Fix HistoryPlugin selection out of sync

[zurfyx]

Need a proper review on this one...

When the sharedHistory is a shared across parent/nested editors, you can reach a state where the HistoryEntry contains an EditorState and a undoSelection that do not point to the editor. This will always cause the editor to crash as it won't be able to find the node to select.

For #4389 in particular it seems like this can ocurr when the you previously had a selection on another editor and you then create the first change onto the new editor. The first change will carry the editor and editorEditor from the previous parent iteration and the selection will be previousSelection of the nestedEditor, regardless of what the selection of the nestedEditor is it will never be the selection of the (parent) editor

What I'm suggesting in this PR is to remove any chances that it can be out-of-sync, by relying on the selection in the same EditorState.

But I'm not sure why undoSelection was created in the first place so I might be missing something important.

Fixes #4389

[vercel]

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

Name	Status	Preview	Updated (UTC)
lexical	✅ Ready (Inspect)	Visit Preview	Apr 22, 2023 1:12am
lexical-playground	✅ Ready (Inspect)	Visit Preview	Apr 22, 2023 1:12am

[github-actions]

size-limit report 📦

Path	Size	Loading time (3g)	Running time (snapdragon)	Total time
packages/lexical/dist/Lexical.js	26.94 KB (0%)	539 ms (0%)	155 ms (-7.53% 🔽)	694 ms
packages/lexical-rich-text/dist/LexicalRichText.js	37.8 KB (0%)	757 ms (0%)	383 ms (+27.41% 🔺)	1.2 s
packages/lexical-plain-text/dist/LexicalPlainText.js	37.78 KB (0%)	756 ms (0%)	327 ms (+46.42% 🔺)	1.1 s

[fantactuka]

See #966 summary recording for a case where undo selection was added. Although it seems to work without undoSelection now, probably addressed with a better calculation of history push vs history merge cases

[zurfyx]

See #966 summary recording for a case where undo selection was added. Although it seems to work without undoSelection now, probably addressed with a better calculation of history push vs history merge cases

@fantactuka - I can repro the same behavior as the test plan without it

Mind sharing more details on why we need undoSelection please?

[fantactuka]

Mind sharing more details on why we need undoSelection please?

IIRC it was specifically added for cases when few operations coalesced including caret move, and during undo caret position supposed to be in (N-1). That's all I can remember :) I suppose it has been fixed with history merge algo adjustments for selection-change cases, so as long as it works as in test plan above I'm more than happe to remove undoSelection

[zurfyx]

Mind sharing more details on why we need undoSelection please?

IIRC it was specifically added for cases when few operations coalesced including caret move, and during undo caret position supposed to be in (N-1). That's all I can remember :) I suppose it has been fixed with history merge algo adjustments for selection-change cases, so as long as it works as in test plan above I'm more than happe to remove undoSelection

That's how it works now which behavior is identical to the previous version. While arguably there's undo nits that don't work like other browsers (line new line) the behavior is the same.

Screen.Recording.2023-04-25.at.10.14.03.AM.mov