diplomat VS marker

The tooling for the first kind -- calling Rust from another language -- is a bit less developed, and tends to rely on code generation that doesn't necessarily produce a natural C API. cbindgen, uniffi, cxx, and Diplomat all take this course.

I don't think that accurately describes what ICU4X does. It generates code directly in the language needed using Diplomat. Which currently supports JS/TS/Wasm, C, C++, and .NET, but we plan to support more (Dart is actively being worked on, there is a lot of interest in Java). Diplomat is designed to be relatively easy to write backends for: you have to write code that converts its internal IR to bindings code, so you mostly need to understand that internal IR and figure out how to idiomatically map it to other languages.

The ICU4X team is developing diplomat - https://github.com/rust-diplomat/diplomat/

I'm not that interested in a separate tool for this but it would be really cool to have a dart plugin as a part of Diplomat. Diplomat's not fully polished yet but it's totally okay to add new language backends to it. (Eventually I plan to restructure it so that it has a cleaner plugin interface, but for now "checked into tree" is fine for backends)

Check out this, which aims to implement said stable interface!

Hey, I might be able to give some input how I deal with it in [rust-linting](https://github.com/rust-linting/rust-linting). For some context, the project needs to load several dynamic libraries and provide each of them with an abstract syntax tree. Serializing and deserializing the types for every step would most likely be too expensive. That's why I opted for a Rust <-> Rust FFI. There are two parts of this: 1. The loaded libraries needed to accept data from a driver. For this, I generate functions in the library crates which are marked as `extern "C"` and only use FFI safe types. Passing information to the loaded crates then always calls the generated functions, which intern call access a thread local struct instance in the dynamic crate. It's important that the instance implement a specific trait. For the library creation, it seems like magic. 2. Callbacks. The loaded libraries need to pass information back to the driver. For this, I use a struct with function pointers. These are also marked as `extern "C"` and need to only use FFI safe types. The definition of FFI safe, is a bit difficult. Slices, `str`, `Option<>` and most of the rusts STD types don't have a stable layout to the point, that it can change between compilations with the same compiler. Therefore, it's required that each passed type is `#[repr(C)]`. Options are wrapped in an enum, which has `#[repr(C)]`, slices and strings are dismantled into a data pointer and a length. On the receiving and they're reconstructed again. A small warning. I'm not an expert on FFI interfaces. My implementation would probably have some problems with lifetimes, if I'd use a slightly different memory model. So far, this has worked well (Besides the required boilerplate). The project is currently sadly lacking documentation, as it's still under heavy development. If you want, feel free to lock around the code base. The stable types and most of the interface is inside the `linter_api` crate.