I have deeply mixed feelings about #ActivityPub's adoption of JSON-LD, as someone who's spent way too long dealing with it while building #Fedify.
-
I would be strongly opposed to any effort to remove JSON-LD from AS2. We use it for a lot of extensions. Every AP server uses the Security vocabulary for public keys.
@evan @kopper @hongminhee The problem is that signing json-ld is extremely hard, because effectively you have to turn to the RDF graph normalization algorithm, which has extremely expensive compute times. The lack of signatures means that when I boost peoples' posts, it takes down their instance, since effectively *every* distributed post on the network doesn't actually get accepted as-is, users dial-back to check its contents.
Which, at that point, we might as well not distribute the contents at all when we post to inboxes! We could just publish with the object of the activity being the object's id uri
-
@evan @kopper @hongminhee The problem is that signing json-ld is extremely hard, because effectively you have to turn to the RDF graph normalization algorithm, which has extremely expensive compute times. The lack of signatures means that when I boost peoples' posts, it takes down their instance, since effectively *every* distributed post on the network doesn't actually get accepted as-is, users dial-back to check its contents.
Which, at that point, we might as well not distribute the contents at all when we post to inboxes! We could just publish with the object of the activity being the object's id uri
@cwebber @hongminhee @evan admittedly, codeberg.org/fediverse/fep/src/branch/main/fep/8b32/fep-8b32.md does kind of solve this specific problem. the json canonicalization used there is much lighter than rdf canonicalization (which iceshrimp had to implement in dotNetRdf specifically for its ld signature support, so tooling availability is not really an excuse in favor of json-ld either!) -
@cwebber @hongminhee @evan admittedly, codeberg.org/fediverse/fep/src/branch/main/fep/8b32/fep-8b32.md does kind of solve this specific problem. the json canonicalization used there is much lighter than rdf canonicalization (which iceshrimp had to implement in dotNetRdf specifically for its ld signature support, so tooling availability is not really an excuse in favor of json-ld either!)
@kopper @hongminhee @evan Interesting... I guess it means you can't re-compact with a new outer context, but maybe that's fine
-
@evan @kopper @hongminhee The problem is that signing json-ld is extremely hard, because effectively you have to turn to the RDF graph normalization algorithm, which has extremely expensive compute times. The lack of signatures means that when I boost peoples' posts, it takes down their instance, since effectively *every* distributed post on the network doesn't actually get accepted as-is, users dial-back to check its contents.
Which, at that point, we might as well not distribute the contents at all when we post to inboxes! We could just publish with the object of the activity being the object's id uri
@cwebber @kopper @hongminhee I talk about this in my book. Unless the receiving user is online at the time the server receives the Announce, it's ridiculous to fetch the content immediately. Receiving servers should pause a random number of minutes and then fetch the content. It avoids the thundering herd problem.
-
@cwebber @kopper @hongminhee I talk about this in my book. Unless the receiving user is online at the time the server receives the Announce, it's ridiculous to fetch the content immediately. Receiving servers should pause a random number of minutes and then fetch the content. It avoids the thundering herd problem.
@evan @cwebber @kopper @hongminhee I think that is a better algorithm than a brain dead exponential back off. Perhaps put the two together.
-
@evan @gugurumbe it's infeasible to preload all contexts, pretty much every pleroma instance hosts their own context on their own instance for example. then there is the obvious interop problems of how to handle contexts for new extensions your software is not aware of (though pretending like they're empty might work i guess?)
@kopper It does not; if a malicious context redefines the security properties then the JSON-LD processor will understand the data differently than the unaware processor.
-
@evan @cwebber @kopper @hongminhee I think that is a better algorithm than a brain dead exponential back off. Perhaps put the two together.
@patmikemid I call it trust, then verify. Usually caching the data with a ttl of a short number of minutes is enough.
-
@evan @gugurumbe it's infeasible to preload all contexts, pretty much every pleroma instance hosts their own context on their own instance for example. then there is the obvious interop problems of how to handle contexts for new extensions your software is not aware of (though pretending like they're empty might work i guess?)
-
@evan @gugurumbe i know what caching is, thanks. in fact, my current project is building one that's tailor made for solving the activitypub thundering herd problem (codeberg.org/KittyShopper/middleap)
i've been trying to keep civil through this thread largely because i started the conversation mentioning software i (temporarily) help maintain and therefore represent it even implicitly, but leaving that aside and letting my own personal thoughts enter the picture:
i think this passive aggressive reply is the last straw. thinking that i somehow know enough to write code for this protocol without knowing what a cache is? plugging your book in a network largely developed by poor minorities (i myself have the rough equivalent of less than 40 USD in my bank account total)? this inability to consider change? ("as2 requires compaction", because you're the one defining the spec saying it does), the inability to consider the people and software producing and building upon the data, as opposed to the data itself? the inability to consider the consequences of your specifications and how they're being used in the real world?
i honestly do not know if this line of thought is truly capable of leading this protocol out the slump it's currently in. if you're insistent on shooting yourself in the foot, so be it, but please take the time to consider how this behavior affects other people.
