February 2026 (2026-02) Monthly Summary for apache/spark: Key features delivered: - Proto2 Extensions Support (ExtensionRegistry) for proto2 extensions in protobuf serialization/deserialization. This enables retention of extension fields during from_protobuf and to_protobuf when a file descriptor set is provided. - Introduced an ExtensionRegistry and a name-to-extensions map, wired through helper classes for schema conversion and serde, and used during DynamicMessage construction. - Feature gating via Spark configuration spark.sql.function.protobufExtensions.enabled to preserve backward compatibility. - Changes are supported by unit tests that validate basic behavior, extension handling in nested messages, and extensions defined across multiple files. Major bugs fixed: - Fixed data loss of proto2 extension fields during protobuf function usage by ensuring extensions are retained instead of being dropped when file descriptor sets are provided; aligns behavior with user expectations and protobuf semantics. This fixes SPARK-55062 related issues and related tests. Overall impact and accomplishments: - Dramatically improves data fidelity and interoperability for protobuf-encoded data in Spark SQL functions, reducing surprises for users importing/exporting proto2 data. - Enhances schema conversion and serde paths to correctly handle extensions, enabling more complete and future-proof protobuf workflows. - Strengthens Spark's protobuf feature parity with Java-based protobuf usage and supports more complex data models. Technologies/skills demonstrated: - Protobuf proto2 extensions, ExtensionRegistry, DynamicMessage, and descriptor handling - Spark SQL function integration and feature gating via configuration - Schema conversion, serde, and cross-file extension support - Unit testing and validation for extension semantics Deliverable trace: - Commit: dd5ce947d80855b3793e5f33e7cf51c593d897e6 - Related PR: SPARK-55062, closes #53828