bookmarks are a way for*users* to hold a list of chats Certainly, but there are valid use-cases for entities to populate lists of chats without user interaction. Aren't server extensions pre-populating rosters of newly registered users pretty popular? My client shows them all, joined or not, in the chat roster. What human-readable name do you display in the UI of your client for not-joined rooms? So to handle that, I'd now need to bookmark bookmarks...
There was a misunderstanding we sorted out OOB. This protocol would let you display a friendlier name than the JID localpart of not-disco-queried(-yet?) rooms in your UI, not propose a mechanism to discover "new" rooms.
* Your client gets told, somehow, about the place to disco for more rooms.
You can also disco#items the MUC/gateway component and you do get named rooms in there, this is true. Maybe this is cleaner than using bookmarks, but it lacks a "push" mechanism. As you said, it's a matter of tradeoffs. It is likely more work for clients to nicely integrate the disco#items replies of MUC-services-that-are-gateways than to process data from a bookmark extension.
* Your client gets told, somehow, about the place to issue a XEP-055 search for more rooms. * Your server gets told that, for you, the XEP-0055 room search should also search these other places. I think room search serves a different purpose. For instance, in the case of a matrix gateway, it should map to the Matrix global room search (if there is such a thing). Here we are talking about a much more limited set of rooms, that clients "should know about" without sending additional IQs, and can do cool things with, eg fuzzy search. I don't propose this as a general way to discover new rooms. I think if the data does not fit, that suggests it's the wrong data.
Can you elaborate on why it does not fit? It's a list of chats, and sometimes the only info we have about them is a not-human-readable JID unless the client does a disco#info. Maybe clients should do these disco#info queries and avoid displaying JID localparts. In practice they don't, but maybe this is what needs to be addressed? -- nicoco