I'm still not clear about the discussion we had yesterday.

Steve Bate

@evan I understand the extension goal, but how does that work without using JSON-LD processing to expand (disambiguate) terms? In other words, how is extensibility supposed to work for developers using plain JSON?

Steve Bate

@evan I think we're conflating data models with JSON(-LD) *serialization* of those models and that's going to cause even more confusion. Without RDF, I think it's not going to be easy to discuss whatever abstract data model AP uses.

Evan Prodromou

@steve I think trying to define where digital objects *really* are is a fool's errand. When I get an inbox, it has activities in it.

julian

@[email protected] there is no extensibility with plain JSON.

Or more specifically there are no guarantees that a property you reference actually refers to what you think, when using plain JSON.

Best you can get is convergence on a property name via popularity, but even then it's a "probably".

Evan Prodromou

@steve you put an URL in your context array that says, "I support this extension."

Then, you use terms from that extension.

And, yes, I'm aware it can get a lot more complicated, but it doesn't have to.

Evan Prodromou

@steve I mean, are apples really "on" your shopping list? No, there is not a pile of fruit on top of the piece of paper. The letters "A-P-P-L-E-S" are not actual apples. They're not even the idea of apples. But saying "I have apples on my shopping list" is useful and meaningful.

Steve Bate

@evan Not necessarily. If you'd like, I can quickly write a simple AP-server that will return a JSON-serialized inbox Collection with URIs in the "orderedItems" array instead of embedded serialized copies of the object referents. It won't interoperate with typical server implementations (Vocata is one exception), although AFAIK it would be conformant with AP/AS2. But again, this is conflating serialization with whatever abstract data model underlies it.

Evan Prodromou

@steve ok. The abstract model is, "the objects are in the collection".

Steve Bate

@evan That's not going to handle the same property names in different extensions (without JSON-LD expansion). If that weren't an issue, there could just be an "extensions" property with a list of URIs and JSON-LD could be eliminated completely.

Steve Bate

@evan One can discuss containment without identifying the *real* location of objects (whatever that even means). You seem to be annoyed with this discussion, so I'm going to end my part in the thread now. However, I think it's important (and maybe I'm the only one that feels this way?) to discuss and fully understand these and other confusing aspects of AP/AS2, so don't be surprised to see related posts in the future.

Evan Prodromou

@steve I think it's a style issue. You started the thread by saying that collections do not contain AS2 objects. This is untrue, unless you twist the definition of "contain".

Steve Bate

@evan I'm open to style suggestions. However, I still think my original statements are true. I've tried to explain why, but if you have questions, let me know.

Steve Bate

@julian Sounds right. The idea that AS2 has meaningful extensibility with plain JSON seems like wishful thinking to me.

Evan Prodromou

@steve for style, I guess I'd suggest softening language or even asking a question when your statement is counterintuitive and provocative. Instead of "The AS2 container type does not contain objects" maybe "I think the AS2 container type does not contain objects" or "Does the AS2 container type contain objects?" or "What does containment mean in AS2?"

Evan Prodromou

@steve I'm also REALLY oversensitive to any statements about allowed representation of objects in AP. As you probably know better than I, a lot of AP consumers fail to robustly handle id representation or object representation of property values.

Steve Bate

@evan I understand. That's why I said repeatedly that I'm not discussing serialization. I agree that serializing Collections with embedded copies of Objects is needed for interoperability with most existing AP S2S implementations. To clarify, are you concerned about statements related to valid, allowed AP (serialized) representations that aren't the ones typically implemented? Versus invalid representations?

Evan Prodromou

@steve so, this is an interesting question. I'd say, as an interoperability standard, AP doesn't make any (well, not many...) assumptions about underlying models. The JSON-LD serialization is all you get; each implementation decides how to manage its own structures internally.

Evan Prodromou

@steve and, yes, what I worry about is when people say, "the value of the `object` property can only be in its JSON object representation" or "the members of the `items` array can only be URLs". It's not true, it's hard for people to understand, and it drives us towards fragile implementations.

Steve Bate

@evan We're back in the circular reasoning since JSON-LD *has* a well-defined data model. It's RDF and graph-based (Chapt 10. JSON-LD 1.1 spec.). The statement that AP uses "JSON-LD" but is not an RDF data model doesn't make sense to me given how JSON-LD is specified. I don't think JSON-LD provides much extensibility benefit to plain JSON apps. Maybe there should be a variant of #ActivityPub that is unambiguously plain JSON with a plain JSON extensibility design? Maybe less confusing for devs?

Steve Bate

@evan Some AP advocates/activities like @smallcircles have been proposing something similar to this for a while now (discussed on SocialHub).