[hutiefang76]

When a command in a batch fails, both Statement and PreparedStatement currently rethrow a plain SQLException. Callers that catch BatchUpdateException cannot inspect which commands executed before the failure.

Wrap the error in BatchUpdateException with the executed prefix of update counts and preserve its SQLState, vendor code and cause. Use the long[] constructor for both executeBatch and executeLargeBatch, keeping the existing stop-at-first-failure and transaction behavior.

The regression tests fail against the original implementation. After the change, TestBatch (17 tests) and TestPrepare (28 tests) pass with the locally built native driver on macOS arm64/JDK 17:

GEN=ninja make release
java -cp build/release/duckdb_jdbc.jar:build/release/duckdb_jdbc_tests.jar org.duckdb.TestDuckDBJDBC TestBatch
java -cp build/release/duckdb_jdbc.jar:build/release/duckdb_jdbc_tests.jar org.duckdb.TestDuckDBJDBC TestPrepare

make test reaches an unrelated failure in test_scalar_function_vectorized_parallel_streams (a result unexpectedly becomes NULL). That test also fails in three isolated runs with the original PreparedStatement implementation. I am investigating that separately; this PR does not claim the full suite passes. The modified Java files pass clang-format.

AI assistance was used for implementation and review.

[staticlibs]

Thanks for the PR! Looks good to me.