local_first_sqlite_storage changelog

0.6.1 #

A write made by somebody else is not part of the batch that is open. runInTransaction used to share its transaction with every call made while it ran, whoever made it: a record written by the person while a page of remote events was being applied joined that batch, and was rolled back with it — silently, its own future having completed — when the batch failed. A batch now owns only what its own action does (the zone it runs in says so). A call from anywhere else waits for the batch to end and commits by itself; a nested call from inside the action still joins it.
A watcher is one for as long as somebody listens. watchQuery and watchChanges registered their watcher when the stream was created and dropped it when the last listener left — for good: a stream listened to again after everybody left answered once and was never told of a write again. A watcher is now registered when its first listener arrives, every time, and one nobody listens to is asked nothing on a write.
The folder a namespace's file lives in is asked of the factory that opens it. With the default factory nothing changes; a test that names its own factory gets a file per namespace without the platform.
Runs the storage contract of local_first (watch_distribution_contract.dart) over real files and real transactions.

0.6.0 #

Requires local_first ^0.10.0, and implements what it added to the storage interface:
- getByIds(tableName, ids) reads several records in one statement, with the set in a WHERE id IN (...), keyed by id. Rows the table does not hold, and rows whose last event is a delete, are left out.
- getAllEvents(pendingOnly: true) carries the condition into the statement: what is waiting to be sent is asked for by its status, and the event history is never read back and sifted.
- LocalFirstQuery.distinctOn(field) is decided in the statement: a NOT EXISTS sub-select keeps, for each value of field, the row that no other row of its group comes before in the order asked for, ties broken by id. field and the first sort field must be declared columns of the repository's schema; anything else throws ArgumentError.
With local_first 0.10.0, a write of one record runs in one SQLite transaction — the row and its event commit together — and watchers are told once.
Added prepareNamespace(namespace), which names the namespace to open before initialize. The account's database is then the file that opens, and its schema is verified there once — not on the shared default file first and again after useNamespace. It throws StateError once the database is open; useNamespace is still what changes the namespace afterwards.
Added getConfigValues<T>(keys), which reads several config values in one statement — every cursor of a sync round at once, not one read per domain. Keys the database does not hold are left out of the answer.
The schema is verified once per database file. What was verified is remembered per namespace and across close, so switching to another account and back creates no table twice; ensureSchema still forces a new verification on every file, and a file reopened without a table it was remembered to hold — an in-memory database, a file removed — is verified again. The metadata table is declared when its database opens, not on every read of a config key.
TestHelperSqliteLocalFirstStorage exposes tableVerifications, metadataEnsured and metadataDeclarations, so a test can count the declarations that really reached the database.
The README's installation snippet names the new versions.

0.5.0 #

Implemented LocalFirstStorage.deleteAllSynced: drops every state row whose events are all synced, with those events, and returns how many rows went. A row with an event still waiting to be sent stays, with its events. Both deletes run in one transaction, joining the one in progress when there is one, and watchers are notified only when something was removed.
Implemented LocalFirstStorage.deleteWhere: drops every state row whose field equals value, and the events logged for those rows, in one transaction. A field declared in the repository's schema is matched on its own indexed column; any other field is read out of the stored payload with json_extract, which scans the table — declare the fields you drop by. A null value matches the rows that do not have the field.
Added SqliteLocalFirstStorage.deleteDatabases(databaseName), a static call that removes the database file of the default namespace and of every other namespace (<namespace>__<databaseName>), with the journal files beside them, and returns how many were removed. For an app whose data model moved on: the new model opens under a new databaseName, and the files nothing will open again leave the device. Never call it with the name of an open database. Takes an optional directory (the platform's databases folder by default) and dbFactory.
Fixed deleteEvent, which asked for id = ?: the event table has no id column, so the statement matched nothing on every database but a legacy one and the event stayed. It now asks by the event id.
A watcher reads one query at a time. Two writes side by side notified twice, the two reads could finish in either order, and the watcher was left showing the older result. A notification that arrives while a read is running asks for exactly one more read, so the last emission is the newest state.
Requires local_first ^0.9.0.
The README documents the new operations, and its installation snippet now names the real package versions.

