You need a stream parser like expat. No need to code your custom parser and once you settle at level n+1 in the XML it is just like parsing independent document.
But again, the parsing is hidden by existing XMPP library, so you do not even need to go to that level of details.
You should compare it to other messaging protocol. You may change opinion, as it is actually elegant, modular and quite versatile.
It managed to adapt over 25 years which I think it is quite a testimony. All messaging apps from that time disappeared.
TL;DR: Matrix protocol is based on document replication and synchronisation, while XMPP is based on message passing. Matrix requires to constantly replicate and merge the data. It is an intended feature, but as it does more, you cannot expect to scale to the same level.
That said, we want ejabberd to be able to suit all types of messaging needs and we have already implemented 1-to-1 gateway with Matrix server inside the protocol. We are fine tuning our chatroom interop with Matrix. It means that we have already most of the code in place to support Matrix chat (as we need to implement the replication part). It means, the next step could be to implement full Matrix support for those who would rather use Matrix clients.
Disclosure: I am ProcessOne Founder, XMPP developer since 1999 and working on ejabberd since 2002.
Yes, I confirm you can build the same starting from ejabberd. A lot of changes are related to the ability to scale horizontally only some subcomponents, to fine tune the processing power, and also things that are related purely to Whatsapp infrastructure.
I have been funding Murena, because it’s probably the most ambitious and challenging project that aims to make the world better.
We need to make the tech an opportunity again for people, to make it work for the people.
Exploring the 'Commons' in the digital economy. I discuss the French government's endorsement of a proprietary messaging platform, highlighting a deep misunderstanding of key digital dynamics.