KDL 2.0.0 support

[zkat]

KDL 2.0.0 has been released!

Along with this new spec release, I've released [email protected], which supports both KDL 1.0.0 and KDL 2.0.0, and is able to translate document text between both versions.

This means that Niri would not need to move away from knuffel to add 2.0 support: by pulling in kdl-rs and enabling the v1-fallback feature, you can basically do something like this:

let mut doc = kdl::parse(input)?;
doc.ensure_v1();

let config: NiriConfig = knuffel::parse("example.kdl", doc.to_string()).expect("This should always succeed because `ensure_v1()` should guarantee a valid v1 doc, if the initial parse succeeded");

...continue as usual.

Nothing else would have to change about Niri's code unless you want to do your own heuristics around error management/fallbacks.

kdl-rs's 2.0 parser is able to collect many syntax errors in a single pass, and so can give richer error messages than either the 1.0 parser (for now), or knuffel's parser.

Always happy to answer questions! I hope this is useful to y'all!

[YaLTeR]

Oh, it's finally out! Congrats on the release.

Along with this new spec release, I've released [email protected], which supports both KDL 1.0.0 and KDL 2.0.0, and is able to translate document text between both versions.

This means that Niri would not need to move away from knuffel to add 2.0 support: by pulling in kdl-rs and enabling the v1-fallback feature, you can basically do something like this:

This is very very cool. Guess we don't have to have a big backwards compatibility break after all?

I will absolutely try this out when I'm finished with the floating PR. I'll also need to see if I can make it so when the config is KDL v2, it doesn't accept a few deprecated features.

[zkat]

Yeah I think it makes sense for Niri to have its own fallback logic, using parse_v1() and parse_v2() and making its own decision about what to do with the result. For example, if you use the fallback as-is, and most folks have v1 configs, they will never get v1 errors if they make a v1-specific mistake. It’ll just pretend it was supposed to be v2 the entire time. Probably something to play with heuristically.

[YaLTeR]

Hmm, thanks for the tip, I'll need to play around with it.

[zkat]

update: 6.2.0 now has "smarter" heuristics about which errors to report.