[lexical-link] Bug Fix: Enable autolink matching when it unlinked

[levensta]

Description

Previously, it was possible to enter invalid characters in the text of an unlinked AutoLinkNode, which made the link invalid. The fix activates parsing of valid links in AutoLinkNode if they are marked as unlinked

Test plan

Before

Screen.Recording.2026-02-24.at.02.14.48.mov

After

Screen.Recording.2026-02-24.at.02.16.35.mov

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
lexical	Ready	Preview, Comment	Feb 28, 2026 7:56pm
lexical-playground	Ready	Preview, Comment	Feb 28, 2026 7:56pm

[etrepum]

I'm not sure this fixes the whole problem, e.g. if you have an unlinked AutoLinkNode and add a decorator like date or emoji it will split the AutoLinkNode instead of destroying any of them

[levensta]

I'm not sure this fixes the whole problem, e.g. if you have an unlinked AutoLinkNode and add a decorator like date or emoji it will split the AutoLinkNode instead of destroying any of them

If during transformation one of the child nodes is not a TextNode or is not simple, the link will be destroyed. I added a test with emoji

[levensta]

The tests are failing for the second time in a specific job e2e-tests / windows (22.x, chromium, rich-text, legacy-events) / e2e-test. I don't understand what the problem might be yet, maybe it's worth marking the tests as flaky or rewriting them as unit tests in LexicalAutoLinkExtension

[etrepum]

I'm not sure either without looking a lot more closely, which is hard for me since I don't have a windows computer to test with. unit tests would be fine, it's unclear if marking them as flaky would be a sufficient workaround. Really depends on what kind of timing issue it is.

[levensta]

Well, it took some time to figure out that the LEGACY_EVENTS variable disables the beforeInput event, and this has nothing to do with the version of Playwright 😅

The truth turned out to be that even in the current version of lexical, without changes from this PR, there is a problem that this test revealed, and it also occurs on MacOS

_Screen.Recording.2026-02-27.at.01.23.01.mov

https://playground.lexical.dev/?disableBeforeInput=true

[levensta]

I also see that in LEGACY_EVENTS mode, there is a delay between key type/press events, which may have affected the fact that the tests passed successfully on other operating systems
https://github.com/facebook/lexical/blob/main/packages/lexical-playground/__tests__/utils/index.mjs#L109-L114

I checked locally that the node transform handler is called once in both cases

[levensta]

As a result, the current changes in this PR do not bring anything new to this scenario. It already fails in prod in legacy_events mode, so I added a condition to skip the test #8165 (comment)

Tests directly related to the new behavior are written below.