This also seems to be describing a querying system, data mapping (e.g. for using csv data), runtime calculated data (likely used to drive the csv reading stuff), etc a number of other things that are less related to just data and more how to get that data, beyond just an RPC API described by types. E.g. the annotations and computed types could describe a fair number of sql queries in your data model, so you could probably just point it at a database with very minor "glue" code.
Whether that second part is relevant probably depends on how much you want to write in your implementation language vs in taxi's language... but it's substantially different than protobuf in any case.
But yes, you do indeed have user-defined structure (message) types in protobuf, and you can bend them quite far. There are also annotations of a sort, though support there seems worse, personally I've only very rarely been able to use them cross-language at all.
[1] yeah, there are alternate code generators, but their support and quality is "mixed" to put it mildly. They do exist though, and many have a plugin system of some kind for making your own less-painful custom types.
Taxi is intended to work with protobuf, rather than replace it.
The idea is teams work on the Taxi type system collaboratively, building a library of terms (FirstName / LastName / etc), which can be applied consistently across multiple team's schemas.
Developers using Protobuf / OpenAPI / RAML / etc., can embed those terms directly into their existing schemas, to improve documentation, and describe interoperability between multple internal systems.
(TeamA's field called "givenName" is the same data as TeamB's field called "FIRST_NAME")
As I write this, I realize our docs are missing how to use Taxi with Protobuf, which is kinda embarrassing. There's examples documented elsewhere[0], but we need to get the Taxi docs updated!
[0] https://docs.vyne.co/reference/message-formats/protobuf/#emb...
Honestly I think this makes it more interesting... though I'm curious how it handles its generics in an agnostic way, as protobuf has some, but they're quite crippled. Or do you just have to know the limitations of your encoding and not use those features, and/or build your own translation layer to handle it? E.g. you can get pretty far with a double-encoded blob of bytes.
Internally, the compiler understands the idea of a generic type A<B>, which is used for Array<T> and Stream<T>, as well as describing generic functions.
I definitely think there's more we could do here, but haven't hit the usecase yet.
Hit me up on Slack[1] if you wanna discuss ideas, as I'd love to hear more about what role generics can play.
[0] https://docs.taxilang.org/language-reference/functions/#gene... [1] https://join.slack.com/t/vyne-dev/shared_invite/zt-697laanr-...
Also, the idea that those types can be embedded across multiple schemas is how you can describe cross-system interoperability, without requiring all systems to have the same Model for "Customer" or "Account"