Support RegExp set notation + properties of strings proposal

[mathiasbynens]

Let’s use this as the tracking bug for this work.

Here’s the high-level steps towards RegExp set notation + properties of strings support in regexpu-core:

• Revise feature-specific options API
  • PR by @nicolo-ribaudo: [breaking] Change the API to choose which features to compile #49
• Add support for the new RegExp flag + grammar to regjsparser
  • PR by @nicolo-ribaudo: Add unicodeSet option for the RegExp set notation proposal jviereck/regjsparser#113
• Add codegen support to regjsgen
  • PR by @nicolo-ribaudo: Add support for the RegExp set notation proposal bnjmnt4n/regjsgen#20
• Implement set notation transpilation
  • PR by @nicolo-ribaudo: Implement support for && and || in v sets #52
• Implement strings support in sets
  • Set operations can be represented as alternations (e.g. def|bc|a) after computing the resulting set.
  • PR by @nicolo-ribaudo: Support strings in sets in v mode #53
• Implement properties of strings transpilation
  • Similar to set operation results, any property of strings can be transpiled into a series of alternations.
  • We could use node-unicode-data packages e.g. unicode-14.0.0 to get the Unicode data for the supported properties.
  • PR by @nicolo-ribaudo: Add support for \p Unicode properties of strings #54
• Integrate with Babel
  • Parser support PR by @nicolo-ribaudo: Add parser support for the "regexp unicode sets" proposal babel/babel#14086
  • Transpilation support PR by @nicolo-ribaudo: Add transform support for the "regexp unicode sets" proposal babel/babel#14125

[nicolo-ribaudo]

regjsgen PR: bnjmnt4n/regjsgen#20

[nicolo-ribaudo]

Currently regexpu-core never introduces new flag in the output regexp. However, when transpiling things like /[^[a-z][f-h]]/v, it would be good to transform it to /[^a-z]/u and not to /(?:(?![a-z])[\s\S])/ (unless unicodeFlag is set to "transform").

Do you think that it would be ok to document that v always compiles regexps to u, unless the regexpu-core user specifies that u needs to be transformed? Or do we need to add a way for regexpu-core to signal the flags for the resulting pattern?

[mathiasbynens]

Transpiling to u seems fine, now that modern environments all support the u flag (and have done so for a while). 👍

[mathiasbynens]

This is shipping in Babel v7.17.0! https://babeljs.io/blog/2022/02/02/7.17.0