Chrome/V8 tailored rules for word/sentence

[cometkim]

While I implement the word segmenter, I met a case 'unicode-segmenter'를 has different result in spec and Node.js

Chrome/V8 (6): ', unicode, -, segmenter, ', 를
Unicode spec-compliant implementation (4): ', unicode, -, segmenter'를

JSC and Spidermonkey follow the spec.

But Intl.Segmenter in Chrome/V8 behave differently:

• can't -> can't
• a'b -> a'b
• a'가 -> a, ', 가
• a'α -> a'α
• a'א -> a'א
• a'中 -> a, ', 中
• a'あ -> a, ', あ
• a'ア -> a, ', ア
• a'ก -> a'ก
• a'ア -> a, ', ア
• 가'나 -> 가, ', 나
• α'β -> α'β
• א'ב -> א'ב

That implies Chrome/V8 is applying tailoring beyond UAX #29 rather than using the ICU library as-is.

As a native Korean user, that makes perfect sense, the apostrophe joiner is a strange rule in CJK.

There may be other examples around the sentence segmentation too.

The question is, what should the unicode-segmenter follow. Should I strictly follow the spec? Or the practical one that matches the most popular environment?

[cometkim]

Filed an issue to the Chromium tracker
https://issues.chromium.org/issues/472065310