refactor!: move Rust code generation logic from candid_parser.

[lwshang]

SDK-1589

Description

We are going to make ic-cdk-bindgen support more use cases, e.g. generating types only, generating bindings for canister implementation.

The first step is to move the Rust code generation logic from the candid_parser crate so that we can have complete control of Rust codegen.

The added code_generator.rs file was the rust/candid_parser/src/bindings/rust.rs in the candid crate.

The only difference:

diff cdk-rs/src/ic-cdk-bindgen/src/code_generator.rs candid/rust/candid_parser/src/bindings/rust.rs
0a1
> use super::analysis::{chase_actor, infer_rec};
3d3
< use candid_parser::bindings::analysis::{chase_actor, infer_rec};

How Has This Been Tested?

Example tests

Checklist:

• The title of this PR complies with Conventional Commits.
• I have edited the CHANGELOG accordingly.
• I have made corresponding changes to the documentation.

[chenyan-dfinity]

How do you make sure the bindgen stays in sync with candid_parser::bindings::rust? candid_parser is currently going through a major revamp, likely the Rust binding module will look completely different soon.

[lwshang]

How do you make sure the bindgen stays in sync with candid_parser::bindings::rust?

There is no need to keep sync. This PR is just setting a starting point for my following rewrite of ic-cdk-bindgen crate. I will soon make it very different from candid_parser::bindings::rust.

---

Currently, we keep the language bindings in candid_parser crate in the candid repo. While some of the language codegen logic was for some particular downstream projects, e.g. cdk-rs. agent-rs, agent-js, etc. This introduced a circular dependency which makes it really hard to maintain both sides.

There were a few cases illustrated why it is not a good idea to have such circular dependency:

Case 1

When we needed to change the generated Rust code in ic-cdk-bindgen. We had to go through following steps:

1. Open a PR in the candid repo which change the codegen logic.
2. Open a cdk-rs PR which updated the candid dependency (with a temporary git commit ref) to see if the change is as expected.
3. Merge the candid PR and make a new release of candid_parser.
4. Change the cdk-rs PR to use the new release of candid_parser and merge the PR.

Case 2

When there was a change in the candid_parser crate that accidentally altered the Rust codegen, a patch release was made. However, this change was blindly passed to the ic-cdk-bindgen, causing it to break. To fix the issue, we had to go through the cumbersome steps in case 1.

---

I'm envisioning a more clear separation of responsibilities which will make the projects more cohesive and reduce coupling.

• The candid repo (candid_parser, didc bind) can still keep some language codegen functionality.
  • But such codegen should be usage agnostic, hence no circular dependency.
• ic-cdk-bindgen is in charge of bindgen from Candid to Rust only for the usage inside Rust CDK.
  • If we want to change the generated Rust code, we will only need one cdk-rs PR and have proper CI check there.

[chenyan-dfinity]

resolved offline