ci: bring back Oracle database coverage in GitHub Actions [full-ci]

[oc-tmueller]

Description

The drone → GitHub Actions migration dropped the Oracle DB pipeline, and #41555 then removed the installer-side Oracle support, so an instance can no longer be installed against Oracle at all — even though the whole Oracle runtime layer (ConnectionFactory, OracleConnection, OracleMigrator, AdapterOCI8, OCIExpressionBuilder, plus Oracle branches in AppConfig, AllConfig, Files\Cache\Cache and many tests) is still shipped and untested.

This restores just enough to get that layer under test again.

1. CLI installer (feat:)

A partial revert of 95e3cde: OC\Setup\OCI, its dbSetupClasses entry, dbconnectionstring handling in AbstractDatabase, and the --database-connection-string option of maintenance:install.

The web installer is deliberately left unchanged — no template or setup.js change, and hasOracle is not re-added. getSupportedDatabases(false) reads the supportedDatabases config, which still defaults to sqlite, mysql, pgsql, so Oracle is not offered in the UI. This exists for CI coverage of code we still ship; it is not a re-declaration of production support for Oracle.

2. CI job (ci:)

php-unit.yml gains a databases input (defaulting to today's five jobs, so nothing changes for the existing matrix) and an oracle service on gvenzl/oracle-free:23-slim-faststart (23ai), using the same "empty image string disables the service" idiom as the existing mysql/postgres services.

Notable details:

• The app user lives in the FREEPDB1 pluggable database, so --database-host is not passed; a connect string carrying SERVICE_NAME is used instead, and ConnectionFactory then takes its connectstring path.
• oci8 is only requested from setup-php for Oracle jobs (it is built from source, so it costs a few minutes).
• DB_TYPE is now exported to the Run PHPUnit step for Oracle only, so test-phpunit.sh narrows the run to --group DB as drone did. The other databases keep running the full suite exactly as today.

As in drone ("Note: do not run Oracle by default in PRs"), Oracle does not run on every PR: it is gated on full-ci in the PR title, plus a new nightly workflow.

3. One Oracle-specific bug found this way (fix:)

Oracle cannot store empty strings, so the file cache converts '' → null before writing a row (#31692). For the storage root, whose path is the empty string, that left md5(null): a PHP 8 deprecation, and a path_hash that was the hash of null rather than of ''. This commit is separable if you would rather see it on its own.

Verification

Locally, against a real gvenzl/oracle-free:23-slim-faststart with oci8 3.4.0 and PHP 8.3:

• the exact maintenance:install line from the workflow succeeds — 48 tables, 35 sequences, 35 triggers, 65 migrations
• phpunit --group DB is green: Tests: 5421, Assertions: 35231, Skipped: 29

The 29 skips are pre-existing platform skips, including the three Oracle ones that were already in the tree (MDB2SchemaManagerTest::testAutoIncrement, MigratorTest::testDuplicateKeyUpgrade, ApiTest::testShareStorageMountPoint → doctrine/dbal#4687).

What only CI can prove is the runner-side oci8 build time and the oracle service's health under Actions — hence full-ci on this PR.

Types of changes

• Bugfix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Technical debt

Checklist

• Code changes
• Changelog items
• Documentation (not needed — no user-facing feature)
• Tests

🤖 Generated with Claude Code