Clone this repo:
  1. d757abe Merge branch 'invalid-unicode' of github.com:adetaylor/serde-jsonrc into all-prs-merged-3 by Adrian Taylor · 6 months ago master
  2. d2f7ef0 Adding new tests by Adrian Taylor · 6 months ago
  3. 4b6d847 Making unicode replacement switchable by Adrian Taylor · 6 months ago
  4. fe4e594 Handling incomplete escapes by Adrian Taylor · 6 months ago
  5. 3ce2ff1 Unicode strategy to substitute invalid chars. by Adrian Taylor · 6 months ago

serde_jsonrc

This is a lenient JSON parser forked from the serde_json crate that is that is designed to parse JSON written by humans (e.g., JSON config files). This means that it supports:

  • /* and // style comments.
  • Trailing commas for object and array literals.
  • \v and \xDD literal escapes (for vertical tab and two-digit hexadecimal characters)
  • [planned] Unquoted object keys (precise spec TBD).

I am still playing with the idea of making it more lenient, which could include taking more features from the JSON5 spec.

Motivation

When I created this crate, my immediate goal was to create a fast parser for a config file for a work project. I wanted a file format that would be familiar to developers, but restrictive in what it accepted. I had encountered this problem several times in my career, which always faced the same set of tradeoffs:

  • JSON: Not easy enough to use because of lack of support for comments and trailing commas.
  • YAML: Input language has too many features.
  • TOML: Use is not yet widespread enough that I would consider it “familiar to developers.” Also, nested objects either end up being verbose or about the same as JSON.

When considering the relative downsides of each of these options, it was clear that what I really wanted was a more lenient JSON. The next question was how to get a lenient JSON parser (in Rust, in my case). I considered the following options:

Make serde_json lenient

This was my first choice, but the maintainer wanted to keep the scope of serde_json limited to strict JSON, so we respectfully agreed that forking was the way to go.

json5 crate

The json5 crate supports the superset of JSON specified at https://json5.org/. In principle, the feature set met my needs, but in practice, I discovered the implementation was not nearly as performant as serde_json, even for small files. (Also, it cannot parse streams: only strings.)

serde-hjson crate

The serde-hjson crate provies a parser for a different superset of JSON named Hjson (“JSON for humans”). I am not a fan of Hjson because the language it accepts is not valid JavaScript, so it's not nearly intuitive as JSON.

Long-Term Goals

Ultimately, I would like to see a more lenient form of JSON standardized that experiences the same level of ubiquity as JSON today. I would like this crate to be a reference implementation of that new, more lenient specification.

I suspect that being more conservative in adding new features to the spec has the best chance of getting widespread buy-in, which is why I am not immediately gravitating towards implementing all of JSON5. Instead, I am starting with my suggested improvements to JSON from way back in 2011.

Finally, my gut feeling is that a new version of JSON should still be valid JavaScript. For example, one other shortcoming of JSON today is a lack of support for multiline strings. JSON5 addresses this by allowing \ for continued lines, but at this point, I think backtick (`) would be a more intuitive solution because that would be consistent with ES6 (though string interpolation would not be supported).

License

Because serde_jsonrc is a fork of serde_json, it maintains the original licence, which means it is licensed under either of

at your option.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in serde_jsonrc by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.