This should fix it for good this time. We’ve had some inconsistency between the Ruby version and the JS version of the regexpes used but that is now unnecessary.
main ← fix/unify-word-boundary-regex
merged 03:30PM - 23 Feb 26 UTC
The CJK fix (d7a53ada16) introduced separate boundary patterns for Ruby and JS e… ngines in `match_word_regexp`. The Ruby engine used `[:word:]` (which includes digits), while the JS engine used `\P{L}` (non-Letter). Since digits are not letters, the JS pattern treated them as valid word boundaries — causing "123Test" to match as "3Test" and standalone number watched words like "123" to match inside "abc123".
Replace both engine-specific patterns with a single unified pattern using Unicode property classes (`\p{L}`, `\p{M}`, `\p{N}`, `\p{Pc}`) that work identically in Ruby and JavaScript. This treats letters, marks, numbers, and connector punctuation as word characters in boundary checks, which fixes the number-matching bug for JS consumers while preserving the existing correct behavior on the Ruby side.
Since `match_word_regexp` no longer branches on engine, remove the now-dead `engine:` parameter from all 5 method signatures that threaded it through (`match_word_regexp`, `word_to_regexp`, `regexps_for_action`, `compiled_regexps_for_action`, `serialized_regexps_for_action`) and all call sites passing `engine: :js` (serializers, pretty_text).
https://meta.discourse.org/t/396110
https://meta.discourse.org/t/396109
Follow-up to d7a53ada16 (#37844)
3 Likes