Modules

To extend the built-in Oracle-to-H2 rewrites without changing this module, subclass H2SqlRewriteDriver, supply your function to super(rewriter, "jdbc:testoracleextended:h2:"), and register the subclass with java.sql.DriverManager before creating the extension (via META-INF/services/java.sql.Driver or an explicit DriverManager.registerDriver call). wrap(String) gives the H2 URL the driver accepts; unwrap(String) recovers the plain H2 URL. H2DatabaseExtension prefers one registered subclass accepting that prefix, rejects multiple such subclasses, and otherwise uses the built-in driver. The built-in driver's connection path also delegates to that registered subclass, so a previously registered built-in instance cannot silently bypass the custom rewrite function. JDBC interface unwrapping returns the rewriting wrapper; intentionally unwrapping an H2 implementation class exposes the underlying object and bypasses rewriting. The Oracle admin hook (OracleAdminProvisioner) lets consumers add owner-scoped provisioning (for example per-schema grants and synonyms) from their own project, keeping the extension generic.

Oracle compatibility is partial: H2 tables are not physically partitioned and the rewrite removes PARTITION(...) on reads, so reads include the whole fixture table, not just the named partition. Partition-targeted writes are rejected instead of silently broadening their effect. Ordered XMLCAST(XMLAGG(XMLELEMENT(name, scalar_text) ORDER BY ...) AS CLOB) is supported for one scalar text argument, preserving surrounding RTRIM expressions, XML characters, nulls and large character results; it is not a general XML engine. The H2 _ROWID_ and ORA_HASH shim support repeatable reader bands for unchanged fixture rows but do not reproduce Oracle row identifiers or hash values. Whole-query application templates are not emulated by this driver. Single-named-source CONNECT BY queries with LEVEL termination are lowered to recursive CTEs, including qualified PRIOR columns, START WITH, post-hierarchy filters and duplicate source rows. This subset does not implement Oracle cycle detection, NOCYCLE, hierarchy path functions, sibling ordering, joined/derived sources, or binds inside the hierarchy block; unsupported shapes must not be treated as Oracle-equivalent. A missing or zero upper bound still emits the root, as on Oracle. Date arithmetic translation recognizes explicit Oracle date expressions and TO_DATE projection aliases, not arbitrary JDBC column types; date differences use numeric days at Oracle DATE second precision. Date-plus/minus-numeric CASE operands support searched, simple and nested cases when every result branch is a numeric literal, bind, TO_NUMBER, numeric cast, arithmetic over those forms, or NULL (including an omitted ELSE). Conditions and selectors are preserved, fractional offsets retain time-of-day, and unknown column/function result types or mixed date/text branches are left untranslated rather than assumed numeric. Date aliases are inferred from their own CTE projections, not unrelated projections sharing the same column name; this is conservative inference, not general SQL type resolution. DBMS_XMLGEN.GETXML executes a single SELECT on the active connection, serializes ROWSET/ROW XML, omits null columns, and returns null for no rows. It excludes writes, multiple statements and WITH queries, but does not sandbox user-installed SQL functions. XMLTABLE/PASSING, namespaces, arbitrary XML column names and complete Oracle XML serialization parity remain unsupported. In ORACLE_EXTENDED mode the ALL_TABLES, ALL_TAB_COLS, ALL_TAB_COLUMNS and ALL_CONSTRAINTS dictionaries are auto-installed as H2 views over INFORMATION_SCHEMA, so they always reflect the real schema (including SA.X FOR PUBLIC.X synonyms reported under the SA owner). The partition/app-metadata dictionaries (ALL_TAB_PARTITIONS, ALL_PART_TABLES, ALL_PART_KEY_COLUMNS, ALL_PART_INDEXES, ALL_IND_PARTITIONS and the USER_* index views) are installed as empty Oracle-typed scaffold tables that the consumer seeds with its own partition data; neither dictionary type should be re-created by the consumer. Schema-qualified references must be backed by real H2 schemas or synonyms; no schema prefix is silently removed. SHA-1 character input is encoded as UTF-8 for H2, which matches Oracle only when the Oracle database uses a compatible character encoding.