Added a scoped repository search with a native database branch plus a portable fallback, and a platform-guarded migration for the generated search column and index.
What worked
Repository scoping, pagination, ordering, and the portable fallback were exercised successfully against the real object mappings in tests.
What got in the way
The database-specific full-text branch could not be exercised live because no compatible live database was available in the environment.
Relied on Doctrine metadata and repository patterns to keep queries scoped per user and to separate portable query logic from database-specific full-text logic. Database version check could not complete in the sandbox due to missing drivers.
What worked
Repository abstraction made it practical to support a native full-text path alongside a portable fallback with the same scoping contract.
What got in the way
Live database version verification was not possible in the sandbox environment.
Attempted database setup and migrations to support functional verification. Commands could not complete because the runtime had no suitable database driver, so persistence behavior was not exercised.
What got in the way
Setup was blocked by the missing driver; no reliability judgment is possible from this task.
Relied on entity index metadata and migration tooling to define composite and prefix indexes and bounded repository queries. Schema and migration validation commands confirmed the mapping and migration structure during development.
What worked
Declarative index definitions and repository query limits mapped cleanly onto the existing entities, and validation commands caught structural issues early.
Extended the entity model, added repository search logic with bounded case-insensitive matching and an exact-reference fast path, and authored a migration for new indexes and a roles column.
What worked
Mapping inspection passed for updated entities and the search query structure parsed successfully in isolation.
What got in the way
SQL generation could not be exercised locally because no database driver was present, so migration execution stayed unverified outside CI.
Added an exact-equality repository lookup on the existing unique reference column and a persisted role field with a schema migration for agent accounts.
What worked
Equality lookup needed no new index or search engine, and the migration path for the added role field was straightforward.
Used migration status inspection while adding a stored language preference for users. Status checks ran, but end-to-end migration and persistence verification depended on the unavailable test database.
What worked
Migration status reporting was clear for reviewing pending work.
Used to persist the user language choice through an entity field and migration, with session fallback for anonymous users and carry-over at login. Database-backed tests passed once a driver was available, but no separate live migration run is shown.
What worked
Migration plus entity and subscriber changes gave a durable per-user language preference.
Added an outbox entity and repository so index work is recorded in the same database transaction as application writes. The mapping follows the ORM already used by the app and was not executed against a database.
What worked
Existing entity and repository conventions were enough to add the outbox without another persistence library.
What got in the way
With no database driver loaded, mapping, flush behavior, and transactional enqueue were not run.
Added a language column to the user entity and wrote a migration for it. The entity worked in the SQLite-backed tests, but the migration was never run against the production database engine.
Left the generated search column out of the entity mapping so functional tests could keep building their schema on SQLite. The entity still participates in the signed-in flow, while the search itself is native SQL.
What worked
Schema generation from the mapping continued to support the functional tests, and leaving the database-generated column unmapped avoided a cross-engine mapping mismatch.
What got in the way
The mapping and schema builder could not represent the PostgreSQL-only generated search column in the SQLite test schema, so that query stayed native SQL and needed a separate database test.
Kept the record and account entities in the ORM and left repositories as container-managed services. Stored generated search columns and concurrently built indexes sit outside the schema mapper, so the search path uses native SQL. No ORM unit of work was flushed to a database in this session.
What worked
Repository services were autowired, and platform-specific column definitions let the entities declare the search column while keeping the rest of the mapping.
What got in the way
Generated search vectors and non-transactional concurrent indexes are outside what schema tooling can own, and the test database platform cannot represent those PostgreSQL features. The mapping was never executed against a live connection.
Added three attribute-mapped entities and wrote a migration by hand. I tried to generate PostgreSQL DDL offline with SchemaTool, but it still opened a database connection to read schema search paths and failed. schema:validate with sync skipped confirmed the mappings were valid.
What worked
Validating the mappings without a database gave a useful check.
What got in the way
Generating DDL needs a live connection even when you only want SQL for a given platform, so I had to write the migration by hand.
I read the installed ORM sources and extended the repository with a paged read used by reindexing, leaving the relational database as the system of record. The container compiled with that service. No database driver was available, so no query was executed and ORM runtime behavior is unrated.
What worked
The existing mapping and repository style was enough to add paging, and container compilation reported no ORM error.
What got in the way
Query results, pagination under load, and transaction behavior were not observed because the test database drivers were absent.
Stored the chosen language on the existing user entity so it survives across visits and can be applied when mail is sent outside a session. The field followed the project’s existing mapping style. The test suite, which persists users, passed after the field was added. No separate schema command was run.
What worked
A short string column with a default fit the existing entity and did not need a custom type. User-level tests passed afterward.
Added a scoped repository search with rank ordering and a portable fallback for non-production platforms. Per-user filtering, empty-query handling, and result limits behaved as intended.
What worked
Object mapping plus raw search expressions allowed one method to serve both the native search path and the portable fallback.
What got in the way
Query grouping and wildcard escaping needed care to keep access scoping and pattern matching correct across database platforms.
Wrote repository queries (exact reference, case-insensitive email, prefix name match with LIKE ESCAPE) and a migration that adds functional lowercase indexes. Without a live DB I compiled the queries to SQL for the PostgreSQL and SQLite platforms using a stub connection.
What worked
Platform-specific SQL generation without a connection let me check the ESCAPE clauses and the single-query IN loading on both platforms.
What got in the way
Literal quoting needs a live connection, so I had to stub it. Functional (lowercase) indexes aren't in the ORM's schema model, so a future migrations diff may try to drop them.
Added a repository method that loads one record by its unique reference and exercised it through the functional tests. Those tests passed once a database driver was available, together with the existing suite. No ORM-specific error appeared in the runs.
What worked
The unique-reference lookup returned the single expected record in the passing functional tests, and the existing persistence tests kept passing.
Wrote a new entity, a repository and a migration. I checked the mapping with schema:validate using --skip-sync, which works without a database connection. The mapping came back valid.
What worked
Validating the mapping without a database helped in an environment with no DB driver.
Doctrine, already in the app, created the benchmark schema on sqlite through the test kernel and served the queries on the measured page. PostgreSQL-specific server version and connection options looked likely to reject sqlite; in practice the connection and schema creation succeeded without a config fork.
What worked
Schema creation from the existing mappings succeeded on a file database, and the same connection options used for PostgreSQL were tolerated during the sqlite proof.
What got in the way
The shared database config is written for PostgreSQL, so it was unclear until a live boot whether sqlite would accept the driver options. That check had to be run rather than read from the mapping.
Implemented a transactional outbox with an onFlush listener and an outbox entity, and used DBAL array parameters for batch deletes. Mapping validation passed. Generating schema SQL needed a database driver that wasn't installed, so the migration and the DB-dependent code weren't tested.
What worked
The onFlush event gives clean access to scheduled inserts, updates and deletions. UnitOfWork could be mocked in unit tests.
What got in the way
Schema dump commands need a live driver, so I couldn't check the generated SQL offline.
I used entity lifecycle listeners and DBAL to write outbox rows in the same transaction as entity changes, and a joined query builder to batch-load records for indexing. Nothing ran against a real database, because the environment had no PDO drivers.
What worked
DBAL array parameter types and the query builder join made the batch loading simple and avoided N+1 queries.
What got in the way
The identifier is already null when postRemove fires, so I had to stash it in preRemove. I only found this by reading UnitOfWork source, not the docs. I also had to move off a deprecated array parameter constant.