The parametrize decorator on line 170 uses [[False, True], [False, True]] which creates two identical test cases with span_streaming=False, send_default_pii=True. This means the new span streaming code path (lines 198-222) will never be executed. The intended parameterization should test all combinations: [[False, False], [False, True], [True, False], [True, True]] or at least include span_streaming=True cases.

_get_request_attributes iterates over headers from _filter_headers and directly assigns values to attributes. When should_send_default_pii() returns False, _filter_headers returns AnnotatedValue objects for sensitive headers (cookies, authorization, etc.). These objects are passed to set_attribute(), which calls format_attribute() -> safe_repr(), resulting in unhelpful string representations like '<AnnotatedValue object>' instead of proper redaction handling. While no actual PII is leaked (AnnotatedValue.value is empty), the behavior is inconsistent with _get_request_data and produces meaningless attribute values.

The code sets network.protocol.name as the attribute key, but the test test_capture_transaction in tests/integrations/asgi/test_asgi.py line 217 expects the attribute to be http.request.protocol.name. This will cause the test to fail when span streaming is enabled, as the test assertion span["attributes"]["http.request.protocol.name"] == "http" will not find the expected key.

The envelopes_to_spans function directly accesses span_json["attributes"] without checking if the key exists. In _span_batcher.py line 116, the attributes field is only included in the transport format when item._attributes is truthy (non-empty). If a span has no attributes set, span_json["attributes"] will not exist, causing a KeyError when the function tries to iterate over it.