[#1075] Test the JDBC guards against a neighbouring MySQL database and Oracle schema

[vharseko]

Fixes #1075

Test-only change: the MySQL/Oracle twin of #902's postgres case (PgSqlTestCase.testAnOpenIsAnsweredForByNoTableOfASchemaOffTheSearchPath), shaped by what the drivers actually return rather than by the case the issue sketched.

What the drivers do (measured)

Probes against the drivers this backend ships (Connector/J 9.2.0 on mysql:9.2, ojdbc8 23.7 on oracle-free 23.26.2), with a neighbour holding the same table and index name:

Setting	getCatalog / getSchema	What the lookup returns for the neighbour	What guards against it
MySQL, default databaseTerm=CATALOG	database / null	nothing with the catalog passed; TABLE_CAT=<neighbour> rows with null	two layers: the catalog passed to getTables/getIndexInfo, and isSameCatalog in TableScope.covers()
MySQL, databaseTerm=SCHEMA	null / database	TABLE_CAT=def, TABLE_SCHEM=<neighbour>	only the mysql branch of schemaPathOf() + the schema half of covers()
Oracle	null / OPENDJ	nothing without a grant; SCHEM=<neighbour> once grant select is given	only the oracle branch of schemaPathOf() + the schema half of covers()

So the case #1075 sketched — two databases on the default MySQL connection — goes red only when both layers give way at once; a loosening of either one alone, which is what the issue worries about, stays green there. Each layer needs a case of its own.

Changes

• JDBCStorageRetryTest — three mock cases on the mysql branch:
  • testTheMySqlGuardsAskInTheDatabaseOfTheConnection — getTables/getIndexInfo are asked with con.getCatalog(), not any();
  • testAnIndexOfAnotherDatabaseAnswersForNoneOfThisOne — an index row of another TABLE_CAT is not this backend's, and create index is issued;
  • testUnderDatabaseTermSchemaAnIndexOfAnotherDatabaseAnswersForNoneOfThisOne — the same with the row shape measured under databaseTerm=SCHEMA.
• MySqlTestCase.testAnOpenIsAnsweredForByNoTableOfAnotherDatabase — live, data-provided over catalog (default) and schema (?databaseTerm=SCHEMA); neighbour database made as root, asserted through information_schema.
• OracleTestCase.testAnOpenIsAnsweredForByNoTableOfAnotherSchema — live; the neighbour user is made by SYSTEM (the container's ORACLE_PASSWORD is the withPassword value — checked in testcontainers 1.20.6) and granted to opendj, without which the dictionary views show nothing and the case would pin nothing. The manual docker run hint now passes -e ORACLE_PASSWORD=password.
• TestCase.dropStaleNeighbours() — a best-effort hook run in setUp() before dropStaleTrees(). That listing spans every database/schema the connection sees, so a neighbour left by a killed run on a long-lived manual container would make its unqualified (mysql) or unprivileged (oracle) drop table fail and skip the whole class on every later run.

No production code changes.

Verification

Mutants of JDBCStorage, each run against the mock class and the live case:

Mutant	Mock cases	Live MySQL	Live Oracle
m1: no catalog passed to getTables/getIndexInfo	red: …AskInTheDatabase…	green	—
m2: covers() ignores the row's database	red: …AnIndexOfAnotherDatabase…	green	—
m3: m1 + m2	both red	red (catalog)	—
m4: mysql branch of schemaPathOf() returns null	red: …UnderDatabaseTermSchema…	red (schema)	—
m5: oracle branch of schemaPathOf() returns null	green	—	red (ORA-00942)

Head, one JVM per class: JDBCStorageRetryTest 116/116, MySqlTestCase 97/97, PgSqlTestCase 94/94, jdbc EncryptedTestCase 38/38, OracleTestCase 92 passed + 1 skip (the existing SkipException of testAConnectionGetsItsLockBoundBackAfterADdl on Oracle — "no session lock bound"). MsSqlTestCase is not runnable locally and is left to CI.