@aaronwinstonsmith Or, even better, "we" would define it...

Assuming anybody, besides me, is interested in such a thing, what's the best (or even just good) way to organize such an effort?

Here's an idea. What if we extend #ActivityPub C2S (C2S++?) with a minimal set of features (FEPs) to provide a reasonable (or even excellent) UX? Along with servers that already support C2S, we could write an external protocol adapter from C2S to the Mastodon client API to increase the number of users that could potential use a C2S client. The C2S API would be general, but the UIs could be domain-specific (microblogging, media sharing, long-form, etc.). Who's with me?
https://www.youtube.com/watch?v=6eX3fiQLo84

@silverpill @trwnh @raucao I don't think this is accurate or helpful. The first sentence of the AP spec: "The ActivityPub protocol is a decentralized social networking protocol based upon the ActivityStreams 2.0 data format.". Later, "ActivityPub uses ActivityStreams for its vocabulary." AS2 is referenced many times in the spec. It definitely *does* matter in an ActivityPub context.

@liquidparasyte AFAIK, the ActivityPub spec doesn't prohibit a change like that. Is this a question about specific implementations of ActivityPub?

@strypey@mastodon.nzoss.nz @naturzukunft @smallcircles @skyfaller @hugh @bob There are other servers that implement C2S support (ActivityPods, Vocata, onepage.pub, ...). Lack of servers implementing C2S is not the problem. See the many other issues described in this thread for examples of why one can't built an *interoperable* AP C2S client with features a typical user would expect.

@cascode @hugh gargron's perspectives are still accurate and gives a good (partial) list of the many underdefined aspects of C2S. Related discussion at SocialHub: https://socialhub.activitypub.rocks/t/nextgen-activitypub-social-api/4733/4

@mariusor Sure it is, dereferencing AP objects is part of AP. WebFinger is not AP per se, but it's part of the actual Fediverse protocol that is commonly referred to as ActivityPub. However, I agree it's in the early stages. The Follow processing and content federation will be more interesting. Like I said in another post, there's very little, if any, C++ AP server code out there, so any LLM "knowledge" based on existing implementations will at least require prog language/idiom translation.

I picked C++ because it's a relatively complex language that's not commonly used to implement AP servers (if any?).
I've learned that Claude 3.7 Sonnet is a much better coding agent than OpenAI 4o.

I've been playing with the experimental GitHub Copilot agent mode in vscode. I've asked it to start from an empty directory and create an #ActivityPub server in C++, directed only by prompts (no human coding). So far, it has created a C++ project and implemented WebFinger, resource persistence, resource retrieval and a decent set of unit tests. At this point, Mastodon is able to discover our actors. The AI is currently implementing Follow processing. I'm curious to learn how far I can push this.