0.2.2

Added

• SQL-native time series UDTF template (time_series_sql_udtf) with predicate pushdown hints
• Support for aggregate hints (avg, sum, min, max, count) in SQL-native UDTF
• Granularity hints for time-based grouping in SQL-native UDTF
• Pushdown optimization to avoid raw datapoint retrieval when using SQL-native UDTF

Improved

• Enhanced time series UDTF generation with SQL-native template support
• Better predicate pushdown support for time series queries

0.2.1

Fixed

• Fixed missing json import in udtf_function.py.jinja template that caused NameError: name 'json' is not defined when processing data model UDTFs with array or relationship properties
• Fixed time_series_datapoints_detailed_udtf to properly parse protobuf responses instead of attempting to parse JSON, which was causing zero-row results
• Implemented proper protobuf parsing with JSON fallback for detailed time series UDTF, extracting instanceId, status codes, and symbols from protobuf responses
• Ensured all helper functions have access to json module by adding import at the beginning of eval() method

Improved

• Enhanced protobuf parsing logic to extract detailed information (status codes, status symbols, external_id, space) for each datapoint in detailed UDTF
• Improved error handling with proper fallback to JSON parsing when protobuf is unavailable
• Better support for all datapoint types (numeric, string, aggregate) in protobuf responses

0.2.0

Added

• Added _error column to UDTF output schemas for better error visibility in query results
• Direct REST API calls in generated UDTFs (no Cognite SDK dependency at runtime)
• Enhanced error messages with error categories (AUTHENTICATION, CONFIGURATION, NETWORK, UNKNOWN)
• Protobuf parser support for time series datapoints with JSON fallback
• HTTP client module with OAuth2 token caching and retry logic
• Support for distributed limit calculation across multiple time series items

Changed

• UDTFs now use direct REST API calls instead of Cognite SDK at runtime
• Improved request payload alignment with SDK behavior (ignoreUnknownIds, limit distribution)
• Updated time series UDTF templates to match SDK's retrieve_arrays behavior
• Enhanced error handling with structured error messages in output

Fixed

• Fixed time series UDTF request payload to match SDK behavior (ignoreUnknownIds: True)
• Fixed limit distribution across multiple time series items (distributes total limit, not per-item)
• Fixed CI/CD compatibility issues (ruff UP038, mypy pyspark imports)
• Fixed dependency resolution for cognite-databricks integration
• Fixed get_file method to handle _session and _catalog suffixes

Improved

• Removed debug functionality from all UDTF templates for cleaner output
• Improved exception handling specificity throughout codebase
• Enhanced code quality (linting, type checking, formatting)
• Updated documentation to reflect direct REST API approach
• Improved import organization and code style alignment with pygen-main

Removed

• Removed time_series_datapoints_long_udtf and time_series_datapoints_multi_udtf templates
• Removed all debug-related code and columns from UDTF templates
• Removed Cognite SDK runtime dependency from generated UDTFs

0.1.0

Added

• Initial release of pygen-spark with UDTF generation for CDF Data Models
• Support for generating Python UDTFs from CDF views
• Time series UDTF support (datapoints, latest datapoints, long format)
• Type conversion utilities using PySpark as source of truth
• Connection configuration management
• Utility functions for consistent UDTF naming
• Support for predicate pushdown
• Configuration file support (TOML/YAML)
• Generic Spark support (works with any Spark cluster)