Question: is generating API spec from fastapi code intentional?

[uncleDecart]

Question

Is generating API spec from fastapi code intentional?

Further Information

TL;DR

Based on my experience, having a spec and then generating data structures for a fairly complex protocol is a good way to go, because

• In multi-language environment (or multi-service environment for that matter) you won't worry about API compatibility when deploying things
• You can have different implementations of client/server

Why do I think those might be applicable

• At some point of scaling you will need backend to be multiple services handling different workflows, written in different languages
• At some point you will most probably will have multiple server instances be it the ones you host, vs the ones which want to be a part of your network, but deployed by another entity and those might be written in different languages.

I'm talking way far into the future. Why on Earth would I suggest it right now?

Based on my experience, the more mature the system is, the harder is to flip that, there are a lot of unknown unknowns out there and the requirements you have right now, will most probably change in the future and programming language is one of those things :)

The use-case I had in mind (the reason I though about it in the first place)

Really nice idea with apis and using file-system as an abstraction over the whole concept. Let's say I have an idea of integrating this into a project, which would gather some data, would have some fixed set of commands to operate over the data and send it to an aggregating server. I do have couple of problems:

1. I want this to be my private server (think private segment of the Internet, SyftBox in a nutshell a toolkit which allows me to build "Federated Networks", not sure if that's the right term, but basically I mean set of compute nodes with storage on which I can execute some commands)
2. I want this client to be part of the system, not an arbitrary binary with python running there (interpreter, plus all the libraries will be a costly thing, plus wrapping it around in a docker image...)

I know that in theory I can do a Rust client with fixed set of those APIs under a MB or two. I have no problem looking at the server OpenAPI and implementing client following that, I know that with my forked server it'll continue working, but might become painful if I want to connect to a public net and/or sync my fork. I'll have to update those things manually.

That's why I wanted to ask this question, before getting into implementation myself. After all, I think that the main value of a system software is a spec or API, there could be multiple implementations, but it's about capabilities and how to use it.

I hope that I'm not way to off from the whole concept and would love to hear your thoughts on this :)

[eelcovdw]

Hey @uncleDecart,

We're still at a pretty early phase and we'll probably have many API changes that are not backwards compatible for the coming months. I think this is a great idea to add, once we are in a less experimental stage