i've largely been burnt out of interacting in socialhub and other official protocol communities due to exactly this behavior, whether from you or others with influence on the final specs, and the only reason i keep trying is because of what's probably a self-destructive autistic hyperfixation on this niche network and trying to make it actually work for me and my friends, as opposed to receiving funding from the well-known genocide enablers at meta and trying to shove failing standards where they don't belong.
please be a better example. if the protocol was actually desirable then sure, you may have earnt it, after all, atproto is teeming with silicon valley e/acc death cult weirdos and yet people seem to prefer it. have you wondered why? or do you prefer to dismiss anything not coming from you without thinking about it -
@cwebber @kopper @hongminhee I talk about this in my book. Unless the receiving user is online at the time the server receives the Announce, it's ridiculous to fetch the content immediately. Receiving servers should pause a random number of minutes and then fetch the content. It avoids the thundering herd problem.
@evan@cosocial.ca @cwebber@social.coop @kopper@not-brain.d.on-t.work @hongminhee@hollo.social shared inboxes are a thing, Evan
The protocol isn't just how it was initially defined. Protocols evolve and change from their ideals to fit the needs of their operation, and getting rid of individual inboxes is one of those changes.
Social media platforms are real-time- you can't just defer stuff like that. -
@evan@cosocial.ca @cwebber@social.coop @kopper@not-brain.d.on-t.work @hongminhee@hollo.social shared inboxes are a thing, Evan
The protocol isn't just how it was initially defined. Protocols evolve and change from their ideals to fit the needs of their operation, and getting rid of individual inboxes is one of those changes.
Social media platforms are real-time- you can't just defer stuff like that.@julia @cwebber @hongminhee @kopper
Hi!
Nice to meet you. I'm well aware of `sharedInbox` and helped design it.Realtime is an illusion. You can make it pretty convincing.
Your users are mostly not online. Remote users are mostly not online. Tracking the last time remote and local users were seen can help you prioritize local and remote delivery.
It's a lot better to deliver to the tiny percent of users currently online first rather than delivering to the user named `aaaaaaaaamng` first.
-
@julia @cwebber @hongminhee @kopper
Hi!
Nice to meet you. I'm well aware of `sharedInbox` and helped design it.Realtime is an illusion. You can make it pretty convincing.
Your users are mostly not online. Remote users are mostly not online. Tracking the last time remote and local users were seen can help you prioritize local and remote delivery.
It's a lot better to deliver to the tiny percent of users currently online first rather than delivering to the user named `aaaaaaaaamng` first.
@evan@cosocial.ca @cwebber@social.coop @hongminhee@hollo.social @kopper@not-brain.d.on-t.work I feel like deferring activity resolution and publishing based on online status would only serve to create more reasons for your average person to feel that the fediverse is unstable- explaining the logistics of the herd problem to someone who doesn't know what a distributed system is is kinda difficult.
-
@evan@cosocial.ca @kopper@not-brain.d.on-t.work @gugurumbe@mastouille.fr Evan, to put it bluntly, the status quo only creates a further divide between the "big certified implementations" and the small implementation independent developers can make without worrying about making the implementation vulnerable-by-default
There could totally be a body that at least attempts to standardize well-known LD prefixes while retaining compatibility with JSON-LD (like IANA and protocol schemes), but there isn't
There could be a subset of JSON-LD that prohibits common pitfalls, but there is not one. In fact, there are very few high-quality openly available libraries that can process ActivityPub objects. There is no way to declare the actual shapes of objects without heavy fuzzing.
There is no safe amount of JSON-LD in a distributed network where context URIs may fade in and out of existence. I'm saying distributed because that's essentially what happens in practice over sufficient time.
Can we for the love of all that's serializable shrink the state space of this mess? It's possible and it's actionable, without anyone left out. We don't need a Rube-Goldberg machine to share a JPEG online -
@evan @gugurumbe i know what caching is, thanks. in fact, my current project is building one that's tailor made for solving the activitypub thundering herd problem (codeberg.org/KittyShopper/middleap)
i've been trying to keep civil through this thread largely because i started the conversation mentioning software i (temporarily) help maintain and therefore represent it even implicitly, but leaving that aside and letting my own personal thoughts enter the picture:
i think this passive aggressive reply is the last straw. thinking that i somehow know enough to write code for this protocol without knowing what a cache is? plugging your book in a network largely developed by poor minorities (i myself have the rough equivalent of less than 40 USD in my bank account total)? this inability to consider change? ("as2 requires compaction", because you're the one defining the spec saying it does), the inability to consider the people and software producing and building upon the data, as opposed to the data itself? the inability to consider the consequences of your specifications and how they're being used in the real world?
i honestly do not know if this line of thought is truly capable of leading this protocol out the slump it's currently in. if you're insistent on shooting yourself in the foot, so be it, but please take the time to consider how this behavior affects other people.
i've largely been burnt out of interacting in socialhub and other official protocol communities due to exactly this behavior, whether from you or others with influence on the final specs, and the only reason i keep trying is because of what's probably a self-destructive autistic hyperfixation on this niche network and trying to make it actually work for me and my friends, as opposed to receiving funding from the well-known genocide enablers at meta and trying to shove failing standards where they don't belong.
please be a better example. if the protocol was actually desirable then sure, you may have earnt it, after all, atproto is teeming with silicon valley e/acc death cult weirdos and yet people seem to prefer it. have you wondered why? or do you prefer to dismiss anything not coming from you without thinking about it@kopper @evan @gugurumbe thank you for saying this; the relatively poor DX is one thing, but i could get around that if it weren't for the insistence by the authors and major developers here on minimizing much of the very real problems of this protocol/network and then complaining about those who look elsewhere without considering why they would want to do that
i continue to maintain existing stuff for this network, but i'm incredibly hesitant to make anything new mostly due to the reasons stated above - especially when we have other options now
-
@evan @kopper @gugurumbe what, you got a book on caching you want to plug too?
