With #110258 a new Doctrine DBAL introspection method was introduced
that replaces the deprecated API calls.
This triggered a code path in Doctrine DBAL that requires an existing
database on its connection for inspection. On a fresh setup of
TYPO3 this can lead to problems.
This patch ensures to set a "dbname" to be used for the
`SetupDatabaseService` that is available.
`information_schema` always exists and is readable by any MySQL/MariaDB
user without special grants, and `getAllDatabaseNames()` ignores
the selected database anyway (it queries information_schema.SCHEMATA
unconditionally). So this only satisfies Doctrine's precondition
without changing what gets returned.
PostgreSQL/SQLite are untouched: Postgres connections always have some
database selected by the driver even when dbname is omitted (unverified),
so `PostgreSQLMetadataProvider` never had this guard, and SQLite skips
this whole method.
Setup via:
```
vendor/bin/typo3 setup --no-interaction --force --server-type="apache"
```
now completes successfully where it previously failed with:
```
A database is required for the method:
MySQLMetadataProvider::__construct.
```
So why did the e2e/acceptance tests not catch this?
Two independent gaps, stacked:
1. The main Playwright acceptance suite never touches this code path
at all. runTests.sh's runPlaywright() hardcodes TYPO3_DB_DRIVER=
sqlite (there's even a @todo: !Add support for all available
database engines" comment on it).
SetupCommand/InstallerController both special-case pdo_sqlite to
skip getDatabaseList() entirely, since SQLite has no concept of
"select a database from a server." So the suite that runs
`vendor/bin/typo3 setup` for every other acceptance test
structurally can't exercise the MySQL branch.
2. The dedicated MariaDB/MySQL installer e2e tests do reach this
code, but two things let it slip through:
- InstallerController::showDatabaseSelectAction() wraps
getDatabaseList() in try { … } catch (\Exception $exception) { ... }
the exception is swallowed into a soft error list, and the endpoint
still returns success with a rendered (if empty/erroring) page. The
CLI path has no such fallback SetupCommand::selectAndImportDatabase()
catches DBALException and hard-fails).
- install-mariadb.spec.ts / install-mysql.spec.ts only assert the
heading "Select a database" is visible, then immediately click
#t3-install-form-db-select-type-new and type a brand-new database
name: the "create new database" shortcut. They never assert the
existing-database list actually populated or that no error banner
appeared, so the swallowed exception is invisible to the test.
The CLI flow the user hits (dbname=db, pre-created by ddev) has no
"new database" escape hatch — it must resolve an existing database
via the now-broken list.
So the bug requires exactly what `ddev tryout delete` does — CLI setup,
MySQL-family driver, against a database that already exists —
and none of the current test coverage reproduces that specific
combination.
This needs coverage in a follow-up patch with proper adjustments.