fix(plugin-sdk-value): only repair divergence that persists, never transient store states#2998
Merged
Merged
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
🦋 Changeset detectedLatest commit: f03fb03 The changes in this PR will be included in the next version bump. This PR includes changesets to release 1 package
Not sure what this means? Click here to learn what changesets are. Click here if you're a maintainer who wants to add another changeset to this PR |
…ansient store states
With two users typing simultaneously (even in different blocks), a
remote transaction arriving interleaved with the listener echoes of
this client's own recent edits leaves the store value transiently
wrong until the rebase corrects it moments later. The whole-value
repair fired inside that window: it copied the transient into the
editor (deleting text the user had legitimately typed), and a
follow-up repair restored the text at a drifted offset because the
user had kept typing, scrambling words ('newsroom ' becoming
'nom ewsro').
The repair now confirms divergence before acting:
- applySync becomes two-phase: it records the exact (editor, store)
state pair and only applies the repair after a confirmation window
(1s in production, sized to outlast a slow echo round trip) if that
pair is byte-for-byte unchanged. Transients self-correct within the
window and produce zero editor writes; genuine divergence is stable
and gets repaired one beat later, losing nothing.
- Repairs never run while local keystrokes are unflushed: unflushed
edits mean the store lags the editor by design, and a repair would
delete them.
Verified with the external two-client human-typing harness: the
separate-blocks duo scenario went from ~3-in-8 trials scrambled to
10/10 clean at aggressive typing rates, solo typing stays 6/6 clean,
and soft-loss rates in the discrete collision matrix are unchanged.
The regression tests pin both sides: a transiently wrong store value
must produce zero sync writes into the editor (fails on the previous
behavior in all three browsers with exactly the whipsaw's two-write
signature), and persistent divergence still gets repaired.
christianhg
force-pushed
the
fix-duo-repair-whipsaw
branch
from
July 21, 2026 18:30
0ee2351 to
f03fb03
Compare
Contributor
📦 Bundle Stats —
|
| Metric | Value | vs main (bbce83b) |
|---|---|---|
| Internal (raw) | 803.9 KB | - |
| Internal (gzip) | 154.3 KB | - |
| Bundled (raw) | 1.41 MB | - |
| Bundled (gzip) | 318.5 KB | - |
| Import time | 99ms | +1ms, +0.8% |
@portabletext/editor/behaviors
| Metric | Value | vs main (bbce83b) |
|---|---|---|
| Internal (raw) | 467 B | - |
| Internal (gzip) | 207 B | - |
| Bundled (raw) | 424 B | - |
| Bundled (gzip) | 171 B | - |
| Import time | 2ms | +0ms, +1.8% |
@portabletext/editor/plugins
| Metric | Value | vs main (bbce83b) |
|---|---|---|
| Internal (raw) | 2.7 KB | - |
| Internal (gzip) | 894 B | - |
| Bundled (raw) | 2.5 KB | - |
| Bundled (gzip) | 827 B | - |
| Import time | 7ms | -0ms, -0.2% |
@portabletext/editor/selectors
| Metric | Value | vs main (bbce83b) |
|---|---|---|
| Internal (raw) | 81.8 KB | - |
| Internal (gzip) | 15.2 KB | - |
| Bundled (raw) | 77.6 KB | - |
| Bundled (gzip) | 14.1 KB | - |
| Import time | 8ms | -0ms, -0.8% |
@portabletext/editor/traversal
| Metric | Value | vs main (bbce83b) |
|---|---|---|
| Internal (raw) | 28.1 KB | - |
| Internal (gzip) | 5.6 KB | - |
| Bundled (raw) | 28.1 KB | - |
| Bundled (gzip) | 5.5 KB | - |
| Import time | 6ms | -0ms, -1.8% |
@portabletext/editor/utils
| Metric | Value | vs main (bbce83b) |
|---|---|---|
| Internal (raw) | 29.7 KB | - |
| Internal (gzip) | 6.2 KB | - |
| Bundled (raw) | 27.4 KB | - |
| Bundled (gzip) | 5.9 KB | - |
| Import time | 6ms | +0ms, +1.0% |
🗺️ . · ./behaviors · ./plugins · ./selectors · ./traversal · ./utils · Artifacts
Details
- Import time regressions over 10% are flagged with
⚠️ - Sizes shown as raw / gzip 🗜️. Internal bytes = own code only. Total bytes = with all dependencies. Import time = Node.js cold-start median.
📦 Bundle Stats — @portabletext/markdown
Compared against main (bbce83b8)
| Metric | Value | vs main (bbce83b) |
|---|---|---|
| Internal (raw) | 53.8 KB | - |
| Internal (gzip) | 9.8 KB | - |
| Bundled (raw) | 348.9 KB | - |
| Bundled (gzip) | 96.3 KB | - |
| Import time | 40ms | +1ms, +1.4% |
🗺️ View treemap · Artifacts
Details
- Import time regressions over 10% are flagged with
⚠️ - Sizes shown as raw / gzip 🗜️. Internal bytes = own code only. Total bytes = with all dependencies. Import time = Node.js cold-start median.
christianhg
approved these changes
Jul 21, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The bug
With two users typing simultaneously, even in separate blocks (the common "two editors, two paragraphs" case, which involves no genuine conflict), one user's words could end up scrambled mid-word: "newsroom " becoming "nom ewsro", "editors" becoming "edit… ors" with another word's restore spliced in between. Letters survive; placement breaks. Found by a new two-client human-typing driver in the external harness (sanity-labs/pte-sdk-concurrent-repro,
duo-typing.mjs): roughly 3 of 8 aggressive trials scrambled on@portabletext/plugin-sdk-value@7.1.1.Root cause (from instrumented sync traces)
When a remote transaction arrives interleaved with the listener echoes of this client's own recent edits, the SDK store value is transiently wrong until the rebase corrects it moments later. The plugin's whole-value repair fired inside exactly that window:
The repair wasn't repairing; it was photographing the store mid-blink, then un-photographing it at a drifted offset because the user kept typing between the two writes.
The fix
Repairs only act on divergence that persists:
applySyncis now two-phase: it records the exact (editor, store) state pair and applies the repair only after a confirmation window (1s in production, sized to outlast a slow echo round trip) if that pair is byte-for-byte unchanged. Transients self-correct within the window and produce zero editor writes; genuine divergence is stable and gets repaired one beat later, losing nothing.Tests
Verification against real Content Lake (external harness)
Related but independent: #2933 fixes an adjacent caret-jump (remote mark toggles replacing a block's children wholesale). Tested in combination: the two compose cleanly (10/10 with both). This fix does not require it, and neither supersedes the other; notably, this plugin's
update valueescalation path is exactly the kind of producer whose replacements #2933 teachestransformPointto tolerate.