[lexical][lexical-rich-text][lexical-utils] Bug Fix: copy and insert keep the original node's state

[LeSingh1]

Description

Four bugs, one shape. An operation that copies or inserts a block
builds the new node by hand instead of deriving it from the node it came
from, so the original's mode, alignment, format or style is silently
dropped — or the split leaves an extra empty block behind. This is the same
defect class already seen in the insertNewAfter paths that construct a
node from scratch rather than reusing the original's state.

• ParagraphNode.extractWithChild
  (packages/lexical/src/nodes/LexicalParagraphNode.ts) did not exist, so
  ParagraphNode inherited the default of false and copying a single,
  fully selected paragraph dropped the paragraph itself from both clipboard
  payloads — its alignment, indent and style had nowhere to live and the
  paste came out left aligned. HeadingNode and QuoteNode already
  override this. ParagraphNode now does too, but only when the selection
  covers the whole paragraph and that paragraph actually carries
  block-level state, so a plain paragraph still copies as the inline-only
  content existing clipboard consumers expect. ([lexical] Bug Fix: keep paragraph alignment when copying a whole paragraph #8944, fixes Bug: Text alignment is lost on copy-paste #8101)

• ParagraphNode.collapseAtStart (same file) decided a paragraph was
  "empty" from children[0] alone. A first paragraph whose first text node
  is blank but whose later children carry the content — a leading space
  followed by a bold run — was treated as empty, so Backspace at its start
  removed the whole paragraph and lost the content; a non-text child such
  as a line break was discarded the same way. It now requires every child
  to be a text node and the paragraph's whole text content to be blank.
  Blank and empty first paragraphs still collapse exactly as before. ([lexical] Bug Fix: collapseAtStart no longer discards a first paragraph that only starts blank #8975)

• $insertNodeToNearestRoot (packages/lexical-utils/src/index.ts) splits
  the enclosing blocks at the selection focus, so a focus at the end of a
  block left an empty right-hand half — a spurious empty paragraph or list
  item after the inserted node. That empty block is only wanted when the
  insertion point is at the very end of its nearest root, where it is the
  only place left for the selection to go, so a $shouldSplit now declines
  the empty 'last' split when the new $hasContentAfter helper reports
  content still following the caret. ([lexical-utils] Bug Fix: no extra paragraph when $insertNodeToNearestRoot splits at the end of a block #8976, fixes Bug: $insertNodeToNearestRoot inserts extra paragraph node if layout node is inserted at the end of text node. #5433)

• QuoteNode.insertNewAfter (packages/lexical-rich-text/src/index.ts)
  built the continuation block by hand, named its RangeSelection argument
  _ and carried across only the direction, so Enter at the end of a
  centred quote produced a left-aligned paragraph while the identical
  keystroke in a centred paragraph did not. It now carries the block format
  and style and the caret's pending text format and style, matching core
  ParagraphNode.insertNewAfter. ([lexical-rich-text] Bug Fix: Enter at the end of a quote keeps the block alignment and style #9028)

Test plan

Each fix brings its own unit tests: the new Issue8101Repro.test.ts
and ParagraphCollapseAtStart.test.ts under packages/lexical, two new
cases in LexicalUtilsInsertNodeToNearestRoot.test.tsx, and the new
QuoteInsertNewAfter.test.ts under packages/lexical-rich-text. The
quote tests run the same helper against a paragraph and against a quote and
require them to agree, so they are pinned to core's behaviour rather than to
a hardcoded expectation. #8975 additionally adds a Playwright case to
packages/lexical-playground/__tests__/e2e/Selection.spec.mjs; that spec
needs a built playground and was not run locally, so it is covered by CI's
e2e job rather than by the output below.

Supersedes #8924, #8944, #8975, #8976, #9028, consolidated per the review
feedback on #9027 and #9035.

[vercel]

@LeSingh1 is attempting to deploy a commit to the Meta Open Source Team on Vercel.

A member of the Team first needs to authorize it.

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
lexical	Ready	Preview	Aug 28, 2026 8:34pm
lexical-playground	Ready	Preview	Aug 28, 2026 8:34pm

[etrepum]

I don't think all of these are the correct approach, I don't think the intended behavior is for splitting a token to create more tokens for example. mode should only be applied to the original node and anything else created from that should be normal unless the caller explicitly wants them to keep the mode. I had made a similar comment on the original #8924

[LeSingh1]

You're right, and I should have taken the #8924 answer the first time. Dropped the splitText mode-propagation change and its test — LexicalTextNode.ts is out of the diff entirely. I don't have the call-site audit or the playground repro you asked for, so I'm not going to keep pushing on it.

What's left is the paragraph/quote/insertNodeToNearestRoot work, 8 files. 123 tests pass locally.

[etrepum]

A lint rule changed since this PR was opened so you'll need to apply pnpm run lint:fix to this (and probably your other PRs) in order to pass integrity checks