Okay, now I want to _query_ this database. So I build a simple traversal algorithm (it's easy, just a few lines of prolog) and…

…I get a result that is PTIME-complete to evaluate?

Uhhhhhhhh.

Well, I suppose with some good caching and some backdrops…

But everything might not even be hosted on my server and require multiple HTTP calls to retrieve, it might be quite large, it might connect to other graphs.

Well this will be fun to preload.

2/

@ireneista

Crap. The Actor is another id. I follow that and look up and that one helpfully downloads the entire public outbox for me into the actor. So now I've gotten the actor plus 5 MB of random post data that I don't want.

So now I need a validator that can pull that apart, or I need an API that is defined in such a way that I can do subselection from the objects.

But neither of those are defined in AP.

So it is both the semantic and the structural

5/5

@risottobias @jenniferplusplus

@ireneista

Another example:

I have an Actor that I want to follow updates on a Note. How do I do this? Do I contact the attributedTo actor? With what? Do I contact the object itself? Where do I send the follow request? Can I even send such a request and expect anything?

There's no documented way to do it in AP, so I end up making it up

To my understanding, this is the layer of semantic confusion that you are discussing: there's a mismatch in expectations

3/

@risottobias @jenniferplusplus

@ireneista

There's another layer, however, that is also important. That's how the data structure actually appears in practice.

So I receive a Create(Note) object, but it doesn't include the entire note, it includes just the _id_ for the note.

Or worse, it contains _multiple_ ids for multiple objects, only some of which are notes.

I need to resolve those ids before I know what types I'm dealing with.

I resolve them, pull them up, and look at the actor…

4/

@risottobias @jenniferplusplus

@ireneista

Someone else builds a bicycle radio app, designed so that groups of cyclists can sync up their music on long trips. This uses Arrive() to mean "arrived at my destination" and uses Listen(SoundFile) to track what you are listening to while driving

So my microblog app is now not going to know what to do with Listen(SoundFile) and has a different semantic meaning for arrive

This creates different expectations and interpretations between the systems

2/

@risottobias @jenniferplusplus

@ireneista

Yes and.

If I'm understanding correctly: That is all correct and describes some of it, and the problem exists on two separate levels. This ties into what is technically called reification and also gets into some of my core objections to the use of JSON-LD.

First layer:

I am building a microblogging app. I interpret "Arrive" activities to be arriving online (presence), and I have no idea what to do with Listen(SoundFile).

1/

@risottobias @jenniferplusplus

@risottobias

If you want to preserve the JSON-LD, you do it by advertising capabilities somehow, and still you'll probably want to constraint JSON-LD to a subset of what it can do in order to make sure that everything is formatted in a way that lets you validate it easily.

@jenniferplusplus @ireneista

there is a metaphor of a "timeline" or "feed" (arguably the same thing, we're still chewing on that sort of question)

and there is a data structure of, like, some specific ActivityPub-defined type that you'd know more about than we do, which needs to be the primary entity that the timeline metaphor organizes