shorthand - in classes should ERROR

[bennypowers]

$ cat shorthand.re
/[\w-.]/
$ tree-sitter parse shorthand.re
(pattern [0, 0] - [1, 0]
  (term [0, 0] - [0, 8]
    (pattern_character [0, 0] - [0, 1])
    (character_class [0, 1] - [0, 7]
      (class_range [0, 2] - [0, 6]
        (character_class_escape [0, 2] - [0, 4])
        (class_character [0, 5] - [0, 6])))
    (pattern_character [0, 7] - [0, 8])))

expected: error

[the-mikedavis]

Tree-sitter grammars are usually very permissive and are not meant to be used for validation.

The docs have some notes on how to structure rules:

• https://github.com/tree-sitter/tree-sitter/blob/ccd6bf554d922596ce905730d98a77af368bba5c/docs/section-3-creating-parsers.md?plain=1#L233
• https://github.com/tree-sitter/tree-sitter/blob/ccd6bf554d922596ce905730d98a77af368bba5c/docs/section-3-creating-parsers.md?plain=1#L342-L376

that suggest optimizing the grammar for easy consumption rather than full spec conformance. In this case, if you were looking to mark this as invalid, you could have a highlight query:

(class_range
  (character_class_escape) @error)

which would highlight the \w. IMO this is a better indication of the problem than what regex101 provides in this example if you interpret this as an invalid range (albeit it's arguable).

See also similar discussion: tree-sitter/tree-sitter-javascript#209

[bennypowers]

The problem isn't the \w it's the invalid -
This is an interesting case. while regex101 flags it as invalid, regexr allows it bennypowers/nvim-regexplainer@98c45db

and firefox and node both run runs it, as well.

Thanks for the background info!