crdt_lf_sqlite 0.3.0

crdt_lf_sqlite: ^0.3.0 copied to clipboard

Date: 2026-09-07

compare to previous release

Changed #

• A change is named by its author and its clock, not by a string: changes has author, hlc_l and hlc_c in place of the one text column. It is the same name written apart, so nothing is stored twice — and SQL can compare it, which is what a version vector asks. The primary key (document_id, author, hlc_l, hlc_c) is already the index that comparison wants.

• The database carries a schema version, in PRAGMA user_version, and this build writes version

  2. There was none before, and CREATE TABLE IF NOT EXISTS never reshapes an existing table, so opening a 0.2.0 database rebuilds changes — from its old change_id column rather than from the stored bytes, so a change this build cannot decode still migrates. It runs in one savepoint, and every later open reads the version and returns.

• The storages implement the shared contract from the new crdt_lf_persistence, re-exported here so one import stays enough. Code written against a storage now runs on any adapter, and CRDTDocumentPersistence keeps a whole document on disk for you — see that package's README.

• CRDTSqlite is now a CRDTStorageBackend. documentIds is a UNION over the three tables, the peers one included, so a document that was created and never written to is still listed. close() is idempotent.

• deleteDocumentData is now deleteDocument, the name the contract uses. Same behavior: the changes, the snapshots and the identity, in one transaction.

• Every storage method stays synchronous — sqlite3 answers on the spot, and the return types say so. await on these calls is no longer valid, drop it.

• A transaction no longer lets another document write inside it. Savepoints are a stack on the connection, so two documents sharing one could interleave and a rollback would silently undo the other's writes. Each transaction now takes a name of its own, a synchronous body is carried through without suspending, and an asynchronous one waits for whatever is already open. A call made from inside a transaction still nests.

• runInTransactionAsync is gone. runInTransaction takes a FutureOr body and covers both.

• CRDTSqlitePeerIdStorage keeps the PeerId a document writes under, in a new peers table, so a restart no longer adds a peer to the version vector. Read it before building the document: CRDTDocument(documentId: id, peerId: database.peerIdStorageForDocument(id).loadOrCreate()). An existing database picks the table up on the next open.

• getChanges takes newerThan and upTo, both VersionVectors. The database answers it: a change outside the range is never read and never decoded.

• CRDTDocumentStorage comes from crdt_lf_persistence and is re-exported, so the import path does not change.

• isEmpty and isNotEmpty are gone from both storages. Use count.

• storageForDocument returns a CRDTSqliteDocumentStorage, whose transaction() is a real SQLite transaction: a prune lands whole or not at all. It is built on savepoints, so a batch that opens a transaction of its own nests instead of failing on a second BEGIN. Its close() does nothing — one file holds every document, so the connection stays CRDTSqlite.close()'s to release.

• Requires crdt_lf: ^4.2.0.

Date: 2026-08-16

• Requires crdt_lf: ^4.0.0 instead of >=3.0.0 <5.0.0.

Date: 2026-07-28

• Widens the crdt_lf constraint to >=3.0.0 <5.0.0. No functional changes, and no migration of existing databases.

Date: 2026-07-19

• Documentation release: refreshes the CHANGELOG and docs published on pub.flutter-io.cn. No functional changes since 0.1.0.

Date: 2026-07-18

• Documentation release: refreshes the CHANGELOG and docs published on pub.flutter-io.cn. No functional changes since 0.1.0.

Initial Release #

• Added CRDTSqlite utility class for opening a SQLite database and creating the CRDT schema
• Added CRDTSqliteChangeStorage and CRDTSqliteSnapshotStorage for persisting Change and Snapshot objects as binary blobs, scoped per document via an indexed document_id column
• Added CRDTDocumentStorage container bundling both storages for a document