Conversation

Notices

1. Embed this notice
   feld (feld@friedcheese.us)'s status on Monday, 23-Sep-2024 06:07:33 JST feld

   • Daniel Barlow
   @dan unless the domain is taken away from you. Or the TLD is retired. Or the registrar shuts down and the domain expires and you can't renew it because it's stuck in limbo.
   
   There are so many possibilities, and expecting people to be able to understand how to purchase domains is still too much. I've said the same thing for years and expected it to get better. It appears no progress has been made there. Too many people are still too ignorant about this.
   • Embed this notice
     feld (feld@friedcheese.us)'s status on Monday, 23-Sep-2024 07:57:42 JST feld

     in reply to

     • Daniel Barlow
     @dan com, net, and org are all ultimately controlled by the US gov via Verisign.
     
     We have already had several fediverse servers lost because someone spammed them with child porn and the server and domain was seized. This hasn't happened in a couple years AFAIK, but it did happen several times and is gonna happen again. Guaranteed.
   • Embed this notice
     Daniel Barlow (dan@brvt.telent.net)'s status on Monday, 23-Sep-2024 07:57:51 JST Daniel Barlow

     in reply to

     @feld

     unless the domain is taken away from you. Or the TLD is retired. Or the registrar shuts down and the domain expires and you can't renew it because it's stuck in limbo.

     Assuming an old-school TLD (com/net/org/iso 3166 country) that's not run by muppets, any of these things happens less often than hosting providers going under and a lot less often than fediverse hosts throwing in the towel.

     I'm not saying it's a panacea nor is it available to everyone, but it's still a shitload easier than self-hosting your fediverse instance. "It's ok to lose your fediverse presence and have to start again because it's also possible to lose your domain name" doesn't strike me as a particularly great argument given the relative probability of those events.

     And it doesn't even seem like it'd be especially hard to fix, if the software folk wanted to. A tool to export the database as plain text (JSON, whatever) and another tool to import it in the new place.

   • Embed this notice
     feld (feld@friedcheese.us)'s status on Monday, 23-Sep-2024 08:30:18 JST feld

     in reply to

     • Daniel Barlow
     @dan
     
     > And it doesn't even seem like it'd be especially hard to fix, if the software folk wanted to. A tool to export the database as plain text (JSON, whatever) and another tool to import it in the new place.
     
     Yeah it always looks easy to people who haven't touched the code.
     
     First problem is that importing gets you nowhere useful as the entire fediverse only knows about the content under the old URLs.
     
     So ok you import your post history ignoring the other technical issues and you've made a mess: you cannot trust timestamps of activities. Everyone sane orders them in the database using a Flake or Snowflake ID which is a type of time-sorted unique hash. So now you have to be able to backdate these hashes so the import doesn't kill the local timelines by flooding them with old garbage.
     
     Ok so now someone added the ability to backdate Flakes. Next problem is the posts have to be imported and data has to be processed. Any mentions of users needs to be resolved. So every user interacted with has to be imported so the links can resolve to the local copy of their profiles. Is their server dead or account deleted? Uhoh, that's gonna break processing the post. Now you've introduced an edge case nobody can really handle reliably.
     
     How about deliveries? Now you don't know who the posts were delivered to so if you do want to delete any of them in the future you're forced to spam the deletes go every known server like Mastodon does which is terrible behavior.
     
     How about media? Now you have to be able to backup all the media attachments too and restore them onto the new server. Every post with media needs to have the URLs rewritten to whatever the new server's media hosting looks like and (ideally) re-run through their media processing pipelines for security reasons or importing media will bypass your security controls.
     
     Likes, favs, boosts? All lost.
     
     Custom instance emojis? Broken too unless you move those over and the posts need to be re-processed to fix those URLs too
     
     
     I guess we just don't care enough to fix this trivial problem 🤷
   • Embed this notice
     feld (feld@friedcheese.us)'s status on Monday, 23-Sep-2024 10:48:29 JST feld

     in reply to

     • Daniel Barlow
     @dan I think it will always be impossible as long as content is web native / addressed by HTTPS URLs
     
     If you put it on a completely different ecosystem but allow people to view the content through public websites that proxy to the existing content it would work. This is essentially what Farcaster and Nostr are. But they're making users maintain private keys so their identity is nomadic and that's generally too complex for the average person
     
     Bluesky is trying to architect around this issue by still using web primitives and I predict they will fail.
   • Embed this notice
     Daniel Barlow (dan@brvt.telent.net)'s status on Monday, 23-Sep-2024 10:48:31 JST Daniel Barlow

     in reply to
     @feld perhaps you should just have said "we've made data portability in the fediverse architecturally impossible, so please don't ask" in the first place instead of trying to argue that it doesn't matter and nobody should want it because other things can break too.