Enable ruff B015 to catch silent no-op comparisons in tests by shahar1 · Pull Request #66977 · apache/airflow

shahar1 · 2026-05-15T04:59:10Z

Bare actual == expected expressions in tests silently evaluate and discard their result — the test passes regardless of correctness. This is exactly what happened in #66894 (four test_serialize methods in the Vertex AI triggers had been vacuously passing for years).

Changes

pyproject.toml: add B015 (useless-comparison) to extend-select. This rule fires on any bare comparison expression used as a statement.
16 test files: add the missing assert to all 26 existing violations found by the rule across airflow-core, task-sdk, and providers amazon, cncf/kubernetes, google).

With this in place, any future bare comparison in a test is a lint error caught before CI runs.

Was generative AI tooling used to co-author this PR?

Yes — Claude Code (Sonnet 4.6)

Generated-by: Claude Code (Sonnet 4.6) following the guidelines

Bare `actual == expected` expressions (missing `assert`) silently pass regardless of correctness. Enable ruff rule B015 (useless-comparison) to make this a lint error, and add `assert` to all 26 existing instances found across airflow-core, task-sdk, and providers (amazon, cncf, google).

`.calls` is not a real MagicMock attribute; accessing it auto-creates a child MagicMock. The expression `mock.calls[0] == []` was a silent no-op that B015 correctly caught. Adding `assert` exposed that the comparison always evaluates to False. Replace with `assert_not_called()`, which correctly verifies that `revoke_task` routes through `kube_scheduler.patch_pod_revoked()` rather than calling `patch_namespaced_pod` directly.

- test_bedrock: set ensure_unique_job_name on the operator before execute so the not-ensure-unique parametrize cases actually test the right path, and use pytest.raises for the cases where a conflict error is expected - test_emr_serverless (create): remove dead get_application mock setup and wrong call-count assertion; the operator uses waiters, not direct polling - test_emr_serverless (start job): correct expected call_count from 2 to 1; the operator calls get_application once (state CREATED is in SUCCESS_STATES so no retry loop)

jscheffl

Cool! I assume you can work on making CI green in parallel :-D

- Add missing `assert` to three bare comparisons caught by the new B015 rule: test_pod_template_file.py, test_otel_logger.py (×2) - Fix the MappedOperator task_type assertion in test_dag_serialization: `type(task).__name__` gives the wrapper class (`DecoratedMappedOperator`) but `operator_class["task_type"]` stores the underlying operator's name (`task.operator_class.__name__`, e.g. `_PythonDecoratedOperator`)

The operator uses json.dump() which writes strings incrementally to the file handle; the previous assertion expected bytes in a single write call. Also adds the missing mock_file.write.reset_mock() between the CSV and JSON sub-tests so write calls don't accumulate across sections.

shahar1 · 2026-05-15T04:59:10Z

Bare actual == expected expressions in tests silently evaluate and discard their result — the test passes regardless of correctness. This is exactly what happened in #66894 (four test_serialize methods in the Vertex AI triggers had been vacuously passing for years).

Changes

pyproject.toml: add B015 (useless-comparison) to extend-select. This rule fires on any bare comparison expression used as a statement.
16 test files: add the missing assert to all 26 existing violations found by the rule across airflow-core, task-sdk, and providers amazon, cncf/kubernetes, google).

With this in place, any future bare comparison in a test is a lint error caught before CI runs.

Was generative AI tooling used to co-author this PR?

Yes — Claude Code (Sonnet 4.6)

Generated-by: Claude Code (Sonnet 4.6) following the guidelines

Bare `actual == expected` expressions (missing `assert`) silently pass regardless of correctness. Enable ruff rule B015 (useless-comparison) to make this a lint error, and add `assert` to all 26 existing instances found across airflow-core, task-sdk, and providers (amazon, cncf, google).

`.calls` is not a real MagicMock attribute; accessing it auto-creates a child MagicMock. The expression `mock.calls[0] == []` was a silent no-op that B015 correctly caught. Adding `assert` exposed that the comparison always evaluates to False. Replace with `assert_not_called()`, which correctly verifies that `revoke_task` routes through `kube_scheduler.patch_pod_revoked()` rather than calling `patch_namespaced_pod` directly.

