On Sat, 19 Nov 2022 at 18:31, Dave Cridland <[email protected]> wrote:
> On Sat, 19 Nov 2022 at 16:33, Daniel Gultsch <[email protected]> wrote:
>>
>> The <inline/> stream feature wrapper is just a neat wrapper for all
>> stream features that can be inlined into SASL to announce themselves.
>> Yes ISR works without inline but that's not the point.
>>
>
> Well, that wasn't in fact the point I was making there - I meant the 
> <inline/> element within the TLEs.
>
> But you're right, neither adds any actionable information whatsoever, so 
> isn't useful and therefore should not be added.
>
>>
>> Imagine we want to make compression inline-able. Simply by looking at
>> the <compression/> stream feature a client doesn’t know if compression
>> can be inlined into SASL. So if we want to make compression inlinable
>> the new XEP (or an amendment to the existing XEP) would have to either
>> announce a new stream feature called <compression-inline/> (This is
>> kind of - but not really - what ISR is doing) or place itself in the
>> <inline> wrapper like this: <inline><compression/></inline>
>>
>
> OK, so compression, let's take that as a worked example.
>
> A client absolutely knows that XEP-0138 isn't a XEP-0388 extension, because 
> it isn't.
>
> You cannot simply advertise it in an <inline/> block and expect it to work, 
> either. When would compression kick in? There's no specification that tells 
> you. It might be obvious to you - it might even be obvious to me - but it's 
> not specified, and my obvious may not match yours. So you need to define what 
> that means, and it's a change in wire protocol, so implicitly that means that 
> something new has to be advertised.

I agree, you cannot just advertise anything in an <inline/> block and
expect it to work. Nobody thinks that. Every protocol that supports
inlining has specified exactly what that means. E.g. XEP-0198 has been
updated with an additional section to cover this.

Now I'll add that your suggestion also doesn't work: you can't just
spec something and then expect it to work. Just because we update
XEP-0198 with a section that describes how it can be inlined, that
doesn't mean every server supporting SASL2 automatically understands
XEP-0198 inlining. And as you pointed out, things may change over
time. In other words, we *need* advertisement and discovery for this
feature.

>> It's just a syntax modification that some people find more pleasent;
>> not a new addition to the XEP to achieve something that wasn’t
>> possible before.
>
> Oh, but it carries all manner of implications.
>
> Supposing that we change the wire protocol of XEP-0198 to support directly 
> using it as a SASL2 extension. So that means advertising the same stream 
> feature under <inline/>, by this design. So far, so good.
>
> But supposing that the SASL2 extension part needs changing - in that case, we 
> need to bump the namespace of that - but we don't want to break compatibility 
> with the old-style stream feature. So now we have to advertise a different 
> stream feature if you're inlineing it, and then if the main one changes but 
> the ... you get the drift, I hope.

I don't really understand this concern. A XEP can specify anything to
be contained in <inline/>. If the inlining portion of a XEP (only)
needs to change, then of course it will need to advertise that
somehow. <inline> doesn't change that either way.

> So repeating the same stream feature inside a SASL2 stream feature as 
> <inline/> isn't actually what you want.
>
> You do, genuinely, just want a new stream feature.

There's no rule that you have to repeat the same stream feature inside
<inline/>.

> Incidentally, if SASL2+BIND2+etc take over the world, then you'll be 
> repeating all these stream features and/or having all stream features nesting 
> eternally inside each other.

I don't see how <inline/> adds any need for repetition. If <inline/>
didn't exist, everything still has to indicate in some way whether it
can be inlined or not. Sure, we can have <compression> and
<compression-inline> features perhaps, but I find <inline> is a much
cleaner and more logically obvious solution to this issue.

Regards,
Matthew
_______________________________________________
Standards mailing list
Info: https://mail.jabber.org/mailman/listinfo/standards
Unsubscribe: [email protected]
_______________________________________________

Reply via email to