Revise Animation example

[balupton]

Reading the paper I had two questions about the animation example:

https://datatracker.ietf.org/doc/html/draft-toomim-httpbis-braid-http-02#section-5.1

https://github.com/braid-work/braid-spec/blob/bf179d8c84ea48b5cb1ea4faf23c6ea9bfc28416/draft-toomim-httpbis-braid-http-03.txt#L580-L608

1. That the paper on IETF should use PNG or JPEG as GIF has native animation built into it, so it conflates the purpose of the feature. I can see however the github version uses PNG, so good work there.

2. How come the example responds with only subsequent data (patches to ranges), and not initial data (the whole data for the latest server-side version). Shouldn't it behave like https://github.com/braid-work/braid-spec/blob/bf179d8c84ea48b5cb1ea4faf23c6ea9bfc28416/draft-toomim-httpbis-braid-http-03.txt#L406-L472 which does respond with the initial data, and the subsequent data. As doing <img src="/animated-braid.png"/> doesn't seem valuable if it is fetching changes without any initial data to base those changes on. Otherwise, the example should have a Version header in the request to indicate why initial data is not being returned.

[canadaduane]

Good point! I agree the animation example should first respond with a complete version before continuing with streaming patched versions.

[josephg]

Yes, and in general this should be something we talk about in the specification.

Different implementations may want to:

• Send the complete set of changes since the start of time (eg automerge)
• Send a current snapshot of the data with no history
• (This is already specified) Send information from some point in time

The "send current snapshot" case shows up a lot when you aren't using a merge type. (Is there a standard way to specify that a patch is just a copy of the data, and not a patch?)

[toomim]

Notes from https://braid.org/meeting-3:

• Yes, let's include the initial version first in this example
• There's also ambiguity in the different ways to respond with initial data:
  • With a snapshot
  • With sequence of versions
  • With just the versions since the last connection

[canadaduane]

Comment from Phoenix on Discord:

the streaming png example feels a bit weak since last i checked every web browser already supports a way of doing that, and has done so for around 20-30 years. multipart/x-mixed-replace is how a bunch of old JPEG webcams stream out their “video”

[toomim]

Another use-case could be to implement progressive loading of images (like how images used to load over dialup), that don't natively support it.

[toomim]

Another use-case is animating HTML from the server, rather than the client. So you could have a dynamic web app without Javascript.

[toomim]

New ideas for examples:

• Mutate a PNG via snapshots instead of patches
• Mutate a text marquee
• Mutate an HTML file over time
• Subscribe to a CPU temperature or bitcoin price
• Document editing:
  • Editing and publishing this spec itself through the IETF
  • Subscribing to a git branch and seeing each commit as a new version
• Chat
• RSS feed

Consensus:

• We want to publish draft 03 before writing new examples.
• We are split on whether to remove this example or keep it for draft 03, but nobody really seems to care a lot. We remember Ben(?) once said he thought the basic idea of this example was good, but wanted it improved.

[balupton]

A web component that takes a url as an attribute or content, and then braid subscribes to the url, embedding the (svg/txt/html) result into its element would probably work today, allowing for a demo repo to be cited. This would offer some really interesting nesting options for dynamically updating websites.

There could be a json or parser variant of it too, where you tell it which json var in the response to embed, or give it a parser function.

[balupton]

The closest example I can think of to the image while avoiding which image format complexity and logistics, would be the same example but for css stylesheet link hrefs

[toomim]

I've updated the example with an initial snapshot in 7589df1.

[toomim]

More discussion here: #94 (comment)

• Its almost impossible to figure out a binary patch to a PNG because the content is internally gzipped iirc. (if you do, you'll probably end up with a patch that just replaces the entire contents of the file)
• Its unclear whether this overwrites the specified bytes or truncates? Like, what happens if the new content is smaller than the old content? (And I might be revealing my ignorance of the content-range: bytes= header by asking this, but how does this format indicate the size of the file should be truncated?

The example might be better either:

• Using a bitmap instead of a PNG - because then the byte ranges at least make a bit of sense, and the file size will stay constant
• Or don't use patches in this example, and simply re-send the content with each change
  • And we can write a second example which uses patches