The test function test_async_transport_concurrent_requests is missing @skip_under_gevent, @pytest.mark.asyncio, and @pytest.mark.skipif(not PY38, ...) decorators that all other async transport tests in this file have. Without @pytest.mark.asyncio, the test will fail or be skipped by pytest-asyncio. Without @skip_under_gevent, the test will run under gevent where async tests are incompatible. Without the Python version skip, the test will fail on Python < 3.8.

The close() method uses warnings.warn() with message 'close() used with AsyncHttpTransport. Use close_async() instead.' but the corresponding test test_close_with_async_transport_warns expects logger.warning() with a different message. This will cause the test to fail. Either change to logger.warning() to match the test expectation, or update the test to use pytest.warns() to capture warnings.

Also found at:

• sentry_sdk/client.py:1095-1100
• tests/test_client.py:1669-1674
• tests/test_client.py:1737-1741

The test test_loop_close_flushes_async_transport creates mock_transport = Mock(spec=AsyncHttpTransport), but the _flush() function in patch_loop_close() checks isinstance(client.transport, AsyncHttpTransport) which will return False for a Mock object, even with spec=. This means _flush() returns early without calling close_async(). The test assertions should fail, or if they pass, the test is not actually verifying the flush-on-close behavior it intends to test.

Also found at:

• tests/test_transport.py:1058

The common: httpcore[asyncio] line at 91 adds the httpcore asyncio dependency for all Python versions, but httpcore 1.x requires Python >=3.8. Since the common test environment includes py3.6 and py3.7 (as defined on line 21), tox will fail to install dependencies for those environments. The dependency should be version-gated like other packages in this file (e.g., {py3.8,py3.9,py3.10,py3.11,py3.12,py3.13,py3.14,py3.14t}-common: httpcore[asyncio]).

Also found at:

• sentry_sdk/client.py:1057

The lambda on line 284 captures queue_ref by reference, not by value. Since queue_ref is reassigned in each loop iteration, all lambdas will see the most recent value of queue_ref when they execute, not the value at the time they were created. If kill()/start() replaces self._queue while tasks are pending, the callbacks will use the wrong queue, leading to calling task_done() on the wrong queue. This directly contradicts the comment on lines 281-282.

The test mocks sentry_sdk.client.logger and asserts mock_logger.warning.assert_called_with(...), but the actual close() implementation at line 1049 uses warnings.warn() instead of logger.warning(). Additionally, the expected message differs from the actual message in the implementation. This test will always pass the assertion incorrectly because the mocked logger is never called.

Also found at:

• tests/test_client.py:1737-1741
• sentry_sdk/client.py:1049-1052

The test function test_async_transport_concurrent_requests is missing the request fixture parameter in its function signature. This parameter is needed to add a finalizer for proper cleanup of the global scope client, as is done in other similar async tests in this file (test_transport_works_async, test_async_transport_rate_limiting_with_concurrency).

Also found at:

• tests/test_transport.py:1067

The test sets the global client via sentry_sdk.get_global_scope().set_client(client) but never resets it to None after the test completes. All other similar tests in this file use either request.addfinalizer() or a try/finally block to reset the client. This can cause test pollution affecting subsequent tests.

The helper function _make_async_transport_options() is defined at line 55 but is never called anywhere in the test file or the entire test suite. This appears to be dead code that was added but never integrated into any test. The async transport tests (e.g., test_transport_works_async) construct their options inline rather than using this helper.