[zurfyx]

Note

Stack, part 4 of 10. Based on #9522 (zurfyx/y-keep-chars-at-format-boundary). Review only this PR's own commits.

The @lexical/y fixes touch the same writer, cursor, reader and binding internals, so they are one linear stack: each PR is based on the one below it, each conflict is resolved once, in the PR whose change causes it, and every CI run tests the real combination.

Merge order: bottom first, #9509 → #9523 → #9522 → #9505 → #9508 → #9502 → #9504 → #9527 → #9507 → #9510. With squash merges, rebase the next PR onto main before merging it (git rebase --onto main <old base branch>).

The stack

#	PR	Branch	Base
1	#9509	zurfyx/update-error-rollback	main
2	#9523	zurfyx/write-before-unread-change	#9509
3	#9522	zurfyx/y-keep-chars-at-format-boundary	#9523
4	#9505 (this PR)	zurfyx/y-undo-member-closer	#9522
5	#9508	zurfyx/y-initial-state-binding	#9505
6	#9502	zurfyx/y-reorder-intent	#9508
7	#9504	zurfyx/y-preserve-unknown-attrs	#9502
8	#9527	zurfyx/y-move-keeps-unknown-fields	#9504
9	#9507	zurfyx/y-composition-holdback	#9527
10	#9510	zurfyx/y-incremental-read-marks	#9507

Follow-ups for after the stack (not in it):

• One "seen of Yjs" view. [lexical-y] Bug Fix: Write a local update made before an unread remote change on the document it was made on #9523's Unread (changes Yjs applied that the binding has not read) and [lexical-playground][lexical-website][lexical-y] Bug Fix: Hold back remote changes to the line being composed until the composition ends #9507's Hidden (changes held back while a line is composed) answer one question, and call sites combine them by hand. One owner module (isGone, isShown, attrOf/attrsOf and the write primitives), an item-level held overlay instead of the was/now maps, and a lint or grep test that allows raw .deleted/getAttr(s) only in the Reader and in named provenance scans. [lexical-website][lexical-y] Bug Fix: Stop a format, comment or link from spreading to the end of the document after an undo or delete races an edit inside it #9505's run-opener scans (liveItemBefore, runOpener, restatedEnd) still read raw deletedness and move there too.
• B-2 at the end of a line. Enter at the very end of a line being composed, while a collaborator's alignment change is held back, leaves that alignment off the new line Lexical copies it to. Enter inside the line keeps it on both halves. It needs a per-key "inherits on split" rule, or a Limit in y.md.
• [lexical-website][lexical-y] Bug Fix: Keep the fields an older client does not know in a block it moves #9527's group fallback chain. A moved item takes its group's fields from a chain of four lookups ($groupReference, $priorGroupItem, the nearest neighbour's marker before the write, priorGroupMarker), with separate predecessor stores for markers and nodes. One stored-predecessor lookup would replace it.

Description

In @lexical/y, a format, comment or link could spread to the end of the
document when a collaborator edited inside it while it was undone, or while
another peer deleted text at its end. All of these converge on main, on
every client and in a fresh read:

• A makes "t56z t57z" italic and undoes it while B types plain text after
  "t56z": everything after B's text, the next paragraph included, turns
  italic.
• B comments "t32z t33z", where a comment of another member of the same
  set ends, and undoes it while A types after "t33z": the removed comment
  covers the rest of the document.
• B bolds "t6z t7z t8z" and undoes it while A removes bold from "t7z":
  " t8z t9z" and the next paragraph turn bold. A removal from the whole
  range, then its undo, does the same.
• A undoes bold or a comment while B pastes a list or paragraphs inside it;
  B undoes a link while A makes text across the link's start italic.
• B pastes inside a comment while C, offline for a long time, deletes text
  at the comment's end: once C is back, the comment covers words far after
  it.

The cause is shared. The client that edits inside the range restates the
value there: the write cursor restores the old value after the text it
inserts or at the end of the range it formats, and an undo brings back the
format items its change deleted. That restated run ended at the closer of
the change that opened the run. When that change is undone, its opener and
its closer are deleted; a concurrent delete can also drop the closer (it
keeps a bounded number of the format items it leaves). Either way the
restatement runs on to the end of the document.

