crdt_lf_persistence 0.2.0

The storage contract behind the crdt_lf persistence adapters, and the consumer that keeps a document on disk as it changes.

0.2.0

Date: 2026-10-04

compare to previous release

Added

• keepSnapshots on open, openSync and openDocument: how many snapshots stay on the disk (1 by default, null for all).
• whenStored(version) completes once the disk can bring back that version, changes or snapshot; storedVersion is what the disk holds now.
• flush(throwOnError: true) throws a failed write instead of returning.

Changed

• compact() throws when the snapshot or the prune fails to write.
• Needs Dart 3.3 and crdt_lf 5.1.0 or later.

Fixed

• open throws when the document and the storage hold concurrent snapshots, instead of dropping one side; a failed open stops following the document.
• A snapshot the document took before open is written; the next open lost it.
• A failed snapshot write no longer lets its prune delete the changes it covers, and is retried.
• Two concurrent flush() calls no longer hide a failed write from each other.
• flush() waits only for what was queued when it was called: with edits arriving faster than the writes it never returned, and neither did dispose() or compact().
• documentAt throws instead of answering a version whose history was pruned with an older state.

0.1.1

Date: 2026-09-19

Widens the crdt_lf constraint to >=4.2.0 <6.0.0, so this package resolves against crdt_lf 5.0.0 as well as 4.x. No functional changes.

0.1.0

Date: 2026-09-07

First release. The storage contract behind the crdt_lf persistence adapters — the README has the API and the examples.

Added

• CRDTChangeStorage, CRDTSnapshotStorage, CRDTDocumentStorage and CRDTPeerIdStorage: what one document has on disk. crdt_lf_hive, crdt_lf_drift and crdt_lf_sqlite implement them, so code written against the contract runs on any adapter.

• CRDTStorageBackend: the database itself — storageForDocument, peerIdStorageForDocument, documentIds, deleteDocument, close.

• CRDTDocumentPersistence: keeps a CRDTDocument on disk as it changes. Writes are batched, a failed write is retried with a growing delay (250 ms, doubling, capped at 30 s), hasUnwrittenChanges says whether anything is still waiting, and compact() snapshots and prunes on demand. openSync restores before it returns, on a backend whose reads are synchronous, so a Flutter app builds its first frame from stored state.

• On the backend, everything an app does with a stored document: openDocument(id) (identity, document and restore in one call), readDocument(id) for a preview or a read-only view, documentAt(id, version) for a history view, and copyDocumentTo(other, id) for a backup or a move to another adapter. The three read-and-copy calls are also on a single CRDTDocumentStorage.

• CRDTChangeStorage.getChanges takes newerThan and upTo, both VersionVectors; filterByVersion is the fallback for a backend that cannot ask its own query language.

• newestSnapshot: which snapshot to restore from when a crash left two. Settled by version vector, then by the smaller id, so every backend answers the same from the same bytes.

• Every storage method returns a FutureOr, so a backend that answers without touching the disk never suspends; CRDTFutureOr.chain works on either.

• CRDTDocumentStorage.close() and transaction(), both with a working default, so an adapter only fills in what its backend can do.