M3 - Create Specification for Peer-to-Peer Extensions to ActivityPub

[maisutton]

Enable “alias URLs” within ActivityPub data to point to their p2p equivalents from the HTTP counterparts. This will enable the ecosystem to gradually adopt more decentralized publishing flows while remaining compatible with the central “instance” model of ActivityPub publishing.

• Draft extensions spec #115
• Update staticpub repo with new P2P URLs and test loading it in the reader #113
• Write new specification based on the format used for the ActivityPub spec, linking to relevant existing specs (e.g. ActivityPub, RDF aliases) and publish under docs website #114
• Write brief description of how it works for docs site
• Blog post announcing the spec
• Reach out to folks in social-wg mailing list to make them aware of the spec

[RangerMauve]

Depends on #102

[RangerMauve]

• Update staticpub repo with new P2P URLs and test loading it in the reader
• Write new specification based on the format used for the ActivityPub spec, linking to relevant existing specs (e.g. ActivityPub, RDF aliases) and publish under docs website
• Write brief description of how it works for docs site
• Blog post announcing the spec
• Reach out to folks in social-wg mailing list to make them aware of the spec

[fauno]

are FEPs a canonical way to do this?

[RangerMauve]

Yeah! We should structure the spec as a FEP

[sutty-coop]

Mauve: I changed the completion date to March 15th bc the reverse timeline said:
| -1 | Extensions spec
If that's not the estimation, let me know or let's take it back to Feb 29th (date stated as completion date here)

[RangerMauve]

So, I think sadly the best way forward would be to publish multiple copies of the ActivityPub files since there's clear way we can represent links to p2p objects.

e.g. even if the Actor had an extra property pointing to it's p2p representation, it'd need to point to the HTTP version of it's outbox items.

I think the safest bet would be to use sameAs functionality. We'd have a separate set of AP objects per p2p/http representations.

Might be good to poke folks in this thread about the subject: https://socialhub.activitypub.rocks/t/nomadic-identity-for-the-fediverse/2101/16?page=2

The identity proofs spec might also be relevant for us: https://codeberg.org/fediverse/fep/src/commit/e98aae3d2c561913d0189064f85236b6adfe871f/feps/fep-c390.md

The alsoKnownAs property seems to be what Mastodon is making use of: https://socialhub.activitypub.rocks/t/defining-alsoknownas/907

Unclear if this could be used for activities and notes.

Maybe we could figure out a spec for relative URLs in IDs to save on duplication?

[RangerMauve]

Relevant thread on non-https IDs: https://socialhub.activitypub.rocks/t/dereferencing-non-https-uris-as-id/116

[RangerMauve]

Post asking for feedback on the property name: https://mastodon.mauve.moe/@mauve/111926822827079732

[RangerMauve]

This is promising: https://codeberg.org/fediverse/fep/src/branch/main/fep/fffd/fep-fffd.md

We could store in the url field with a rel: alternate and keep the mime type. Fits well with webmention.

[RangerMauve]

Initial PR to staticpub based on fep-fffd: https://github.com/RangerMauve/staticpub.mauve.moe/pull/2/files

[fauno]

Regarding FEP fffd I checked Mastodon's code and when it finds that the url attribute of an activity/actor is an array of objects, it'll pick the first one with mimeType equal "text/html" and ignore the rest.

So if we list the HTTPS versions first, anything that comes afterwards should be compatible, even if it's the HTML version on IPFS.

[fauno]

Media attachments are expected to be downloadable via HTTP and if they fail it'll keep retrying three times. It's hard-coded that it'll only download four if there are more.

The current way to work around this would be to produce an unparsable URL, use a mime type like "image/png; ipfs" or let the Mastodon queue fail three times. Or send a patch!

Files involved:

app/workers/redownload_media_worker.rb
app/lib/activitypub/activity/create.rb
app/lib/activitypub/parser/media_attachment_parser.rb

[fauno]

Looks like others would still expect the url attribute to be a string: https://github.com/superseriousbusiness/gotosocial/blob/main/internal/api/model/status.go#L58

[fauno]

I'm not profficient enough in Elixir to check what Pleroma does

[RangerMauve]

We expect using arrays of objects in the url field can bust some impls, but we can see about sending patches for that since reusing this FEP will strengthen us longer term

[RangerMauve]

Gotta make sure we specify html links should go first

[fauno]

I'm thinking of adding it to the Jekyll plugin as sub-plugin, what do you think?

[RangerMauve]

Opened a PR, waiting for review / feedback.

https://codeberg.org/fediverse/fep/pulls/379