Skip to content

Fix sorting of records when processing batches. - #571

Merged
mbrandonw merged 1 commit into
mainfrom
fix-batch-sorting
Nov 3, 2026
Merged

mbrandonw merged 1 commit into
mainfrom
fix-batch-sorting

Conversation

@mbrandonw

Copy link
Copy Markdown
Member

#568 brought up that there was a problem with sorting records when sending off batches, and this is an a simpler approach to that problem. Test taken from that PR.

Comment on lines -1094 to -1097
let lhsRecordType = lhs.tableName,
let lhsIndex = tablesByOrder[lhsRecordType],
let rhsRecordType = rhs.tableName,
let rhsIndex = tablesByOrder[rhsRecordType]

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For tables that have no references, these values would be nil, and we would fallback to returning true below, which ultimately can cause any value to be shifted around while sorting. We already have a helper for properly comparing tablesByOrder values, which is topologicallyAscending, and it further sorts tables with no references alphabetically after all tables with references.

@mbrandonw
mbrandonw merged commit 43de040 into main Nov 3, 2026
5 checks passed
@mbrandonw
mbrandonw deleted the fix-batch-sorting branch November 3, 2026 17:45
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.

1 participant