Library generator removes all comments on Cargo.toml

[abbudao]

Hey @Cammisuli, how is it going?
I just found out that the generator for new libraries is removing all comments on Cargo.toml. This is a bit bothersome because I find it very useful when running a monorepo to explain some dependencies and group them, for instance:

[workspace.dependencies]
# log/tracing
tracing = "0.1"
tracing-subscriber = "0.3"

# datetime
chrono = { version = "0.4", features = ["serde"] }

# database
sqlx = { version = "0.7", features = [ "runtime-tokio", "postgres", "macros", "chrono" ] }

# testing
approx = "0.5"
ntest = "0.9"
axum-test= "13.1.1"

This has been extremely helpful for new Rust developers who still need to learn the ecosystem.

To reproduce, paste the above dependencies on root Cargo.toml with comments and run a generator:
yarn nx g @monodon/rust:lib my-lib

I need to dig more, but it's probably related to LongTengDao/j-toml#19 ?

[Cammisuli]

yea, its definitely related to the toml parsing library I'm using. I'm open to changing it if there's a better one that you know of.

[LongTengDao]

To remain all possible comments is really a hard work, and finally I'm coding a full support DOM-like thing to support unlimited use case, which already cost me weeks, a bit crazy fight.

Just wonder, maybe I got complicated thinking, is there any serialize tool in the world can remain full comment, for example in rust environment, even for other format like yaml or json5? Maybe I can be inspired, thank you!

[abbudao]

I am not an expert in this field but preserving comments in Rust tooling is not uncommon. One example is cargo-edit which uses toml_edit under the hood to achieve that.
AFAIK only order on dotted keys is not preserved.
Sorry, but I'm not familiar with the internals to give you some insights firsthand.

[JamesHenry]

@LongTengDao Thanks a lot, will you also allow for configuring the quotes used when stringifying? That would help reduce diffs

[JamesHenry]

For example right now when implementing a versioning tool for this library I get a large diff because of the comment lost (as already discussed) and also the quotes being different post-serialization:

(The only genuine diff here should be the two version fields)