- test_bedrock: set ensure_unique_job_name on the operator before execute so the not-ensure-unique parametrize cases actually test the right path, and use pytest.raises for the cases where a conflict error is expected - test_emr_serverless (create): remove dead get_application mock setup and wrong call-count assertion; the operator uses waiters, not direct polling - test_emr_serverless (start job): correct expected call_count from 2 to 1; the operator calls get_application once (state CREATED is in SUCCESS_STATES so no retry loop)

jscheffl

Cool! I assume you can work on making CI green in parallel :-D

- Add missing `assert` to three bare comparisons caught by the new B015 rule: test_pod_template_file.py, test_otel_logger.py (×2) - Fix the MappedOperator task_type assertion in test_dag_serialization: `type(task).__name__` gives the wrapper class (`DecoratedMappedOperator`) but `operator_class["task_type"]` stores the underlying operator's name (`task.operator_class.__name__`, e.g. `_PythonDecoratedOperator`)

The operator uses json.dump() which writes strings incrementally to the file handle; the previous assertion expected bytes in a single write call. Also adds the missing mock_file.write.reset_mock() between the CSV and JSON sub-tests so write calls don't accumulate across sections.

shahar1 · 2026-05-15T04:59:10Z

Bare actual == expected expressions in tests silently evaluate and discard their result — the test passes regardless of correctness. This is exactly what happened in #66894 (four test_serialize methods in the Vertex AI triggers had been vacuously passing for years).

Changes

pyproject.toml: add B015 (useless-comparison) to extend-select. This rule fires on any bare comparison expression used as a statement.
16 test files: add the missing assert to all 26 existing violations found by the rule across airflow-core, task-sdk, and providers amazon, cncf/kubernetes, google).

With this in place, any future bare comparison in a test is a lint error caught before CI runs.

Was generative AI tooling used to co-author this PR?

Yes — Claude Code (Sonnet 4.6)

Generated-by: Claude Code (Sonnet 4.6) following the guidelines

Bare `actual == expected` expressions (missing `assert`) silently pass regardless of correctness. Enable ruff rule B015 (useless-comparison) to make this a lint error, and add `assert` to all 26 existing instances found across airflow-core, task-sdk, and providers (amazon, cncf, google).

`.calls` is not a real MagicMock attribute; accessing it auto-creates a child MagicMock. The expression `mock.calls[0] == []` was a silent no-op that B015 correctly caught. Adding `assert` exposed that the comparison always evaluates to False. Replace with `assert_not_called()`, which correctly verifies that `revoke_task` routes through `kube_scheduler.patch_pod_revoked()` rather than calling `patch_namespaced_pod` directly.

- test_bedrock: set ensure_unique_job_name on the operator before execute so the not-ensure-unique parametrize cases actually test the right path, and use pytest.raises for the cases where a conflict error is expected - test_emr_serverless (create): remove dead get_application mock setup and wrong call-count assertion; the operator uses waiters, not direct polling - test_emr_serverless (start job): correct expected call_count from 2 to 1; the operator calls get_application once (state CREATED is in SUCCESS_STATES so no retry loop)

jscheffl

Cool! I assume you can work on making CI green in parallel :-D

- Add missing `assert` to three bare comparisons caught by the new B015 rule: test_pod_template_file.py, test_otel_logger.py (×2) - Fix the MappedOperator task_type assertion in test_dag_serialization: `type(task).__name__` gives the wrapper class (`DecoratedMappedOperator`) but `operator_class["task_type"]` stores the underlying operator's name (`task.operator_class.__name__`, e.g. `_PythonDecoratedOperator`)

The operator uses json.dump() which writes strings incrementally to the file handle; the previous assertion expected bytes in a single write call. Also adds the missing mock_file.write.reset_mock() between the CSV and JSON sub-tests so write calls don't accumulate across sections.

shahar1 · 2026-05-15T04:59:10Z

Bare actual == expected expressions in tests silently evaluate and discard their result — the test passes regardless of correctness. This is exactly what happened in #66894 (four test_serialize methods in the Vertex AI triggers had been vacuously passing for years).

