Manish Goregaokar

I think this looks pretty promising!

Personally I'm in favor of this dependency; it's a build-time only dep and it doesn't add deps other than itself. @sffc ?

Hmmm, I hadn't seen this issue before. Hashmaps are super complicated; I'm not convinced we should be maintaining one. Do we have a strong use case yet where we have...

Ah, that's quite reasonable. Are there other ways to do maps that might work? Perhaps a BTreeMap-like solution? I suspect perfect hashing is the right solution here, but also mutation...

That could work well provided we use perfect hashing which I believe is the plan.

I think a really nice property of that system is that you can unwrap such a map to the underlying ZeroMap, mutate it, and then re-wrap it. Though ZeroMap mutation...

I think it's fine to delete if codgen failed, but we can have an env var disable it perhaps. I assume duckscript has enough stuff to make this work cross...

The way I'd prefer to do this is: ```rust #[databake(path = foo)] #[databake::constructor = from_fields] pub struct Foo Foo

oops, yeah. Your proposal does decouple things completely, but at the cost of a new public type; which I consider to be pretty high. I would rather ICU4X not be...

I mean, part of the thing is that we can have _both_ these schemes, like serde does. I just don't want us to use from_fields in ICU4X. Perhaps with a...