[Breaking Change][lexical] Bug Fix: Commit updates on editor.setRootElement(null)

[etrepum]

Description

setRootElement(null) now triggers the same sort of reconciliation with a history-merge tag that setRootElement(element) triggered.

Previously, when setRootElement(null) was called with any pending updates, it would put the editor in a situation where editor.getEditorState() would return a writable EditorState. Now the invariant is preserved such that editor.getEditorState() always returns a fully reconciled and read-only EditorState, but to match previous semantics such that the current editorState would synchronously get set from the pendingEditorState, setRootElement now always causes a commit when the root element changes instead of just when it changes to a non-null element.

I think there were likely other subtle bugs due to the previous behavior, since setting _editorState directly from _pendingEditorState also won't run transforms, but it would only be possible to notice if you were inspecting this state while still "headless" since a full reconcile would be performed once a new root element is attached.

Closes #7022

Upgrade Information

This change means that a reconciliation will happen on setRootElement(null). Previously, this call would set the pending editor state to the current editor state which means that update listeners, transforms, etc. would not run in this particular scenario. Now they do.

This should not have any effect on normal code, but it is a breaking change because it can trigger test failures if you are counting the number of times that an update listener is called, or doing something similar to that.

Test plan

Before

If setRootElement(null) was called with any pending updates, it would put the editor in a situation where editor.getEditorState() would return a writable EditorState. Now the invariant is preserved such that editor.getEditorState() always returns a fully reconciled and read-only EditorState, but to match previous semantics setRootElement now always causes a commit when the root element changes.

The problems due to this invariant being violated surfaced after #6894 was released

After

Unit test coverage for this readOnly scenario.

[vercel]

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

Name	Status	Preview	Comments	Updated (UTC)
lexical	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Jan 7, 2025 7:16pm
lexical-playground	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Jan 7, 2025 7:16pm

[etrepum]

/cc @potatowagon

[github-actions]

size-limit report 📦

Path	Size
lexical - cjs	31.39 KB (0%)
lexical - esm	31.19 KB (0%)
@lexical/rich-text - cjs	40.47 KB (0%)
@lexical/rich-text - esm	33.12 KB (0%)
@lexical/plain-text - cjs	39.04 KB (0%)
@lexical/plain-text - esm	30.37 KB (0%)
@lexical/react - cjs	42.3 KB (0%)
@lexical/react - esm	34.42 KB (0%)

[potatowagon]

This should not have any effect on normal code, but it is a breaking change because it can trigger test failures if you are counting the number of times that an update listener is called, or doing something similar to that.

thanks for clarifying the effects of the change

[potatowagon]

would this be considered a minor breaking change? given it should work for most cases. edge case would be an implementation that depends on the number of times an update listener is called

[etrepum]

Yes, a very minor breaking change. If it breaks anything, the code was either counting calls or doing something that's probably wrong.