feat(fs): preserve original formatting in Rust implementation

[JamieMason]

From #197

One thing I'm missing in the Rust implementation is #241 is regressed and I can't find anything in Rust that maintains the original formatting of the parsed in JSON file

@JamieMason

I'm still curious about the format-preservation regression -- to avoid mixing two discussions in this thread, do you want to make a separate issue to track that one? Or maybe there already is one?

@tylerd-canva

The gist of it is that I can't find a library in Rust which will parse JSON, let you modify that JSON in memory, and then write it back to disk while preserving the original indentation etc of the original file. In TypeScript I used jsonc-parser but I've not been able to find an equivalent in Rust.

[tylerd-canva]

I think, worst case, that parsing and editing the JSON using tree-sitter would solve this problem, but it's not very appealing. If it was just a matter of preserving the order of keys in JSON map, that would be one thing, but preserving all formatting like indentation is a bit more complex.

[JamieMason]

While I remember I'll add this for visibility, @anthonyshew did mention to me back in November to take a look at Biome's crates as there's a good chance they have what we need – I just haven't had a chance to do that yet. The Biome people are really helpful on their Discord I hear, so they'd be good to chat to whenever I or someone else gets a chance to look into this.

Thanks 👍

[JamieMason]

Quoting the reply here from Emanuele at Biome:

Hey! Welcome to the server 🙂
Yes, Biome infra doesn't use an AST for its analysis, but uses a CST, which keeps track of all information such as spaces, comments, etc. As a result, the Rust APIs provide utilities that allow you to inspect that information (generally called trivia). As for keeping the formatting: yes, that's one to results of having such information at hand, however working with them becomes much more tedious. You have more control, and as a result you need to handle more edge cases.

To give you an example, let's say your want to replace a JSON object with a new one, and the new one needs to have the same indentation of the former one (and possibly also nested children of this new object), you need to pass that information yourself during the creation of the node.

Here's an example from our code base, coming from a lint rule that adds a new JSON object that contains another object as a child.

https://github.com/biomejs/biome/blob/7243cce87617988a09527c09e7296238e79de8ed/crates/biome_migrate/src/analyzers/no_var.rs#L154-L167

You can see that " ".repeat(4). That's the indentation of the new node
– https://discord.com/channels/1132231889290285117/1138936775105052692/1338500615566921728