Parse content again for ActivityPub

[julianlam]

The ActivityStreams Vocabulary defines content thusly:

The content or textual representation of the Object encoded as a JSON string. By default, the value of content is HTML. The mediaType property can be used in the object to indicate a different content type.

The leads to various implementors just sending straight HTML through; the same HTML they'd use to render on their frontend UI. We also do this currently, but it leads to some issues experienced:

• Mastodon sends superfluous tags, attributes, and classes that have no meaning in NodeBB (e.g. mention, wrapping URLs in tags, etc.)
• NodeBB link previews will send the full HTML if a preview is available, but this renders poorly on Mastodon (most of the tags get stripped out)

Discourse was sending a severely stripped down variant, but has since updated their parsing to allow some basic tags. It would be wise to follow suit at least for now:

https://socialhub.activitypub.rocks/t/content-formatting-when-federating-out/4150

[julianlam]

The existing filter:parse.post hook can still be used, and some additional options passed in.

A good starting point might be

{
  type: enum [default, plaintext, activitypub.note, activitypub.article]
}

Whereby;

• if type is default or falsy, then regular post parsing is carried out (so no change for existing plugins)
• if type is activitypub.note, then a severely stripped down version is sent (minimal allowed tags, most plugins would opt out of parsing, inline images removed)
• if type is activitypub.article, then a less stripped down version is sent (inline images, minimal allowed tags, some plugins opt out of parsing)
• type.plaintext is added for future functionality but is outside the scope of this issue at this time

[julianlam]

8cf9617

[julianlam]

The above commit allows plugins to alter behaviour based on the passed-in flag, but stricter sanitization is required at the end as well.

[julianlam]

An additional sanitization step was added on the way in, so that useless classes are removed prior to being saved in database. This only affects content but not sourceContent.