0.4.2 #

Table/schema verification (_ensureTables) now runs once per repository per open and concurrent first touches share one run. Previously every CRUD/query call re-ran the PRAGMA table_info + CREATE INDEX pass, and N concurrent calls on a table that needed a column migration all read the PRAGMA before any ALTER TABLE ran — all but the first then failed with duplicate column name, while the others queued behind the lock ("database has been locked for 10s"). Seen in the chat app when several chats loaded their history at once after a schema that added chat_id.
The column migration tolerates a duplicate column error from a competing verification (e.g. one running inside a transaction, which cannot join the shared run without deadlocking) instead of surfacing it.
The memo is reset by close() (namespace switch) and by ensureSchema() (a re-declared schema may need new columns).

0.4.1 #

Implemented LocalFirstStorage.runInTransaction: applies a batch inside a single db.transaction, routing all CRUD/queries through the transaction executor so the whole batch commits once (a single fsync) and rolls back atomically on error. Watcher notifications are deferred and flushed once after commit instead of re-emitting on every write.
getAllEvents now accepts an optional dataId and pushes it down to a SQL WHERE, so callers that only need one record's history don't scan the whole event log.
Together these make a large cold sync dramatically faster on slower devices (profiling: applying ~98 records ~8s → ~3s), by cutting per-row fsyncs and redundant watcher re-queries.
Added an indexed getById (primary-key lookup) and watchChanges, a broadcast change signal per namespace, plus optional query profiling. Kept WAL journal mode after A/B profiling (WAL ~2121ms vs DELETE ~2437ms on the sync-apply benchmark).
Schema-driven migration: a table created by an older app version is upgraded in place (ALTER TABLE + backfill) when its schema gains columns, instead of failing on the missing column.
Reads (getById / getAll / getAllEvents / getEventById) now retry on the base connection when a concurrent runInTransaction batch commits/closes the shared transaction mid-read (transaction_closed) instead of throwing — e.g. a read triggered while a reconnect sync storm is applying a batch.

0.4.0 #

Added setPassword() to allow changing the SQLCipher encryption password at runtime.
Fixed database_closed race condition during namespace switch.
Fixed SqlCipherOpenDatabaseOptions usage and improved namespace switch protection.
Updated documentation with encryption setup and ProGuard configuration.

0.3.0 #

Enhanced SQL query construction for better performance
Fixed handling of lastEventId field in insert and upsert operations
Improved query observer implementation with proper type preservation
Fixed includeDeleted filter argument ordering in SQL queries
Updated dependency: local_first ^0.7.0

0.2.0 #

Add supported config types table and example app/run instructions to README.
Link to root contributing guide; keep docs aligned with other adapters.

0.1.0 #

Fix pubspec metadata and align workspace configuration.
Bundle example app copy configured for the SQLite adapter.

0.0.1 #

Initial release of the SQLite adapter for local_first.
Supports schema/index creation, rich filtering/sorting, metadata storage, and reactive queries.

local_first_sqlite_storage 0.6.1
local_first_sqlite_storage: ^0.6.1 copied to clipboard

Metadata

0.6.1 #

0.6.0 #

0.5.0 #

0.4.2 #

0.4.1 #

0.4.0 #

0.3.0 #

0.2.0 #

0.1.0 #

0.0.1 #

← Metadata

Documentation

Publisher

Weekly Downloads

Metadata

Topics

License

Dependencies

More

local_first_sqlite_storage 0.6.1 local_first_sqlite_storage: ^0.6.1 copied to clipboard

Metadata

0.6.1 #

0.6.0 #

0.5.0 #

0.4.2 #

0.4.1 #

0.4.0 #

0.3.0 #

0.2.0 #

0.1.0 #

0.0.1 #

← Metadata

Documentation

Publisher

Weekly Downloads

Metadata

Topics

License

Dependencies

More

local_first_sqlite_storage 0.6.1
local_first_sqlite_storage: ^0.6.1 copied to clipboard