Changes

pyproject.toml: add B015 (useless-comparison) to extend-select. This rule fires on any bare comparison expression used as a statement.
16 test files: add the missing assert to all 26 existing violations found by the rule across airflow-core, task-sdk, and providers amazon, cncf/kubernetes, google).

With this in place, any future bare comparison in a test is a lint error caught before CI runs.

Was generative AI tooling used to co-author this PR?

Yes — Claude Code (Sonnet 4.6)

Generated-by: Claude Code (Sonnet 4.6) following the guidelines

Bare `actual == expected` expressions (missing `assert`) silently pass regardless of correctness. Enable ruff rule B015 (useless-comparison) to make this a lint error, and add `assert` to all 26 existing instances found across airflow-core, task-sdk, and providers (amazon, cncf, google).

`.calls` is not a real MagicMock attribute; accessing it auto-creates a child MagicMock. The expression `mock.calls[0] == []` was a silent no-op that B015 correctly caught. Adding `assert` exposed that the comparison always evaluates to False. Replace with `assert_not_called()`, which correctly verifies that `revoke_task` routes through `kube_scheduler.patch_pod_revoked()` rather than calling `patch_namespaced_pod` directly.

- test_bedrock: set ensure_unique_job_name on the operator before execute so the not-ensure-unique parametrize cases actually test the right path, and use pytest.raises for the cases where a conflict error is expected - test_emr_serverless (create): remove dead get_application mock setup and wrong call-count assertion; the operator uses waiters, not direct polling - test_emr_serverless (start job): correct expected call_count from 2 to 1; the operator calls get_application once (state CREATED is in SUCCESS_STATES so no retry loop)

jscheffl

Cool! I assume you can work on making CI green in parallel :-D

- Add missing `assert` to three bare comparisons caught by the new B015 rule: test_pod_template_file.py, test_otel_logger.py (×2) - Fix the MappedOperator task_type assertion in test_dag_serialization: `type(task).__name__` gives the wrapper class (`DecoratedMappedOperator`) but `operator_class["task_type"]` stores the underlying operator's name (`task.operator_class.__name__`, e.g. `_PythonDecoratedOperator`)

The operator uses json.dump() which writes strings incrementally to the file handle; the previous assertion expected bytes in a single write call. Also adds the missing mock_file.write.reset_mock() between the CSV and JSON sub-tests so write calls don't accumulate across sections.

shahar1 · 2026-05-15T04:59:10Z

Bare actual == expected expressions in tests silently evaluate and discard their result — the test passes regardless of correctness. This is exactly what happened in #66894 (four test_serialize methods in the Vertex AI triggers had been vacuously passing for years).

Changes

pyproject.toml: add B015 (useless-comparison) to extend-select. This rule fires on any bare comparison expression used as a statement.
16 test files: add the missing assert to all 26 existing violations found by the rule across airflow-core, task-sdk, and providers amazon, cncf/kubernetes, google).

With this in place, any future bare comparison in a test is a lint error caught before CI runs.

Was generative AI tooling used to co-author this PR?

Yes — Claude Code (Sonnet 4.6)

Generated-by: Claude Code (Sonnet 4.6) following the guidelines

Bare `actual == expected` expressions (missing `assert`) silently pass regardless of correctness. Enable ruff rule B015 (useless-comparison) to make this a lint error, and add `assert` to all 26 existing instances found across airflow-core, task-sdk, and providers (amazon, cncf, google).

`.calls` is not a real MagicMock attribute; accessing it auto-creates a child MagicMock. The expression `mock.calls[0] == []` was a silent no-op that B015 correctly caught. Adding `assert` exposed that the comparison always evaluates to False. Replace with `assert_not_called()`, which correctly verifies that `revoke_task` routes through `kube_scheduler.patch_pod_revoked()` rather than calling `patch_namespaced_pod` directly.

