`cargo add`: Nicer TOML Files

[killercup]

• Don't mess up Cargo.toml structure too much, e.g. keep the ordering
• Don't remove comments.
• Use inline tables whenever possible, e.g. clippy = {version = "0.0.19", optional = true}

[euclio]

+1 to this. Running using cargo add on https://github.com/Hoverbear/rust-rosetta/blob/master/Cargo.toml was a minor disaster (comments removed, binaries reordered, etc.). The toml linked above might make an interesting test case.

[flying-sheep]

another annoyance:

• my single quoted strings get converted to double quoted ones (ideally cargo-add would detect the predominate string style and use it)

we should really parse this into an extensive AST and operate on that level instead of using hashmaps without stylistic and meta information.

i didn’t need that yet, but i can also imagine a complex Cargo.toml having some sections divided by blank lines that should be preserved

[killercup]

Like @rgardner mentioned in #57, there was some effort by @vosen to implement some of this in toml-rs/toml-rs#64.

It might be a good idea to use that code as a starting point for a new crate that parses toml into an AST that maintains more stylistic information but that also offers a "high level" interface that can be used like a nested hashmap to easily insert/update values.

[flying-sheep]

toml-rs was written when the rust ecosystem wasn’t as far so maybe someone would have fun just implementing a new TOML parser in nom or peresil or so.

toml-rs is fine for reading TOML, but not for editing it.

[killercup]

Good idea, @flying-sheep. There are always people asking for suggestions on what to implement to learn Rust. A TOML parser that can write nice files is probably both easy to start and enough of a challenge to spend some time with. I think I'm going to post this in a few places 😄

[flying-sheep]

yup, toml is really easy to parse. no whitespace, no ambiguities. therefore no state and no backtracking.

whatever you choose (parser combinator or classic tokenizer + AST parser), it shouldn’t be much of a problem

[vosen]

Hi guys. This discussion got a little bit fragmented. I've summed up the current state of toml-rs/toml-rs#64 on Discourse (tl;dr It's not dead). But I think this place is slightly more fitting for planning what to actually do. Since my goal is the same as yours (having TOML parsing that doesn't mangle Cargo.toml, part of PistonDevelopers/VisualRust#196), I want to invite you to join me in toml-rs/toml-rs#64 (or take over). From what I can tell this code is furthest ahead when it comes to handling TOML gracefully: ordering, whitrespace, comments are all preserved. What's missing is a public interface for querying/editing/adding/removing elements.