WIP: Array types and interned schema

[jamesmunns]

Still in progress, I have some string interning working, and starting to figure out the type interning.

This attempts to make a third representation for schemas, intended to make a much smaller wire form. This pulls all strings and subtypes into a deduplicated list, then describes types using references to the attached str/ty data.

I also hope to see if I can emit this to a serialized &'static [u8] at compile time so you don't need to actually serialize the schema at runtime.

CC @max-heller if you have any thoughts on the approach. I also introduced an explicit "array" type to the data model, to get away from using tuple for that.

IMO: this is a breaking "Schema" type change, but is NOT a breaking postcard wire change: fixed size arrays are encoded the same way, they are just described differently.

[netlify]

✅ Deploy Preview for cute-starship-2d9c9b canceled.

Name	Link
🔨 Latest commit	8ef9e57
🔍 Latest deploy log	https://app.netlify.com/projects/cute-starship-2d9c9b/deploys/6870e69fb8acd70008afdf0c

[jamesmunns]

So, it works, but it makes small + reasonable schemas a bit bigger (10% or so) on the wire (and a lot more complicated), it's a draw/slight win for "medium" schemas, and really only shines against silly nested/repetitive types.

Hmmm. Bummer.

[max-heller]

Can we split the array schema change out of this? That seems like a relatively straightforward addition, though I'm wondering if we want to make [T; N] and (T, T, ..., T) (N times) interchangeable or not.

Re. interning, I haven't had time to look into this implementation yet. I thought a lot about interning trees like this a while ago, maybe I can dig up some tricks