- test_bedrock: set ensure_unique_job_name on the operator before execute so the not-ensure-unique parametrize cases actually test the right path, and use pytest.raises for the cases where a conflict error is expected - test_emr_serverless (create): remove dead get_application mock setup and wrong call-count assertion; the operator uses waiters, not direct polling - test_emr_serverless (start job): correct expected call_count from 2 to 1; the operator calls get_application once (state CREATED is in SUCCESS_STATES so no retry loop)

jscheffl

Cool! I assume you can work on making CI green in parallel :-D

- Add missing `assert` to three bare comparisons caught by the new B015 rule: test_pod_template_file.py, test_otel_logger.py (×2) - Fix the MappedOperator task_type assertion in test_dag_serialization: `type(task).__name__` gives the wrapper class (`DecoratedMappedOperator`) but `operator_class["task_type"]` stores the underlying operator's name (`task.operator_class.__name__`, e.g. `_PythonDecoratedOperator`)

The operator uses json.dump() which writes strings incrementally to the file handle; the previous assertion expected bytes in a single write call. Also adds the missing mock_file.write.reset_mock() between the CSV and JSON sub-tests so write calls don't accumulate across sections.

shahar1 · 2026-05-15T04:59:10Z

Bare actual == expected expressions in tests silently evaluate and discard their result — the test passes regardless of correctness. This is exactly what happened in #66894 (four test_serialize methods in the Vertex AI triggers had been vacuously passing for years).

Changes

pyproject.toml: add B015 (useless-comparison) to extend-select. This rule fires on any bare comparison expression used as a statement.
16 test files: add the missing assert to all 26 existing violations found by the rule across airflow-core, task-sdk, and providers amazon, cncf/kubernetes, google).

With this in place, any future bare comparison in a test is a lint error caught before CI runs.

Was generative AI tooling used to co-author this PR?

Yes — Claude Code (Sonnet 4.6)

Generated-by: Claude Code (Sonnet 4.6) following the guidelines

Bare `actual == expected` expressions (missing `assert`) silently pass regardless of correctness. Enable ruff rule B015 (useless-comparison) to make this a lint error, and add `assert` to all 26 existing instances found across airflow-core, task-sdk, and providers (amazon, cncf, google).

`.calls` is not a real MagicMock attribute; accessing it auto-creates a child MagicMock. The expression `mock.calls[0] == []` was a silent no-op that B015 correctly caught. Adding `assert` exposed that the comparison always evaluates to False. Replace with `assert_not_called()`, which correctly verifies that `revoke_task` routes through `kube_scheduler.patch_pod_revoked()` rather than calling `patch_namespaced_pod` directly.

- test_bedrock: set ensure_unique_job_name on the operator before execute so the not-ensure-unique parametrize cases actually test the right path, and use pytest.raises for the cases where a conflict error is expected - test_emr_serverless (create): remove dead get_application mock setup and wrong call-count assertion; the operator uses waiters, not direct polling - test_emr_serverless (start job): correct expected call_count from 2 to 1; the operator calls get_application once (state CREATED is in SUCCESS_STATES so no retry loop)

jscheffl

Cool! I assume you can work on making CI green in parallel :-D

- Add missing `assert` to three bare comparisons caught by the new B015 rule: test_pod_template_file.py, test_otel_logger.py (×2) - Fix the MappedOperator task_type assertion in test_dag_serialization: `type(task).__name__` gives the wrapper class (`DecoratedMappedOperator`) but `operator_class["task_type"]` stores the underlying operator's name (`task.operator_class.__name__`, e.g. `_PythonDecoratedOperator`)

The operator uses json.dump() which writes strings incrementally to the file handle; the previous assertion expected bytes in a single write call. Also adds the missing mock_file.write.reset_mock() between the CSV and JSON sub-tests so write calls don't accumulate across sections.

shahar1 · 2026-05-15T04:59:10Z

Bare actual == expected expressions in tests silently evaluate and discard their result — the test passes regardless of correctness. This is exactly what happened in #66894 (four test_serialize methods in the Vertex AI triggers had been vacuously passing for years).

