Lee-W commented on PR #71477: URL: https://github.com/apache/airflow/pull/71477#issuecomment-5740313838
`read_guide_docs` now uses `git cat-file --batch`: at `providers-amazon/9.36.0` that is one subprocess and ~0.05s against 107 and ~1.9s, for a dict that compares equal to the loop's key for key. Parsing stays on bytes since the header counts them, and content decodes as explicit UTF-8 rather than following locale, with a decode failure left to propagate. The `discover_classes_from_provider` docstring now uses "all 12 Module fields, plus `guide_url` when a how-to guide documents the class". `test_module_contract_omits_guide_url_for_undocumented_classes` now asserts on `ModuleContract.model_validate(_module_payload()).guide_url`, and the round-trip one is renamed `test_module_contract_preserves_guide_url_value` and reads the attribute off the model as well. `_validate` still returns the payload it was handed — switching it to `model_dump()` would inject defaults into every emitted entry — so the tests moved to the model instead. `AGENTS.md` drops "Nothing declares that link" for the justification that holds: `provider.yaml`'s `how-to-guide` is CI-enforced by `check_doc_files`, but it names a whole page and never a section, and it covers only operators, sensors and transfers — not the toolset, hook and decorator pages this reads. --- `.provider-detail-page .module-actions` gets `flex-wrap: wrap`, and `git_ls_tree` runs with `core.quotePath=false`. I kept your framing on the second one: no non-ASCII path exists under `providers/` today, so it closes the divergence between the two readers rather than fixing an observed break. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
