Good morning Fediverse.
-
@helge it's easy for data to be valid. it's another thing entirely for it to be useful.
All data is valid is an easy statement to make. It also turns everything into useless trash.
That's the point.
-
Good morning Fediverse. It's raining.
As a developer, I care about valid data. I don't care if it's linked or not. Make sure it's valid. Tell me how to tell it's valid. Tell me how to tell it's invalid. Tell me it's ok to trash the data if I find it invalid.
Just another post about ActivityPub's failings.
helge@mymath.rocks hmm, this I think I disagree with.
I'm all about valid data, sure. It's annoying when data isn't valid, sure.
I think being accepting of less-than-perfect data is something we can all strive for because it opens the door for so much more.
HTML parsing isn't perfect, it has its lumps (some fairly big ones too), but imagine how much further behind we'd be if syntactically imperfect html led to a blank page.
-
helge@mymath.rocks hmm, this I think I disagree with.
I'm all about valid data, sure. It's annoying when data isn't valid, sure.
I think being accepting of less-than-perfect data is something we can all strive for because it opens the door for so much more.
HTML parsing isn't perfect, it has its lumps (some fairly big ones too), but imagine how much further behind we'd be if syntactically imperfect html led to a blank page.
There is plenty of discussions of the robustness principle. See here for a take that it leads to protocol decay.
In the Fediverse context, enforcing things like fep-2277 activitypub core types should be a no brainer.
The trigger for my post was something different, but there are so many issues involving the lack of formal definition of "what is valid" and thus everybody making it up as they go. That's why I say that we should use custom activities for everything with strong validation conditions.
-
There is plenty of discussions of the robustness principle. See here for a take that it leads to protocol decay.
In the Fediverse context, enforcing things like fep-2277 activitypub core types should be a no brainer.
The trigger for my post was something different, but there are so many issues involving the lack of formal definition of "what is valid" and thus everybody making it up as they go. That's why I say that we should use custom activities for everything with strong validation conditions.
@helge @julian what does fep-2277 say that you can "enforce"? reading over the fep, it seems to not have any requirements, but it's also misguided in its rationale -- it presents things as problems when they aren't problems, and concludes things that aren't conclusive. why should actor/activity/object/collection/etc be disjoint? why can't you Update a Collection? why can't you Announce an actor?
the issues are not of "validity" but of "understanding". people misuse and ignore definitions.
-
@helge @julian what does fep-2277 say that you can "enforce"? reading over the fep, it seems to not have any requirements, but it's also misguided in its rationale -- it presents things as problems when they aren't problems, and concludes things that aren't conclusive. why should actor/activity/object/collection/etc be disjoint? why can't you Update a Collection? why can't you Announce an actor?
the issues are not of "validity" but of "understanding". people misuse and ignore definitions.
@helge @julian for example, consider that activitystreams was designed to describe streams of activities, and activitypub was designed to publish activities. yet fedi discards activities and only unwraps them for their "side effects". then why use activitystreams or activitypub at all? any information model, data model, processing model, etc inherent to activity streams is being ignored.
-
@helge @julian for example, consider that activitystreams was designed to describe streams of activities, and activitypub was designed to publish activities. yet fedi discards activities and only unwraps them for their "side effects". then why use activitystreams or activitypub at all? any information model, data model, processing model, etc inherent to activity streams is being ignored.
@helge @julian most "fedi" instances would be better served by a state synchronization protocol, but activitypub is Not That. you can't replay an outbox as a chronological event log. activitypub is better fit to be a notification protocol with built-in persistence (as opposed to linked data notifications which are just direct HTTP POST requests and do not automatically persist).
-
@helge @julian for example, consider that activitystreams was designed to describe streams of activities, and activitypub was designed to publish activities. yet fedi discards activities and only unwraps them for their "side effects". then why use activitystreams or activitypub at all? any information model, data model, processing model, etc inherent to activity streams is being ignored.
trwnh@mastodon.social for the same reason we don't store HTTP requests in full in perpetuity and instead parse it's headers and body for relevant items and act accordingly.
ActivityStreams is a transport protocol.
-
trwnh@mastodon.social for the same reason we don't store HTTP requests in full in perpetuity and instead parse it's headers and body for relevant items and act accordingly.
ActivityStreams is a transport protocol.
@julian @helge no, activitystreams is a data format for *describing streams of activities*, and activitypub is a protocol for *publishing streams of activities*. if you're not working directly with activities then you're going to have a bad time because you are actively fighting the specs and the inherent design of those specs, which are built on activities as the primary unit of content.
-
@julian @helge no, activitystreams is a data format for *describing streams of activities*, and activitypub is a protocol for *publishing streams of activities*. if you're not working directly with activities then you're going to have a bad time because you are actively fighting the specs and the inherent design of those specs, which are built on activities as the primary unit of content.
-
@silverpill @julian @helge #ActivityPub today is where HTML was in the early 2000s - nobody agreed on what it should look like, and consumers had to go to heroic lengths to interoperate with all the different ways that producers interpreted the spec. (I was paid to work on one at the time. It was ... an experience.)
HTML has come a long way since then - but it took rather more than five years!
