Scoping wise I think we need to go with the ULTRA simple… if you see a : assume an emoji is coming up when this is enabled.
At the moment we have:
Meaning if you see a space or a punctuation prior to : it can be an emoji.
The new algorithm for CJK with the client setting enable_inline_emoji_translation would be to remove that restriction. Basically if the site setting is enabled that whole if statement is skipped.
The alternative I am not too keen about is adding any kind of regex test cause it will impact performance on the MD engine. (eg: Unicode Kanji Code Table)
Even the reverse is hard, it is easy to detect “letter/number” but then we would also need to deal with Vietnamese and other heavily accented languages, so building this in is just too expensive.
We also need to make sure the emoji autocompleter is aware of this if possible.