fix: regression of issue: ShadowHost can't be a string (issue 941)

[YunFeng0817]

Bug

This PR is to fix a bug which is a regression of #941. The original one has been fixed but imported again in #1049

Reason

Given the text node of a detached HTMLAnchorElement, getRootNode() will return the anchor element itself and its host property is a string.

Fix

The fix is like #1020. I added unit tests to cover all key codes and I hope this bug won't show up anymore.

[Juice10]

The problem here is that ShadowRoot and HTMLAnchorElement have colliding host properties. In the jsdom unit tests created this passes. But re-reading the code a couple times I can’t see how the host collision issue is bypassed.

[YunFeng0817]

The problem here is that ShadowRoot and HTMLAnchorElement have colliding host properties. In the jsdom unit tests created this passes. But re-reading the code a couple times I can’t see how the host collision issue is bypassed.

I think the problem in the left side is this line

https://github.com/rrweb-io/rrweb/pull/1092/files#diff-048da15f20b717b1ea5aa5346398dfed2fe4e24a7d38445b94cbe707640c9fc5L523
It retrieves the host property without checking the element type.

Here is the process how the error happens:
The input of inDom() is a text node in <a>. This <a> is special because it is not mounted on the Dom.

Then we try to get its rootShadowHost.

Finally the string 'github.com' is returned as the "rootShadowHost".

[bhavitsharma]

@Mark-Fenng I am a bit lost here, we do check if the getRootNode().nodeType is a Document Fragment so why did the check fail?

[YunFeng0817]

@Mark-Fenng I am a bit lost here, we do check if the getRootNode().nodeType is a Document Fragment so why did the check fail?

https://github.com/rrweb-io/rrweb/pull/1092/files#diff-048da15f20b717b1ea5aa5346398dfed2fe4e24a7d38445b94cbe707640c9fc5L523
This line doesn't check the node type.

[bhavitsharma]

Ah, I understand it better now. Thanks. Can we merge this PR then?

[YunFeng0817]

Ah, I understand it better now. Thanks. Can we merge this PR then?

We need at least one reviewer to approve this before merging it.