Changes

pyproject.toml: add B015 (useless-comparison) to extend-select. This rule fires on any bare comparison expression used as a statement.
16 test files: add the missing assert to all 26 existing violations found by the rule across airflow-core, task-sdk, and providers amazon, cncf/kubernetes, google).

With this in place, any future bare comparison in a test is a lint error caught before CI runs.

Was generative AI tooling used to co-author this PR?

Yes — Claude Code (Sonnet 4.6)

Generated-by: Claude Code (Sonnet 4.6) following the guidelines

Bare `actual == expected` expressions (missing `assert`) silently pass regardless of correctness. Enable ruff rule B015 (useless-comparison) to make this a lint error, and add `assert` to all 26 existing instances found across airflow-core, task-sdk, and providers (amazon, cncf, google).

`.calls` is not a real MagicMock attribute; accessing it auto-creates a child MagicMock. The expression `mock.calls[0] == []` was a silent no-op that B015 correctly caught. Adding `assert` exposed that the comparison always evaluates to False. Replace with `assert_not_called()`, which correctly verifies that `revoke_task` routes through `kube_scheduler.patch_pod_revoked()` rather than calling `patch_namespaced_pod` directly.

- test_bedrock: set ensure_unique_job_name on the operator before execute so the not-ensure-unique parametrize cases actually test the right path, and use pytest.raises for the cases where a conflict error is expected - test_emr_serverless (create): remove dead get_application mock setup and wrong call-count assertion; the operator uses waiters, not direct polling - test_emr_serverless (start job): correct expected call_count from 2 to 1; the operator calls get_application once (state CREATED is in SUCCESS_STATES so no retry loop)

jscheffl

Cool! I assume you can work on making CI green in parallel :-D

- Add missing `assert` to three bare comparisons caught by the new B015 rule: test_pod_template_file.py, test_otel_logger.py (×2) - Fix the MappedOperator task_type assertion in test_dag_serialization: `type(task).__name__` gives the wrapper class (`DecoratedMappedOperator`) but `operator_class["task_type"]` stores the underlying operator's name (`task.operator_class.__name__`, e.g. `_PythonDecoratedOperator`)

The operator uses json.dump() which writes strings incrementally to the file handle; the previous assertion expected bytes in a single write call. Also adds the missing mock_file.write.reset_mock() between the CSV and JSON sub-tests so write calls don't accumulate across sections.

shahar1 requested review from amoghrajesh, ashb, hussein-awala, jedcunningham, jscheffl, kaxil and o-nikolas as code owners May 15, 2026 04:59

shahar1 requested a review from vatsrahul1001 May 15, 2026 04:59

shahar1 force-pushed the enable-ruff-b015-useless-comparison branch from ff2e36f to 7dfec04Compare May 15, 2026 05:01

shahar1 requested review from Lee-W and jason810496 May 15, 2026 05:20

shahar1 marked this pull request as draft May 15, 2026 05:36

shahar1 marked this pull request as ready for review May 15, 2026 05:50

shahar1 marked this pull request as draft May 15, 2026 06:29

shahar1 marked this pull request as ready for review May 15, 2026 06:37

Lee-W reviewed May 15, 2026
View reviewed changes

Comment threadproviders/amazon/tests/unit/amazon/aws/operators/test_emr_serverless.py

jscheffl approved these changes May 15, 2026
View reviewed changes

shahar1 requested a review from potiuk as a code owner May 15, 2026 15:55

shahar1 requested a review from bugraoz93 as a code owner May 15, 2026 15:55

shahar1 merged commit 2ed6805 into apache:mainMay 16, 2026
292 checks passed

shahar1 deleted the enable-ruff-b015-useless-comparison branch May 16, 2026 05:24

dkranchii mentioned this pull request Jul 28, 2026
Enable ruff B023 (function-uses-loop-variable) and fix violations #70640
Merged
1 task