serde-yaml icon indicating copy to clipboard operation
serde-yaml copied to clipboard

Quote YAML 1.1 bools during serialization

Open NiklasRosenstein opened this issue 1 year ago • 4 comments

Closes #319

This PR handles pre-YAML 1.2 bool keywords (on, off, yes, no, etc.) during serialization to add single quotes which adds backwards compatibility when the serialized YAML is later loaded by a YAML parser for an older spec.

1.1 Reference: https://yaml.org/type/bool.html

Notable changes

  • Examples/tests often have a field called y which is now serialized with quotes as well (e.g. 'y': 0). I think this is desirable because some YAML implementations can interpret the field as a regular YAML value (and you end up with a boolean key), however I have yet to find that in the YAML spec.
Previous content

This PR modifies crate::de::parse_bool() to treat on/ON/yes/YES/off/OFF/no/NO as bool as per the YAML 1.1 spec. The same change also causes string values with these values to be quoted, which fixes #319.

I couldn't find whether serde-yaml aims to be compatible with a particular YAML spec version; the fact that it did not support on/off/etc. before suggest that it may be targeting YAML 1.2(.2?).

Serializing these words with quotes is backwards compatible, however if we start to deserialize them as bool this may be breaking to consumers of serde-yaml. I would suggest to consider either

  1. We introduce some kind of YamlVersion enum and default to YamlVersion::V1_2, which foregoes matching the on/off/etc. words
  2. We pass a flag to parse_bool() during serialization so that we serialize treating these words as bool but don't during deserialization

NiklasRosenstein avatar Feb 22 '24 10:02 NiklasRosenstein

Now I find the test_tag_resolution() test, which makes it pretty clear the crate is targeting YAML 1.2.2. I would still like to go forward with this PR, and at least ensure that the serialized result is backwards compatible with old YAML specs.

NiklasRosenstein avatar Feb 22 '24 10:02 NiklasRosenstein

Ok, so now I just add a special case for handling the pre-YAML 1.2.2 keywords in the serializer. This avoids a breaking change on the deserializer while adding backwards compatibility to the serialized output.

NiklasRosenstein avatar Feb 22 '24 11:02 NiklasRosenstein

@NiklasRosenstein I suspect you'll have to find a way (cargo feature?) to make the behavior change opt-in. (Assuming here that @dtolnay isn't super keen on rolling out a change that will potentially break a lot of ossified/broken YAML users...) David?

uschi2000 avatar Feb 26 '24 18:02 uschi2000

Also, what does this change mean for serde roundtrips? (ie, de(ser(yaml)) ?= yaml) (I'm not sure if there are any roundtrip guarantees at the moment.)

uschi2000 avatar Feb 26 '24 18:02 uschi2000