This change adds one rule, for every stored inline kind (text formats,
styles, NodeState and its set members, links): a run that a client
restates ends with an item of that client's own.

• Cursor, format rule 5 (also used after inserted text): where a restore
  restates a value whose run another client's change opened (following
  restores back to the run's opener), and the item that ends the restored
  run is that change's own end of the run, the writer puts a null right
  after that end. If the change is undone, or a delete drops that end, the
  null ends the restored run there. It goes after the end, not before it,
  so text a collaborator types at the end of the run at the same time stays
  inside the run. A null by which a change made text inside the run plain
  is not such an end, so undoing that change still brings the value back.
• YBinding#closeRestoredRuns, the undo path: a format item an undo brings
  back gets a null of the undoer's own right after the end its original
  change gave it, where that end is gone. This generalizes the guard added
  for Bug: [lexical-y] Concurrent undo racing a collaborator's change of the same range can spread a format or comment to the end of the document #9493. The editor reads those nulls as part of the undo, with the
  selection the undo puts back, so the undoer's editor matches a fresh read.

The stored layout does not change, and nothing more is written where no
other client's change is involved. The restored value can still stay on
the part of the range after the collaborator's edit, and an undone link
stays on its own text when a collaborator's format change rewrote the
link's textFormat (both documented in y.md); neither reaches past the
range any more.

Restack (October 29)

Rebased onto #9522 with no conflicts (it was based on a main one commit older). Left for the "seen of Yjs" follow-up in the stack notes above: the run-opener scans this PR adds (liveItemBefore, runOpener, restatedEnd) read raw deletedness, so inside #9523's unread window they see the unread change's items; not reproduced as a wrong result.

Fixed after a randomized soak test of the whole stack:

• L1B-1: a comment could spread to the end of the document when a user undoes text typed right after a word two comments cover while a collaborator pastes over that text. The paste's delete tidied the run ends left after the deleted text and kept only the typist's null, which the undo then deleted. The delete now also keeps the last end of the same value another client wrote before it, so each client's undo leaves the other's end in place (L1B-1: in regressions/Formats.test.ts, 4 client-id orders, comments from one client and from two).

Changeset

.changeset/lexical-y-undo-run-end.md: [fix] for @lexical/y, with a
Verify: paragraph.

Test plan

New regressions in
packages/lexical-y/src/__tests__/unit/regressions/Formats.test.ts: each
case in both client orders, with comment variants (another member of the
set, three comments, comment and bold), the selection an undo puts back
(direct and inside an editor update), and a matrix of bold or a comment
removed (by undo or by the opposite change) while a collaborator types,
presses Enter, pastes paragraphs or a list, or deletes inside the range.

Before

On main (0cd7448):

$ npx vitest --project unit --no-watch packages/lexical-y/src/__tests__/unit/regressions/Formats.test.ts
 Test Files  1 failed (1)
      Tests  38 failed | 146 passed | 4 expected fail (188)

AssertionError: expected '"t56z t85z" " t57z t59z t60z"[I] / "t…' to be '"t56z t85z" " t57z"[I] " t59z t60z" /…' // Object.is equality
Expected: ""t56z t85z" " t57z"[I] " t59z t60z" / "t78z t79z""
Received: ""t56z t85z" " t57z t59z t60z"[I] / "t78z t79z"[I]"

AssertionError: expected '"t5z t6z t7z" " t8z t9z"[B] / "tail l…' to be '"t5z t6z t7z" " t8z"[B] " t9z" / "tai…' // Object.is equality
Expected: ""t5z t6z t7z" " t8z"[B] " t9z" / "tail line""
Received: ""t5z t6z t7z" " t8z t9z"[B] / "tail line"[B]"

AssertionError: expected 't7z t8z t9z / tail line' to be 't7z t8z t9z / ' // Object.is equality

AssertionError: expected [ ' t57z t59z t60z', 't78z t79z' ] to deeply equal []

After

$ npx vitest --project unit --no-watch packages/lexical-y/src/__tests__/unit/regressions/Formats.test.ts
 Test Files  1 passed (1)
      Tests  184 passed | 4 expected fail (188)

