that's all so 2-D ...
carbon_based
Umm... I was not so very clear perhaps. The idea would still be that user accounts as well as forums all contain their domain name, as their site of origin rather than a location identifier. Just that the host could change to any other domain (after negociation with the new host, that is). So it's not about domains being tied to specific hosts/IPs but entities being tied to domains. It would be up for design discussion if that identifier should change or not, iin the case of a migration. The idea would be to give entities the ability to roam or be resurrected from any federated copy in case they are dissatisfied with the policies of their hosts, or in the event a domain gets taken down by authoritrian actors. (That's why this actually is off-topic here)
From my glance into the ActivityPub doc, I concluded that it's really only about the data exchange protocol, yet I might have overlooked something as I never had an in-depth talk with people who implement the thing. Yet, just because many do it in a certain way does not mean to me that this is written in stone somewhere. :-)
[OT; tl/dr: the issues with forums and user accounts being under hegemony of server instances is by design but it's not actually the way one would design a truely de-centralised network]
It's a feature but not the best practice if the idea would be forums (and users) being free of domains (and the dangers of domains being taken down, and host admins' whims). The design approach of Lemmy however, speaks "hegemony" all over. It says a lot about the mindset of its creators.
An alternative would be indeed distributed directory systems, employing concepts like DHT ... well proven de-centralized resiliency for quite a while. Would it have been done in such a way, there would be no difficulty with migrating forums and users across instances, and even a domain getting lost would not necessarily lead to all forums/accounts there-on to be lost. Also the issues with link creation across instances were due to forums being bound to domain names instead of them having Universal IDs thus being agnostic of which node they are actually hosted on.
ActivityPub, AFAIK only defines a protocol for communicating datasets between instances, not the structures in which federation should be done.
It came to me today that I should play the devil's advocate. Everyone who comments seems to like this, therefore there has to be someone to take the job of un-hypeing it. So this is just a different POV ... because I can.
It's a kind of social experiment which could be fun for some. The idea is to try to make a drawing with very limited resources (a single pixel per message, perhaps a few hundred pixels over a couple of days), and concurrently with hundreds/thousands of participants.
The outcome is less an artwork than it is a work of art that shows the (dire) state of collective creativity and imaginative capabilities of people who have been "rised" in dominance-hierarchic, competition-centric, societies: imagine one of these squatted houses where the walls quickly get filled with tags and political messages. While they in theory could have collectively put on an artwork that really speaks, they'd rather endulge in "us vs. them" group think, competing for the space and overdrawing each others messages.
-- Thus, in a "canvas" action, those who can captivate the minds of the masses, and those who have the proficiency to have armies of machines fight for them, will win the most space. People will say this is fun but it's rather the same as in politics. The outcome is perhaps pretty or hilarious in some details but it's mostly logos and flags and symbolism, flat colours and two-dimensional cartoon caracters. So rather boring; rarely one will see a detail where one would say, "wow, someone really did put some thought and skill into this".
Regarding me, I could imagine taking part in a project to create a bot that would facilitate collective decision making (about division of labour among drawing project participants and on-the-fly decisions about how to interact with neighbouring bits -- stop at boundary, colour-mix, or overdraw -- and drawing of colour-dithered, possibly three-dimensional, pre-planned graphic design (or algorithmic graphics), while at the same time automating the tedious sending of draw commands from many locations (it's the most inefficient way one could do it but who cares in times of HD video streaming).
... Another fun idea that just pops into my mind, would be programming a "game of life" automaton which respects pixels that are already occupied, or overdraws them then re-draws them in their original colour. It could be made to completely vanish until closure time. :-D