[magnusp]

What

Tests only, for the four behaviors Wave 0 of #195 names. No production code, pom.xml, or dependency change; no existing assertion edited.

1. Attached-schema table metadata (LibsqlResultSetMetaDataTest). SELECT id, name FROM aux.People drives pragma_database_list and the per-schema lookup, so findSchema and loadAttachedTableInfo run. The stub answers a different People in main (nullable, no autoincrement) and in the attached schema Aux (NOT NULL, INTEGER PRIMARY KEY AUTOINCREMENT), so the assertions on getTableName/isNullable/isAutoIncrement only pass when the attached catalogue wins. A second test queries ghost.People with ghost absent from the database list and asserts no table metadata.
2. Date/time tails (LibsqlResultSetTest). +02:00 suffix, Z suffix, bare 10:30:45 defaulting to 2000-01-01, and BLOB rejection with "Bad value for type LocalDate" / LocalTime / LocalDateTime (exact target-type spelling from the code).
3. Savepoint edges (LibsqlSavepointTest). Two savepoints named dup: releasing the later one must keep the earlier active, so the next statement runs on the same stream with no BEGIN, and rollback to the earlier one still works. Releasing the only savepoint of a transaction started in auto-commit ends the transaction, so the following commit() sends nothing and exactly one stream close is issued.
4. Batch vs close() (LibsqlStatementConcurrencyTest). A batch is held inside the connection round trip with the existing Pause seam while close() lands on the statement, so the interleaving is deterministic; executeBatch then fails with Statement is closed.

Mutation probe

Each new test was run alone against a scratch break of the behavior it pins, then the break was reverted with git checkout -- <file>. All nine bite; none stayed green.

Probe lane: ./mvnw -pl libsql-jdbc -am -Dtest='<Class>#<method>' -DfailIfNoTests=false -Dsurefire.failIfNoSpecifiedTests=false test

Test	Break	Observed failure
LibsqlResultSetMetaDataTest#resolvesAttachedTableMetadataThroughPragmaDatabaseList	findSchema: name.equalsIgnoreCase(schema) → name.equals(schema)	:65 expected "People" but was ""
LibsqlResultSetMetaDataTest#reportsNoTableMetadataForASchemaMissingFromPragmaDatabaseList	findSchema: return null → return schema when no database matches	:81 Expecting empty but was: "People"
LibsqlResultSetTest#textDateTimeWithAnOffsetSuffixIsConvertedToTheSystemZone	parseSqliteDateTime: guard if (sign > 9) → if (false && sign > 9)	:290 SQLException: Bad value for type LocalDateTime: 2024-03-15 10:30:00+02:00
LibsqlResultSetTest#textDateTimeWithAZSuffixIsReadAsUtc	parseSqliteDateTime: skip the Z branch (if (false && ...))	:305 SQLException: Bad value for type LocalDateTime: 2024-03-15T10:30:00Z
LibsqlResultSetTest#bareTimeTextGetsTheYear2000DefaultDate	LocalDate.of(2000, 1, 1) → LocalDate.of(1970, 1, 1)	:316 expected 2000-01-01T10:30:45 but was 1970-01-01T10:30:45
LibsqlResultSetTest#blobRejectsDateAndTimeGettersWithTheTargetTypeName	BLOB branch message → "Cannot read " + targetType + " from a BLOB"	:328 Expecting message "Bad value for type LocalDate"
LibsqlSavepointTest#releasingTheLaterOfTwoSavepointsWithTheSameNameKeepsTheEarlierOneActive	releaseSavepoint: savepointNames.lastIndexOf → indexOf	:225 actual statements contain a BEGIN before the first INSERT (the transaction was ended early)
LibsqlSavepointTest#releasingTheOnlySavepointOfAnAutoCommitTransactionLeavesCommitWithNothingToSend	releaseSavepoint: drop closeStreamQuietly() in the outermost branch	:250 actual statements end with "COMMIT" (unexpected)
LibsqlStatementConcurrencyTest#executeBatchThatLosesTheRaceToCloseFailsWithStatementClosed	executeBatch finally: clearBatch() → batchStatements.clear()	:200 Expecting code to raise a throwable

Lanes

Command	Result
./mvnw -pl libsql-jdbc -am test	EXIT=0, 383 tests, 0 failures, 0 errors (baseline 374 + 9 new)
./mvnw spotless:apply	no changes
./mvnw spotless:check	BUILD SUCCESS
probe lane (9 runs, table above)	each EXIT=1, production file reverted after every run

Discovery

executeBatch's finally { clearBatch(); } calls checkClosed() through clearBatch. When a statement close() lands while the batch round trip is in flight, a batch whose steps all succeeded reports SQLException("Statement is closed") instead of the update counts. Pinned as the current behavior and not fixed here, since a fix changes behavior under a race; it sits next to the #197 update-count snapshot and #211 close-path work, so the root can decide.

The socketTimeout URL case from the wave list is deliberately absent: #196 owns it.

Fixes #195

[magnusp]

Root verdict: PASS at 4df54b1.

Independent verifier re-ran the gates on a clean worktree: 383 tests, 0 failures, spotless green, the concurrency class 7 times with no flake, and two mutation probes that bit and were reverted. Diff is 235 insertions across four test files, no production change, no existing assertion edited. CI is green and the PR is CLEAN.

One note carried forward: the duplicate-savepoint test name promises more than observable behavior can distinguish. Its mutation does bite the truncation path, so it stays as written. Countersign: root, 4df54b1.