TL;DR

Refactored legacy filter query parameter to be filter_query JSON format.

Before

• filter
  • GET /api/pipeline_runs/?filter=created_by:alice
  • GET /api/pipeline_runs/?filter=created_by:me
• filter_query
  • GET /api/pipeline_runs/?filter_query={"and":[{"value_equals":{"key":"system/pipeline_run.created_by","value":"alice"}}]}

After

• filter
  • Converts created_by:alice or created_by:me to be {"and":[{"value_equals":{"key":"system/pipeline_run.created_by","value":"alice"}}]}
• filter_query no change.

What changed?

Functional

• None. From a user perspective, filter and filter_query behaves the same.

Other

• Replaced _build_filter_where_clauses function with _convert_legacy_filter_to_filter_query that converts legacy filter strings to filter_query JSON format
• Modified build_list_filters to convert legacy filters to filter_query format and process them through the unified filter_query pipeline
• Updated PageToken to store filter as None and use filter_query for all filtering operations
• Added validation to reject empty created_by filter values with a clear error message
• The "me" value in created_by:me filters is no longer resolved during conversion but handled downstream by _maybe_resolve_system_values

How to test?

uv run pytest tests/test_api_server_sql.py tests/test_filter_query_sql.py

• Test legacy filter conversion with created_by:alice filters
• Verify that created_by:me filters are converted but not immediately resolved
• Confirm that empty created_by: filters raise ApiValidationError
• Test pagination with legacy filters to ensure they're properly converted in page tokens
• Verify that unsupported filter keys still raise NotImplementedError

Why make this change?

This change unifies the filter processing pipeline by converting legacy filters to the modern filter_query format, eliminating duplicate SQL generation logic and ensuring consistent behavior between legacy and modern filter formats.