Remove and rewrite unnecessary tests in the SQLAlchemy dialect tests #1242

Summary

Part of #1239. Files: sqlalchemy/test_base.py, sqlalchemy/test_compiler.py, sqlalchemy/test_array.py, sqlalchemy/test_struct.py, sqlalchemy/test_map.py, sqlalchemy/test_types.py, aio/sqlalchemy/test_base.py (sqlalchemy/test_temporal.py has no findings). At master 59169c5, with the cross-check corrections, the files have 333 test functions: 237 KEEP, 62 REMOVE, 15 MERGE, 19 REWRITE. The changes save about 140 Athena queries and 2 metadata API calls per run (100 from the binary merges, about 31 from the across-cursor reductions, 5 from the nested-struct removals, 2 from the test_filter_func rewrite, 1 each from the connection-level callback and aio unicode removals), plus 3 queries where S3 Tables are configured. The compliance suite under tests/sqlalchemy/ is not changed.

Paths are relative to tests/pyathena/. [original] marks a test that existed at 21cbf05 (v3.15.0).

Order

• No open PR (Move S3File writes onto a multipart write session #1232, Read UNLOAD timestamps at microseconds #1233, Break result sets after failed or cancelled fetches and drop the cross-thread guards #1238) changes these files. Read CSV dates as date32 and temporal text NULL as null in ArrowCursor #1236, merged on master at 7e5d9ee, does not change them either.
• Section 1, TestSQLAlchemyAthena::test_select_nested_struct_query and ::test_select_array_with_nested_struct: one cover, test_cursor.py::TestCursor::test_complex_with_type_hints, is a REWRITE in the top-level sub-issue. That rewrite keeps the three hinted expressions in one query, so the removal can land before or after it.
• Within this sub-issue, these removals land with or after the change that keeps their cover:
  • TestAthenaDialect::test_get_view_definition_keeps_blank_lines after the reduction of TestSQLAlchemyAthena::test_get_view_definition_across_cursor_types (the kept case still checks the blank lines).
  • TestAthenaTypeCompiler::test_visit_struct_single_field after the MERGE of the test_visit_struct_*/test_visit_map_* tests.
  • TestArrayWriteIndexType::* after the REWRITE of TestArrayUpdateCompiler::test_bound_index_uses_write_index_processor.
  • The aio test_binary_parameters_and_literals MERGE with the sync MERGE of the same name.

Findings

1. sqlalchemy/test_base.py

2. sqlalchemy/test_compiler.py

3. sqlalchemy/test_array.py

4. sqlalchemy/test_struct.py

5. sqlalchemy/test_map.py

6. sqlalchemy/test_types.py

7. aio/sqlalchemy/test_base.py

Notes

• Borderline KEEPs:
  • sqlalchemy/test_base.py::TestAthenaDialect::test_compliance_suite_registry_matches_entry_points checks test infrastructure; a drift between tests/sqlalchemy/__init__.py and the entry points makes the compliance suite test the wrong dialects without a failure.
  • TestAthenaDialect::test_conn_str_boolean_options overlaps the original test_conn_str_verify, test_conn_str_kill_on_interrupt and test_conn_str_result_reuse_enable; it is offline and adds the true values, glue_metadata_fallback and the aio dialect.
  • TestAthenaDialect::test_internal_cursor_leaves_out_default_cursor_options calls the private _internal_cursor; it is the regression test for the TypeError with DataFrame-only cursor_kwargs (commit 45639cd).
  • TestAthenaDialect::test_column_lookup_retry_config captures the retry_config kwarg; its [cursor_kwargs] case is the only test of a user retry policy overriding the dialect's (commit 461c0d1).
  • TestAthenaDialect::test_table_level_reflection_of_failed_lookup and ::test_empty_information_schema_result_is_a_missing_table assert query counts and info_cache == {}; they are the federated-catalog and throttling-versus-absence contracts of Read the dialect's own queries through an API cursor and resolve federated absence from information_schema #798.
  • TestSQLAlchemyAthena::test_throttled_reflection_reads_glue (sync and aio) asserts the list of patched client calls; "Glue answered each throttled request once, without retrying" is the contract of Fix metadata reflection throttling and reuse listed metadata #777/Answer throttled metadata requests from Glue in the cursor #803.
  • sqlalchemy/test_compiler.py::TestAthenaDDLCompiler::test_scalar_and_float_column_ddl_is_unchanged is the only test of column-level Float(24) -> FLOAT after the SQLAlchemy type compiler emits invalid DDL for empty STRUCT, JSON, and CLOB columns #886 split.
  • sqlalchemy/test_array.py::TestArrayValueProcessor::test_array_result_conversion probes the internal {"_pyathena_array": ...} wrapper in two lines; the other cases are result-processor contracts.
  • aio/sqlalchemy/test_base.py::TestAthenaAioAdaptedCursor::test_execute_closes_cursor_off_event_loop_when_fetch_fails asserts the thread of close(), which is the documented contract of AthenaAioAdaptedCursor.execute.
• Unused after the changes:
  • sqlalchemy/test_base.py: the imports DefaultDialect, AthenaAioAdaptedCursor and AthenaTypeCompiler.
  • sqlalchemy/test_array.py: the imports sqltypes and _ArrayWriteIndexType; the class TestArrayWriteIndexType.
  • sqlalchemy/test_map.py: the whole file.
  • sqlalchemy/test_struct.py: the import STRUCT.
  • sqlalchemy/test_types.py: the import ischema_names.
  • sqlalchemy/test_compiler.py: STRUCT and MAP stay in use if the merged table keeps their cases.
  • The _bind_processors REWRITEs use recording_engine; move it from sqlalchemy/test_base.py to tests/pyathena/util.py.
• Each row's cover is re-checked before the row is applied. A row whose cover does not hold becomes MERGE or KEEP.