Consider using more accurate `wordPattern`

[lierdakil]

Currently, language-haskell sets the word pattern to

language-haskell/src/extension.ts

Line 17 in d20d1c7

wordPattern: /([\w'_][\w'_\d]*)|([0-9]+\.[0-9]+([eE][+-]?[0-9]+)?|[0-9]+[eE][+-]?[0-9]+)/

Would be nice to use something that matches the syntax definition itself, f.ex.

/(?:[\p{Ll}_\p{Lu}\p{Lt}][\p{Ll}_\p{Lu}\p{Lt}\p{Nd}']*)|(?:[\p{S}\p{P}](?<![(),;\[\]`{}_"']))+/u

Feel free to omit operator regex if you feel it's extraneous, I'm on the fence about it.

Side note: numbers are apparently filtered by vscode anyway, so matching on numeric literals is kinda pointless.

[JustusAdam]

Hmm, I would be willing to try a different word pattern and I think including operators is not a bad idea. Are you saying the new pattern you propose is just the one copied from the syntax file?

[lierdakil]

copied from the syntax file?

For the most part. I've adapted it slightly, as JS regex doesn't support range boolean operations, so it's using negative lookbehind in the operator pattern instead. I've also merged data/type constructor pattern [\p{Lu}\p{Lt}][\p{Ll}_\p{Lu}\p{Lt}\p{Nd}']* and identifier/tyvar pattern [\p{Ll}_][\p{Ll}_\p{Lu}\p{Lt}\p{Nd}']* (notice the only difference is in the first range).

FWIW I've patched language-haskell locally, and it works well enough for my purposes (i.e. autocomplete). However, operator pattern picks up things like .., which can be a little annoying sometimes. Perhaps limiting operators to length 3 and up (i.e. use {3,} instead of +) could be beneficial.