[magnusp]

Description

Adds Liquibase integration tests to libsql-jdbc, derived from a throwaway prototype and then pulled up into the module.

The tests drive the in-tree driver through Liquibase's update, validate, snapshot-driven table rebuild, rollback, changeLogSync and typed-insert paths against a real sqld container. They live under io.github.magnusp.libsql.jdbc.liquibase and run under Failsafe (*IT), so ./mvnw test stays Docker-free.

No tracking issue; this is the test half of a driver-compatibility probe.

Driver gaps the prototype found

1. Product name prevents Liquibase from selecting its SQLite dialect. getDatabaseProductName() returns SQLite (libSQL), while SQLiteDatabase.isCorrectDatabaseImplementation requires exactly SQLite (equalsIgnoreCase). Liquibase falls back to UnsupportedDatabase and emits INT GENERATED BY DEFAULT AS IDENTITY ..., which sqld rejects. Returning exactly SQLite and keeping the libSQL identity in getDriverName() would close this. Pinned by databaseFactoryFallsBackToUnsupportedDatabase(); the working tests construct SQLiteDatabase explicitly until then.
2. jdbc:libsql: is not resolvable through Liquibase's URL path. DatabaseFactory.openDatabase(...) cannot derive a driver class because Liquibase's SQLite handler only claims jdbc:sqlite:. Applications must set driver=io.github.magnusp.libsql.jdbc.LibsqlDriver or pass a DataSource. Pinned by databaseFactoryCannotResolveALibsqlUrlToADriver().

Separately, Liquibase 5.0.3's SQLite dropColumn is broken (the single-column form fails validation and the nested form rebuilds without dropping the column), so the changelogs avoid it.

What is covered

• update + re-run idempotency and validate()
• add column, rename column (Liquibase snapshot-driven table rebuild), unique and non-unique indexes
• inline foreign key creation and getImportedKeys round-trip
• rollback(N) in reverse order
• typed inserts: boolean, decimal, double, bigint, timestamp, date, time, clob
• changeLogSync

libsql-jdbc/pom.xml now copies the shared Docker config to target/docker-config and sets DOCKER_CONFIG for the Failsafe fork, so the sandbox workaround is part of the build rather than a shell preamble.

Type of Change

• Bug fix (non-breaking change fixing an issue)
• New feature (non-breaking change adding functionality)
• Breaking change (fix or feature causing existing functionality to not work as expected)
• Documentation update
• Refactoring / tooling / CI improvement

Checklist

• My code follows the code style of this project (./mvnw spotless:check).
• I have added tests that prove my fix is effective or that my feature works.
• All new and existing tests passed in the affected reactor (./mvnw -pl libsql-jdbc -am verify).
• I have updated documentation where appropriate. (No user-facing behavior changed.)

Verification

./mvnw -B -pl libsql-jdbc -am verify -Dit.test=LiquibaseIT

• LiquibaseIT: 8 tests, 0 failures
• Affected reactor: 1511 unit tests + 8 ITs, 0 failures
• ./mvnw spotless:check: passing

The integration tests start the pinned ghcr.io/tursodatabase/libsql-server image and do not require a pre-existing sqld.

Tracked gaps

• Report database product name as exactly SQLite so dialect detection works #261 — product name must be exactly SQLite for dialect detection
• Document the supported Liquibase bootstrap for jdbc:libsql: URLs #262 — Liquibase cannot derive a driver from jdbc:libsql:; document explicit driver/DataSource setup

[magnusp]

The two driver gaps this PR pins are now tracked as #261 (product name / dialect detection) and #262 (Liquibase URL bootstrap and documentation).