Skip to content

Always enqueue pending record changes when preparing batches - #576

Merged
mbrandonw merged 10 commits into
mainfrom
better-way-to-always-enqueue
Nov 5, 2026
Merged

mbrandonw merged 10 commits into
mainfrom
better-way-to-always-enqueue

Conversation

@mbrandonw

Copy link
Copy Markdown
Member

Fixes the issue mentioned here.

We now have a few code paths that can cause "pending" record changes to be added to the metadatabase. Previously it was ok to enqueue those records just at start up, but now it is more appropriate for us to send them off anytime changes are sent to CloudKit because they could contain references that are necessary to save other records.

@mbrandonw
mbrandonw merged commit 5c1548a into main Nov 5, 2026
0 of 5 checks passed
@mbrandonw
mbrandonw deleted the better-way-to-always-enqueue branch November 5, 2026 21:57
mario-luis-dev added a commit to mario-luis-dev/sqlite-data that referenced this pull request Nov 9, 2026
Brings in what upstream merged after 1.12.1, unreleased: the upload sort
(pointfreeco#571), quotaExceeded saves parked and sent again (pointfreeco#335, pointfreeco#576), cached
records waiting for their parent (pointfreeco#567), CloudKit rate limiting (pointfreeco#569, pointfreeco#564),
the mock database's modification dates (pointfreeco#574) and the @fetch load race
(pointfreeco#536).

Conflicts, resolved:
- Upload sort: upstream's topologicallyAscending replaces the fork's own
  comparator; the fork's reference violation guard stays.
- enqueueLocallyPendingChanges takes upstream's scope, and still reads,
  queues and removes the queued notes in one write, so a note written while
  sync starts is queued exactly once. Notes written while sync is stopped
  record their database, as upstream's do.
- MockSyncEngineState takes upstream's single State, and keeps the fork's
  hook before adding changes and its rule that only the last save or
  deletion of a record stays queued.
- MockCloudDatabase takes upstream's records(for:) and keeps the fork's
  recordsLookupError and deletionLog.

The fork's own server lookups (settleDependents, the lost deletion and
leftover record checks) go through upstream's batched records(for:), so
they wait and try again when CloudKit limits the request rate.

Tests: MyDelegate is upstream's AccountChangedDelegate now; a share's
snapshot shows cloudkit.title as in upstream's; PreviewTests.delete() waits
for the preview timer's send to return before it ends (the fork's sends do
more after the server changes, and the test ended while one was still
returning in 4 of 5 full runs). Full suite: 421 pass, 6 runs in a row.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants