[ctodTT]

This change adds a secondary parser for pytest JUnit XML files specifically created by the tt-mlir/test/python/golden/ tests. This is part of a broader effort to get the results of these tests into Superset, and the collect_data action is a key part of that pipeline.

Changes

• The OpTests pydantic model has been modified to fit the one in data-airflow/dags/pipelines/ml_kernel_op_test/pydantic_models.py This is to ensure a common schema across the sftp call that the collect_data action performs
• A new parser specific to the bespoke rich test result schema defined by tt-mlir/test/python/golden/conftest.py. It contains information about input tensors, hardware, git sha, and any other parameters that the test is run with.
• A new filter to parser selection has been added that will pass reports that are both from main and part of a job with "builder" in the name to the BuilderPytestParser.

[vmilosevic]

Parser Registration

Issue: Parsers registered with test_parser.py produce pydantic_models.Test objects, but your parser is producing pydantic_models.OpTest objects, which have a separate collection path and are uploaded in separate files.

Data Flow Paths:

pydantic_models.Test path:
generate_data → create_pipeline_json → create_cicd_json_for_data_analysis → ...

pydantic_models.OpTest path:
generate_data → create_optest_json → create_optest_reports

Solution: Register your parser in optests.py instead of test_parser.py

Testing

This project has multiple contributors, and the only way to ensure that others won't break the changes you made is to add regression tests. Once changes are merged to main, all pipelines start using the new collection code, so it's easy not to notice that something got broken.

Required Test Cases:

Unit tests for builder_pytest_parser.py

Integration tests: Add sample data files to collect_data/test/data/ and add test cases that try to read the actual files, similar to the test here

[vmilosevic]

I believe pipeline object already contains job information you need

[vmilosevic]

job_id_to_name = {j.github_job_id:j.name for j in pipeline.jobs}

[vmilosevic]

Is this exclusive or can both parsers extract data from same files?

[ctodTT]

Technically both could extract data from the same file, but we only want this data to specifically be moved through the builder pytest parser pipeline. The other parser will currently fail due to the suite-wide properties I define to mark this as a builder test artifact.