Click recording://entity/path links in markdown

[nikolausWest]

What

You can now embed links to entities in markdown documents using recording://entity/path, and link to components using recording://entity/path.Color.

In order to support this, a new DataPath is introduced, with a stricter parsing of entity paths. Previously foo bar was a valid path, but now you are only allowed ASCII, numbers, underscore and dash as names (outside of quotes).

A (uncompleted) description of the SfM example as markdown to

• Test how that feels as documentation
• Try out the ability to link directly to entities and components from within the markdown

Checklist

• I have read and agree to Contributor Guide and the Code of Conduct
• I've included a screenshot or gif (if applicable)
• I have tested demo.rerun.io (if applicable)

• PR Build Summary
• Docs preview
• Examples preview
• Recent benchmark results
• Wasm size tracking

[nikolausWest]

Would have loved to be able to write

[`rr.Points3D` archetype](https://www.rerun.io/docs/reference/data_types/point3d)

here instead to make the rr.Points3D be formatted as both code and a link but not super important.

[nikolausWest]

Would be super cool to also be able to refer to a specific time point but I'd need input on what that uri should look like

[nikolausWest]

Something tells me it would be nicer to not add randomness to these but probably no big deal

[emilk]

Someone else should review this, since I wrote most of the code in it 😬

[Wumpf]

oof, that was a lot of new path handling to take in, but looked all sound to me :)

Given that path parsing is quite central, I'm advocating for filling a few test gaps there.

now we need a Rerun markdown quine 😄

[Wumpf]

just noticed that InstancePath is the odd one out - it's in re_data_store unlike the others and has an overlapping relationship with DataPath. Might be nice to put them all next to each other in a follow-up PR and have an overview over all types of paths.

[emilk]

It is odd because it depends on InstanceKey which is a code-genned thing that lives in re_types. I guess we could move all paths to re_types though at some point soon.

[Wumpf]

Can we come up with something that makes it easy to distinguish weburls and internal urls on first sight? 🤔
Looking at this I wouldn't ever asssume that points entity is internal

Relatedly, our design makes it really hard to see what is a url and what is just bold faced :(

[emilk]

Styling links based on the URI is not really possible at this point

[emilk]

I'm a bit hesitant merging this, as it will break existing code.

I'm leaning towards implementing FroStr for EntityPath in a much more lenient way (basically same as on main) that will automatically convert /Hallå Där to /"Hallå Där"

[Wumpf]

if we want to disallow /Griaß di in the long run we could also issue a warning (_once) when converting it to /"Griaß di".

If on the other hand we make this defined behavior, we might need to take special care that we firmly establish /"Griaß di" == /Griaß di for all usages. But since all parsing goes through this that should be a given, right? 🤔
As I understand this now, quotes would be only really necessary in situations where we struggle delimiting/tokenizing the path parts, right?

[emilk]

The core problem is that foo.bar used to be a valid entity-path, but if we try to parse recording://foo.bar we cannot figure out "Is this the entity-path foo.bar, or is it the component path with entity-path=foo and component=bar?

So we either need to forbid . outside of quotes, or use different parsing for things we know are entity paths and things we don't know yet (URIs)

[emilk]

Ok, parsing entity paths is very forgiving again, but logs a warning when not written in their "proper" form.

We could consider making entity parsing forgiving of any error, thus solving #3393 by simply allowing anything as an entity path.

[emilk]

something went wrong when merging.

Rebased in #3442