Modern XMPP project discussion - 2023-09-20

will the "enhancement" of the signal protocol "quantum resistance" also adopted in the xmpp world? https://signal.org/blog/pqxdh/

I suspect moving to MLS is more likely to happen sooner than significant changes to OMEMO

i cant help but feel that is avoiding the problem though :/

Which problem?

it the sense that the problems of lack of implementation, verification and everything else is still gonna exist with MLS

in client implementations that is

MLS has multiple independent implementations, it has formal verification, etc. and is a proper scrutinized standard outside of the XMPP sphere

OMEMO is basically just us

so either all clients use one implementation which is problematic (see webp) and magically everything is fixed. (i doubt it though because clients still need to implement protocol specific stuff)

MLS also adds features, such as proper support for groups

MSavoritias (fae,ve), there are other webp implementations than libwebp.

my main issue i guess is that the xmpp community hasnt shown much willingness to fix the omemo stuff or adopt so moving to mls wont really fix the cultural problems

The two browsers use it, but it is far from the only existing one.

> MSavoritias (fae,ve), there are other webp implementations than libwebp. maybe. from what ive read half the internet was affected anyway

XMPP's problem isn't always the willingness

true. it will fix the implementation part as you said

I think if we could click our fingers and magic OMEMO2 (and OMEMO3) into existence in all implementations, we would

But the reality is that it is *extremely* hard to do that

Not impossible, but hard

im not talking about only adoption of omemo2 though

there has been problems with verification of a lot of devices, old keys still appearing and groups not working for years

or the this message cant be decrypted

Right, because E2EE is super hard, and we're not a community of cryptographers

so i fear that MLS will fix the implementation part

but all the rest protocol/client specific problems will still be there

MLS is the work of multiple actual cryptographers working together to solve the problems that OMEMO has

and get that into a standard that people can use

you mean it solves also the random decryption problems and the too many keys thing among others?

Yes

well lets wait and see then

OMEMO (and Signal) don't *really* support groups, they hack it by pretending everyone is sending to everyone else

i did hear that it has worse metadata protection than omemo

MLS has native support for groups, everything is well-defined

omemo 2 that is

> MLS has native support for groups, everything is well-defined yeah i am aware it fixes groups.

I don't know about that, I haven't studied OMEMO1 vs OMEMO2

MLS has (theoretically) better metadata protection, such as being able to hide group membership

But I don't know if we'll achieve that in XMPP

i do think that trying to fix omemo at this point for groups is probably stupid

when MLS exists

I don't know how it could reasonably be fixed

but i need to read if its worse than omemo:2 or the same. if it can have the same metadata protection i dont see anypoint then even keeping omemo for 1:1 chats

even if it can i dont think i see the point personally

we are just duplicating what is already there

okay so I read the MLS rfc and the omemo rfc a bit. you were right MattJ reading through mls it does have metadata protection. the only thing that it doesnt have is deniability which is not mentioned. but omemo has weak deniability either way.

so Im guessing there is no point then imo to look into omemo for any future projects at this point. ill skip it and go straight to mls

although im not sure what library im going to use either way.

MSavoritias (fae,ve), before a library you will probably want to define a protocol mapping it to XMPP.

I’ve had good feedback from a friend about openmls, but I haven’t used it yet.

of course. but at least i know i dont have to bother with two encryptions now

omemo and mls. and can just go full mls

im guessing a xep will also define the cyphers

Depends whether you want to be compatible with any current client or not.

of course :P