I have released PyAthena 3.37.0, a minor release in the 3.x maintenance line. It backports SQLAlchemy ARRAY support and several fixes from master. The release notes contain the complete list of changes.

SQLAlchemy ARRAY support

PyAthena now supports SQLAlchemy's standard ARRAY type alongside AthenaArray, with typed reads and writes for native Athena arrays. Reflection preserves the element types of ARRAY columns instead of reporting them as String. Typed SELECT results preserve nested arrays and strings such as "a,b", "001", and "null", including the distinction between a string and SQL NULL.

ARRAY expressions support indexing, slicing, concatenation, and ANY/ALL comparisons in SELECT and WHERE clauses. On Iceberg tables, indexed and sliced UPDATE assignments can replace individual elements or resize a range within an array. Indices are one-based by default, and slice stops are inclusive.

Typed ARRAY results use a JSON projection in the generated SQL. Ordered, DISTINCT, and compound typed ARRAY queries require explicit SELECT columns. Label literal SQL expressions and prefer SQLAlchemy expressions for ORDER BY. Decimal ARRAY binds require an explicit Numeric(precision, scale) to preserve fractional values. Raw SQL queries and arrays with unknown element types retain the cursor's existing conversion behavior. The ARRAY guide for this release covers examples and the query restrictions in detail.

Type and cursor fixes

• STRUCT DDL: CREATE TABLE now renders Hive STRUCT<name:type> syntax, including STRUCT values nested inside MAP and ARRAY columns. Athena rejected the previous ROW(...) output. Integer types inside ARRAY, MAP, and STRUCT column definitions now render as INT.
• DOUBLE casts: Casting to Double or DOUBLE_PRECISION now produces DOUBLE, fixing the previous use of the 32-bit REAL type.
• Binary NULLs: The pandas and Arrow cursors now distinguish binary NULLs (None) from empty binary values (b"") when fetching CSV results with default settings and converters.

Thanks to mvanhorn for the STRUCT DDL fix and aminghadersohi for reporting the issue.

Upgrading

Install this release with:

python -m pip install --upgrade 'PyAthena[SQLAlchemy]==3.37.0'

The 3.x line continues to support Python 3.10 and keeps its sqlalchemy>=1.0.0 dependency requirement. The ARRAY changes were checked at compilation level with SQLAlchemy 1.4.54; the test suite runs on SQLAlchemy 2.0. Applications that inspect reflected types or compare compiled DDL strings should account for the ARRAY, STRUCT, and INT changes above.

The release workflow now runs every test suite on every supported Python version before publishing. The full changelog includes the CI and test coverage improvements as well.