Fuzzy emails ignore the TLD list since 6.0.0

[davidtaylorhq]

Since 6.0.0, bare emails are linkified regardless of TLD.

import { LinkifyIt } from "linkify-it";

new LinkifyIt().test("user@example.badtld"); // 6.1.0: true, 5.0.2: false
new LinkifyIt().test("www.example.badtld");  // false on both

.tlds([...]) makes no difference for emails.

I think the regression is in d4a73a4

Potential fix opened at #124

[puzrin]

Hi David,

That was the intended decision in the process of refactoring, when rules were decoupled to more strict and isolated parts.

1. Ordinary fuzzy links have too much collisions with things like readme.md. The whitelist of TLDs is the only way to normalize situation somehow and get the anchor.
2. Fussy emails already have anchor @ and I expected the chances of collision to be about zero.

So, the only question is - do you have collisions in the real world or not? Returning the old behaviour is not a problem, but it will limit fuzzy emails recognition to TLD-listed only (this may be not nice too).

Could you show examples of false positives in your real messages? I need to understand the context where this happens. May be, a more smart solution can be found (if not, i will return the old behaviour).

[davidtaylorhq]

Ah I see, thanks for explaining! I couldn't find the behaviour-change in the changelog, so assumed it was accidental.

The difference was picked up in our CI at Discourse. We allow community administrators to configure their own set of TLDs to be linkified, which is passed to .tlds(...), and used to apply to both regular links and emails. We had tests to verify both link & email cases.

Your reasoning about emails being much-less-likely to collide makes sense. Let's try it, and I'll come back to you if we do come across any real-world issues.

[puzrin]

Ideally, it would be to pipe big multi-lingual corpus of messages via markdown-it v14 / v15 and send me the diverged messages. Then I'll be able to investigate all at once.

That's the only thing I can't do independent because have no access to such data.