tech, Mastodon, fediverse blocks, long
Mastodon currently features two ways, as a user, to limit interactions with a given user, and one way to limit interactions with a given instance.
Regarding users, they are mute and blocks.
Muting an user means you don't receive notifications from them and you don't see them in your timelines. It doesn't give that user (or their instance) any information or require any collaboration from their instance.
Blocking goes a bit further by doing a few things:
1. Forcibly remove them from your followers and auto-reject follow requests (whether or not your account is locked)
2. Hide *your* content from them
3. (it's a bit of a side effect to 2) prevent them from interacting with you at all
Every single of those things can give out to that user that you're blocking them, though, and if they are remote, every single of those requires some kind of cooperation from their end:
1. If you have non-blocked followers on their instance, you're still sending your toots (including private) to that instance, so if the instance doesn't respect the “forcibly remove them from your followers” part, you're fucked. Ultimately, there is no way around it, but the protocol design could have been less error-prone.
2. This is mildly useful, can be very easily bypassed (just opening the profile in a browser for instance), and requires full collaboration from the instance their are on. Ultimately we can't do much better with regards to that particular limitation.
3. I'd argue this is more useful than 2. as preventing people from boosting and replying from blocked actors reduces the amount of exposure your content is likely to get in those circles you want to avoid. However, no matter how you look at it, this requires collaboration as well. But it should be possible (although very involved) to not require the collaboration of the very instance that is hosting offending actors (basically, you'd ask the instance of the blocked user if they accept the interaction and ask them to give proof that they do, then attach it to the interaction, and every collaborating instance would only accept it if it comes with proof)
Now, instance blocks. They're a mess. They are called “Hide instance” and it's not obvious if they block or mute instances. What they actually do, as far as I can tell, is:
1. Forcibly remove followers from that instance (thus requiring collaboration as stated previously, even if that's less of an issue because of 2.), reject any pending follow request from that instance
2. Stop sending anything from your account directly to that instance
3. Avoid showing you content or notifications from that instance
4. [in the development version only] if the “authorized fetch mode” is enabled, do not allow that instance to fetch anything (even public statuses)
The first and last point possibly give away (the first one proactively if you have any follower from that instance, and the last one passively) that you are blocking that instance, although it's not as obvious as the `Block` activity from user blocks.
3 is not a side effect of 2. 3 is the primary purpose. it is a bug and a huge mistake to treat it as a mere "side effect". 1 is a side effect of 3, if anything.
also, your issue with 1 is not wholly a mistake of the protocol in full. the mistake is sharedInbox for private posts --private posts should *not* be delivered to people outside their audience.
> the mistake is sharedInbox for private posts --private posts should *not* be delivered to people outside their audience.
Right but even if it's not a sharedInbox it needs to exist on the shared server. If I block X@Y, and A@Y is a friend who can get private posts, no matter what happens, server Y is going to need to handle my private messages I send to A. And if Y is malicious, well.
This might be solved with some kind of end to end encryption.
but yes, encrypting inboxes with your key could be done (shifting the problem slightly, from "malicious software" to "malicious admin"). you'd have to run your own client for it to matter, and monolithic apps like mastodon effectively bundle the client with the server.
@trwnh @darius @Thib @cwebber
Maliciousness is not an actual requirement for harmful behavior. It's enough to do unwanted favors with the best intentions. Mastodon makes some compromises for performance and other compromises to support interaction. The sum of these compromises is a problem
First, you cannot send "Reject: Follow" when you mean "Block" because you're still federating block information even if you're not using the syntax described in the spec. You're also expecting the target server to retain that information, which is neither reasonable nor possible to assurance
Second, the origin has to provide explicit routing when message delivery may be affected by information that the destination server doesn't have. If you need to deliver to collections in most cases, then you need to identify the exceptions and handle them at origin
When someone on an instance is blocked by the sender then that instance can't be relied on to expand collections in its shared inbox. I'm not addressing implementation details beyond that. Perimeter security with current practices means that instances supporting vulnerable people need to block instances that allow toxic members or federate with toxic instances. That's what needs to be improved immediately
Maliciousness is a problem, but it's not *this* problem. Vulnerable people are capable of dealing with harmful situations decisively. The problem is dealing with centrists leaking information to harmful people while they waffle about the nature of the threat. If implementations deal with the leakage, theorists have a grace period to work on more comprehensive ways to address active threats
if the Follow is stored (as it should be, since Follow must be Accepted and therefore cannot be transient) then you should be able to send Undo Accept Follow to return that account to the "follow requested" state, and Reject Follow to explicitly deny the request.
>instance can't be relied on to expand collections in its shared inbox
i'd argue this is true *always*, not just with blocks.
This is a small personal instance running on a couple small ARM servers at home.