-
@evan @gugurumbe i know what caching is, thanks. in fact, my current project is building one that's tailor made for solving the activitypub thundering herd problem (codeberg.org/KittyShopper/middleap)
i've been trying to keep civil through this thread largely because i started the conversation mentioning software i (temporarily) help maintain and therefore represent it even implicitly, but leaving that aside and letting my own personal thoughts enter the picture:
i think this passive aggressive reply is the last straw. thinking that i somehow know enough to write code for this protocol without knowing what a cache is? plugging your book in a network largely developed by poor minorities (i myself have the rough equivalent of less than 40 USD in my bank account total)? this inability to consider change? ("as2 requires compaction", because you're the one defining the spec saying it does), the inability to consider the people and software producing and building upon the data, as opposed to the data itself? the inability to consider the consequences of your specifications and how they're being used in the real world?
i honestly do not know if this line of thought is truly capable of leading this protocol out the slump it's currently in. if you're insistent on shooting yourself in the foot, so be it, but please take the time to consider how this behavior affects other people.
i've largely been burnt out of interacting in socialhub and other official protocol communities due to exactly this behavior, whether from you or others with influence on the final specs, and the only reason i keep trying is because of what's probably a self-destructive autistic hyperfixation on this niche network and trying to make it actually work for me and my friends, as opposed to receiving funding from the well-known genocide enablers at meta and trying to shove failing standards where they don't belong.
please be a better example. if the protocol was actually desirable then sure, you may have earnt it, after all, atproto is teeming with silicon valley e/acc death cult weirdos and yet people seem to prefer it. have you wondered why? or do you prefer to dismiss anything not coming from you without thinking about it@kopper @gugurumbe sorry, friend, for the curt response. I'm flying today for a death in the family, and I'm having a hard time keeping a lot of conversations going. You should have heard me trying to chair a meeting as I went through airport security!
-
@kopper @gugurumbe sorry, friend, for the curt response. I'm flying today for a death in the family, and I'm having a hard time keeping a lot of conversations going. You should have heard me trying to chair a meeting as I went through airport security!
Anyway, to me, a backwards-incompatible change is absolutely the worst possible choice we could make for the Fediverse. It splits the network, possibly permanently. We have about 100 implementations of ActivityPub, and they can't all upgrade at the same time.
-
Anyway, to me, a backwards-incompatible change is absolutely the worst possible choice we could make for the Fediverse. It splits the network, possibly permanently. We have about 100 implementations of ActivityPub, and they can't all upgrade at the same time.
@kopper @gugurumbe I just don't think the downside of having to cache the results of context URL fetches outweighs that.
-
Anyway, to me, a backwards-incompatible change is absolutely the worst possible choice we could make for the Fediverse. It splits the network, possibly permanently. We have about 100 implementations of ActivityPub, and they can't all upgrade at the same time.
@evan @gugurumbe
here is a backwards incompatible change in a fep you authored: codeberg.org/fediverse/fep/src/branch/main/fep/b2b8/fep-b2b8.md#attributedto (specifically, the Link-and-name bit. mobile Firefox does not let me send highlights apparently)
the http signature draft->rfc change is backwards incompatible.
mastodon api to c2s is backwards incompatible for client developers (and, if done correctly, would require long and unwieldy migrations on servers. ask firefish.social users how those kinds of migrations end up)
whatever the replacement for as:summary as content warnings would be backwards incompatible. replacing as:name with as:description for media alt text is backwards incompatible (gotosocial did it, and we adapted)
making webfinger optional is backwards incompatible
backwards compatibility is not here yet. now is the second best time to get rid of legacy cruft -
@cwebber @kopper @hongminhee I talk about this in my book. Unless the receiving user is online at the time the server receives the Announce, it's ridiculous to fetch the content immediately. Receiving servers should pause a random number of minutes and then fetch the content. It avoids the thundering herd problem.
@evan @kopper @hongminhee But that means either:
- Users don't get to see content that has been federated to them for *minutes*
- Unless we show unverified messages, allowing for windows of impersonation attacks, in which substantial reputational damage can be done!And also:
- Whenever I boost several of @vv's posts, her server can be down *for a while*. Random delays can help reduce load but not as substantially as signature verification
- This has to be done for both the activity *and* the object
- And there's no reason to include either the activity or the object if you care about not risking impersonation attacks, because you might as well just send {"@id": "https://example.org/post/12345/"}
