Fixed #

• getOrCreate/updateOrCreate race (#15): the check-then-insert sequence now runs inside a transaction; if the insert loses a race against a concurrent writer (unique violation), the row is re-fetched and returned with created: false instead of surfacing the error. Nests cleanly inside caller transactions (savepoints on SQLite).

Fixed #

• count() ignores limit/offset (#10): the count aggregation no longer includes paging, so a queryset with OFFSET beyond the row count returns the filtered total instead of throwing BloomOrmNotFoundError.
• LIKE lookups escape wildcards (#12): contains/icontains/startsWith/endsWith/iexact escape %, _, and \ in user input and emit ESCAPE '\', so input matches literally instead of over-matching.
• Unfiltered bulk writes refused (#14): QuerySet.update/delete without filters throw BloomOrmInvalidQueryError unless allowUnfiltered: true is passed, so one call can no longer wipe a table by accident.
• Data-layer hardening (#15): QuerySet.limit/offset reject negative values; PostgreSQL-only regex/iregex lookups fail clearly on SQLite; each executor bounds queryLog to kMaxQueryLogEntries (1000); nested SQLite transactions use savepoints instead of issuing nested BEGIN; PostgreSQL TLS setup is documented for production.
• valuesList honors flat (#19): valuesList(db, field, flat: false) (and the values_list alias) now returns one {field: value} map per row — the same shape as .values(db, [field]) — instead of silently ignoring the parameter and always returning bare values. flat: true (the default) behavior is unchanged.

• Maintenance release for the current ORM runtime and transaction API.

Added #

• DbExecutor.transaction<R>(callback) — atomic multi-statement transactions for both SqliteDbExecutor and PostgresDbExecutor. Commits on success, rolls back and rethrows the original exception on failure. PostgresDbExecutor delegates to package:postgres's real Connection.runTx().

• Internal query-builder fixes. No public API changes.

Initial release.

• QuerySet<T> API: filter/exclude/orderBy/limit/offset/get/first/all/exists/ count/update/delete/bulkCreate/getOrCreate/updateOrCreate/values/valuesList.
• Q() filter expressions with &/|/~ composition; F() field expressions for atomic updates.
• @BloomModel/@BloomField annotations and manual ModelMeta declaration.
• PostgresDbExecutor and SqliteDbExecutor, symmetric dialect support.
• Parameterized SQL generation throughout — no string-interpolated filter values.