$ npx vitest --project unit --no-watch --maxWorkers=6 packages/lexical-y
 Test Files  148 passed | 2 skipped (150)
      Tests  4983 passed | 16 expected fail | 2 skipped (5001)

$ pnpm run ci-check
Changeset check passed: 5 pending changeset(s) valid.
Found 0 errors

Randomized collaborative sessions with formats, comments and undo (a local
soak harness, not committed), compared with main: concurrent undo sessions
112 -> 107 failing of 4000 (none new), sequential undo sessions 200 -> 199
of 4000 (none new), format sessions and the format growth probes
unchanged. An undo-vs-comment fuzz spread in 9 of 300 seeds on main and in
0 of 1000 with this change. A replay of the 4-hour session where a peer
returned after 81 minutes offline no longer spreads the comment (main: 4
untouched words get it). Long sessions store about 2.5% more format items.
Browser e2e was not run.

[vercel]

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

Project	Deployment	Actions	Updated
lexical	Ready	Preview	Oct 29, 2026 6:14pm UTC
lexical-playground	Ready	Preview	Oct 29, 2026 6:14pm UTC

[pkg-pr-new]

lexical

npm i https://pkg.pr.new/lexical@9505

@lexical/a11y

npm i https://pkg.pr.new/@lexical/a11y@9505

@lexical/clipboard

npm i https://pkg.pr.new/@lexical/clipboard@9505

@lexical/code

npm i https://pkg.pr.new/@lexical/code@9505

@lexical/code-core

npm i https://pkg.pr.new/@lexical/code-core@9505

@lexical/code-prism

npm i https://pkg.pr.new/@lexical/code-prism@9505

@lexical/code-shiki

npm i https://pkg.pr.new/@lexical/code-shiki@9505

@lexical/compiler

npm i https://pkg.pr.new/@lexical/compiler@9505

@lexical/devtools-core

npm i https://pkg.pr.new/@lexical/devtools-core@9505

@lexical/dragon

npm i https://pkg.pr.new/@lexical/dragon@9505

@lexical/eslint-plugin

npm i https://pkg.pr.new/@lexical/eslint-plugin@9505

@lexical/extension

npm i https://pkg.pr.new/@lexical/extension@9505

@lexical/fast-check

npm i https://pkg.pr.new/@lexical/fast-check@9505

@lexical/file

npm i https://pkg.pr.new/@lexical/file@9505

@lexical/hashtag

npm i https://pkg.pr.new/@lexical/hashtag@9505

@lexical/headless

npm i https://pkg.pr.new/@lexical/headless@9505

@lexical/history

npm i https://pkg.pr.new/@lexical/history@9505

@lexical/html

npm i https://pkg.pr.new/@lexical/html@9505

@lexical/internal

npm i https://pkg.pr.new/@lexical/internal@9505

@lexical/link

npm i https://pkg.pr.new/@lexical/link@9505

@lexical/list

npm i https://pkg.pr.new/@lexical/list@9505

@lexical/mark

npm i https://pkg.pr.new/@lexical/mark@9505

@lexical/markdown

npm i https://pkg.pr.new/@lexical/markdown@9505

@lexical/mdast

npm i https://pkg.pr.new/@lexical/mdast@9505

@lexical/offset

npm i https://pkg.pr.new/@lexical/offset@9505

@lexical/overflow

npm i https://pkg.pr.new/@lexical/overflow@9505

@lexical/plain-text

npm i https://pkg.pr.new/@lexical/plain-text@9505

@lexical/react

npm i https://pkg.pr.new/@lexical/react@9505

@lexical/rich-text

npm i https://pkg.pr.new/@lexical/rich-text@9505

@lexical/selection

npm i https://pkg.pr.new/@lexical/selection@9505

@lexical/table

npm i https://pkg.pr.new/@lexical/table@9505

@lexical/tailwind

npm i https://pkg.pr.new/@lexical/tailwind@9505

@lexical/text

npm i https://pkg.pr.new/@lexical/text@9505

@lexical/utils

npm i https://pkg.pr.new/@lexical/utils@9505

@lexical/y

npm i https://pkg.pr.new/@lexical/y@9505

@lexical/yjs

npm i https://pkg.pr.new/@lexical/yjs@9505

commit: 603812b