The GitHub Actions job "Tests (AMD)" on airflow.git/fix-hitl-review-output-type-loss has failed. Run started by GitHub user ColtenOuO (triggered by ColtenOuO).
Head commit for run: a632d04087adc871a7d1687bb99f58ec12d9bfcc / ColtenOuO <[email protected]> Fix output type loss in AgentOperator's human-in-the-loop review Add dump_output_to_json() to utils/output_type.py, alongside rehydrate_pydantic_output(), so both directions of the review round-trip live in one module. Route all three serialization sites through it: HITLReviewMixin._to_string, AgentOperator.regenerate_with_feedback and LLMApprovalMixin.defer_for_approval. Serialize non-str, non-BaseModel output with TypeAdapter(...).dump_json() rather than str(output), which produced a Python repr instead of JSON, and fall back to str(output) when the value's type has no pydantic schema. Split schema build from validation in rehydrate_pydantic_output(). An output_type that TypeAdapter cannot build a schema for now falls back to json.loads() instead of raising PydanticSchemaGenerationError or AttributeError, neither of which the previous except clause caught. Use rehydrate_pydantic_output() in AgentOperator.execute()'s HITL branch in place of the bespoke json.loads()/except fallback. With the default output_type=str the approved string is no longer parsed as JSON. Add tests for both schema-build guards, for dump_output_to_json(), and for output_type=str with a JSON-parseable approved string, and fold the four TestToString cases into a single delegation test. Report URL: https://github.com/apache/airflow/actions/runs/30743035213 With regards, GitHub Actions via GitBox --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
