to implement XEPs.

jonas’: mod_smacks has issues (but it works), I don't think they necessarily stem from it being a plugin

The hard thing about 0198 is counting!

counting the right things

I disagree, we haven't had a counting bug for a very long time

"1 stanza", "2 stanzas", "3 stanzas", "oh wait let's do something else" ... "2-and-a-half stanzas" ... "damn how many stanzas have I seen again I'm lost!"

The XEP says what to count and (now) when counters get reset

yep, the xep198 clarifications that where added likely helped to improve the situation

How do you get distracted during stanzas_received++? :)

Threads and stuff.

No idea actually :-)

I'm just not sure I've ever seen an implementation that got it right from start. And I think at least ChatSecure is still borked.

We did have bugs in the early days, but again I don't see that they were caused by it being a plugin

I am not sure if what holger described actually borks the implementations, but it sure was/is identifying the point where to reset the counter and start counting

MattJ, same

I think the main reason it shouldn't be a plugin is for the library maintainer's sanity

flow: That was one multiple problems I've seem implementations running into, and it's good this one has been clarified in the XEP, yes.

*one of

It's hard to ensure you have the right plugin API and all the plugins interact correctly

But e.g. both ejabberd and Conversations were running into threading foo that was unrelated to "when to start counting", initially.

Threads are the root of all evil

MattJ, one thing which went "wrong" in aioxmpp was that a stanza which caused an exception which terminated the stream (really bad) during processing would not be counted ✏

this would cause the stream to reconnect and resume

causing a fun tight reconnect loop

MattJ: Yeah.

admittedly, many problems at play here, but still.

jonas': Exactly, stuff like that.

(reconnect loop because the stanza would be re-received during resumption)

jonas’: I'd argue that not counting a stanza you haven't processed is sensible from a data loss perspective

MattJ, define "have not processed" please

In Prosody we are moving 198 to core, it's my ideal that we wouldn't back until the stanza is committed to storage or delivered, for example

Not sure we'll achieve that, it's Hard

if I auto reply with <service-unavailable/>, is that "processed"?

Sure

if I reply with (I don’t recall what we do) because of a deserialiser error, is that "processed"?

because that’s exactly what (should’ve) happened in the above example (that the deserialiser error broke the XMLStream instead of properly being handled was a different issue)

If you reply of course it's processed

(scrolling through `git log aioxmpp/stream.py`)

MattJ, I mean, what you say makes sense, but it’s hard to get that into code right from the beginning

there are dragons and nasal demons very close due to fun race conditions (even without threads)

maybe race condition isn’t the right term, but edge case is.

https://github.com/horazont/aioxmpp/commit/f9af3cb485e9b6d7bc0d76d10e919aa76e822872 that’s the one

oh, no, that’s a different one

that was when parsing errors didn’t even propagate to the SM implementation

finding the right place to count "stanzas" wasn’t easy, and it took more than one iteration

what impl get wrong mostly is not how to count

is when to start

and its not when you receive the first stanza

its also not after the server sent you <enabled>

which probably is the most common pitfall

what?

I’m pretty sure you start to count inbound after you received <enabled/>, and outbound after you sent <enable/> ✏

which is the only way to make this reliable

jonas’, probably lovetox was talking about counting outbound stanzas ✏

flow, I got confused by "its not when you receive the first stanza", sounds like inbound to me

Guys i need your help

.setXmppDomain("admin@192.168.1.9") .setPort(5280) .setHost("192.168.1.9") .setResource("Rooster")

when connecting with client it showing error as (client not connected to server message not sent ) how to resolve it?

what library is that even?

smack library

java?

yes

flow, ^

though one thing which stands out right away is that using bare IP addresses is asking for trouble.

and a nodepart passed to something called set domain

and the standard bosh port

that’s probably not good either

and also everything

DomainBareJid domainname = JidCreate.domainBareFrom("192.168.1.9"); XMPPTCPConnectionConfiguration conf = XMPPTCPConnectionConfiguration.builder() .setXmppDomain("admin@192.168.1.9") .setPort(5280) .setHost("192.168.1.9") .setResource("Rooster")

i have my server in another system

how to set connection with server and client(mobile)

yeah jonas’ but thats counter to most of other XMPP flows

normally you send a IQ get, receive a set, then something happens

or you negotiate something with, <xxx> and get back <succeeded>

etc

and thats probably the first intention when implementing smacks

we enable, wait for response <enabled> then we start doing our thing

hi everyone. not sure if this is the right place for the questions i have, but maybe you can give me some pointers. i'm trying to build an xmpp client for my organization (web and mobile), on mobile i'm using smack for android which is great (haven't started ios yet...). on web, my search has led me to strophe library, which sadly i coudn't even compile (dirs and files unexistent) and xmpp.js, (node-xmpp rewrite) which i'm currently testing, the basic demo covers connection, send and receive messages, pretty useful, but i'm struggling on finding documentation on some other topics (roster management, file transfer, etc.), do you know and recommend some other library?

Hi xmppnoob

Take a look at converse.js <https://conversejs.org>

strophe.js is generally a low-level library, so there is a lot of DIY involved, though it does have a bunch of plugins that handle some tasks for you

converse.js is a more featured client on top of it, which handles lots more stuff, and is easily customizable/embeddable

Also if you're working on clients, take a look at https://docs.modernxmpp.org/client/protocol/ which has a bunch of relevant notes

thanks MattJ, i'm a little confused about conversejs, it is a client or a library?

It's primarily a client, but it has an API that you can use to control/customize it with your own code

great! i'll give it a try