fix(qwen-agent): accept tuple inputs for messages and content by somuai · Pull Request #913 · open-telemetry/opentelemetry-python-genai

somuai · 2026-10-17T20:57:53Z

Description

This pull request resolves #904 by updating the qwen-agent instrumentation conversion helpers to recognize and handle messages, content items, and tool call histories provided as tuples or general Sequences.

Previously, convert_to_input_messages, convert_to_output_messages, and convert_to_final_output_messages checked if not isinstance(messages, list): messages = [messages]. When messages were provided as a tuple, the entire tuple was wrapped as a single element ([(m1, m2)]), causing property extraction to fail and dropping or corrupting telemetry. Similarly, _extract_content_text only inspected isinstance(content, list), and find_tool_call_id only checked isinstance(messages, list).

Changes

utils.py:
- Added from collections.abc import Sequence.
- Updated _extract_content_text to check isinstance(content, (list, tuple)) and annotated with Sequence[QwenContentItem].
- Updated find_tool_call_id to check isinstance(messages, (list, tuple)) and annotated with Sequence[QwenMessage].
- Updated convert_to_input_messages, convert_to_output_messages, and convert_to_final_output_messages to check not isinstance(messages, (list, tuple)) before wrapping, and updated type signatures to Sequence[QwenMessage].
- Updated create_agent_invocation messages type signature to Sequence[QwenMessage].
Tests (test_agent.py):
- Added test_convert_to_input_messages_handles_tuples verifying tuple messages and tuple content items.
- Added test_convert_to_output_messages_handles_tuples verifying tuple output messages and tuple content items.
- Added test_find_tool_call_id_handles_tuples verifying function id resolution over tuple message histories.
- Added test_agent_run_with_tuple_messages verifying end-to-end Agent.run() span generation when messages and yielded responses are provided as tuples.
Changelog: Added instrumentation/opentelemetry-instrumentation-genai-qwen-agent/.changelog/904.fixed.

Testing

Unit tests: All 28 tests in instrumentation/opentelemetry-instrumentation-genai-qwen-agent/tests/ passed cleanly (27 passed, 1 skipped conformance).
Lint and formatting: 100% clean under uv run --with "ruff==0.16.1" ruff check and uv run --with "ruff==0.16.1" ruff format --check.

Fixes #904

opentelemetry-pr-dashboard · 2026-10-17T20:58:42Z

Pull request dashboard status

Merged · refreshed 2026-10-22 03:17 UTC

Status above doesn't look right?

Anything look wrong? Report it with what you expected; it helps us improve the dashboard.

Copilot

🟡 Changes recommended

General sequences and dictionary-based tuple content still produce missing or incorrect telemetry.

2 open issues

What changed in this PR

Updates qwen-agent telemetry conversion to support tuple-based messages and content for #904.

Changes:

Adds tuple handling and widens annotations to Sequence.
Adds helper and agent-run regression tests.
Records the fix in the package changelog.

File	Description
instrumentation/opentelemetry-instrumentation-genai-qwen-agent/tests/test_agent.py	Adds tuple-input regression tests.
instrumentation/opentelemetry-instrumentation-genai-qwen-agent/src/opentelemetry/instrumentation/genai/qwen_agent/utils.py	Extends message, content, and tool-history handling.
instrumentation/opentelemetry-instrumentation-genai-qwen-agent/.changelog/904.fixed	Documents tuple capture support.

_{Review effort: Balanced}

somuai · 2026-10-17T21:18:33Z

+    if isinstance(content, (list, tuple)):
        texts: list[str] = []
        for item in content:
            text = getattr(item, "text", None)


Resolved in commit 0efdb34: updated _extract_content_text to read text with _field_value(item, "text") and convert to str, ensuring raw dictionary content items like ({"text": "Hello"},) are extracted properly. Added regression test test_convert_to_input_messages_handles_raw_dict_tuple_content.

somuai · 2026-10-17T21:18:41Z

        return []

-    if not isinstance(messages, list):
+    if not isinstance(messages, (list, tuple)):


Resolved in commit 0efdb34: defined _is_sequence helper recognizing general Sequence types while excluding scalar strings, bytes, and mappings. Applied _is_sequence across _extract_content_text, find_tool_call_id, convert_to_input_messages, convert_to_output_messages, and convert_to_final_output_messages. Added regression tests test_conversion_helpers_handle_user_list_sequences and test_agent_run_with_user_list_messages using collections.UserList.

eternalcuriouslearner · 2026-10-19T01:32:35Z


 def convert_to_output_messages(
-    messages: QwenMessage | list[QwenMessage] | None,
+    messages: QwenMessage | Sequence[QwenMessage] | None,


question(blocking): Could we scope this to tuple inputs? Agent.run() rebuilds each output as a list, so is tuple support needed in the output converters? Similarly, since it normalizes inputs into a list before calling _run(), is there a path where find_tool_call_id() receives a tuple history?

Updated in commit acfe76f: scoped tuple inputs strictly to convert_to_input_messages and message content extraction, while keeping convert_to_output_messages, convert_to_final_output_messages, and find_tool_call_id scoped to standard lists as outputs and histories are normalized by Agent.run().

eternalcuriouslearner · 2026-10-19T01:32:52Z


 def find_tool_call_id(
-    messages: list[QwenMessage] | None, tool_name: str
+    messages: Sequence[QwenMessage] | None, tool_name: str


question(blocking): Could we scope this to tuple inputs? Agent.run() rebuilds each output as a list, so is tuple support needed in the output converters? Similarly, since it normalizes inputs into a list before calling _run(), is there a path where find_tool_call_id() receives a tuple history?

eternalcuriouslearner · 2026-10-19T01:34:39Z



+def _is_sequence(val: object) -> bool:
+    """Check if a value is a non-string, non-mapping sequence."""


question(blocking): Could we fix the type narrowing? _is_sequence() -> bool doesn’t narrow the callers’ union types, causing pyright errors. Use TypeGuard[Sequence[Any]] for positive checks, and normalize converter inputs separately because TypeGuard doesn’t narrow false branches:

def _as_message_sequence( messages: QwenMessage | Sequence[QwenMessage] | None, ) -> Sequence[QwenMessage]: if messages is None: return () if isinstance(messages, Sequence): return messages return (messages,)

Then iterate over _as_message_sequence(messages) instead of reassigning messages. Could you verify with uv run tox -e typecheck?

Resolved in commit acfe76f: annotated _is_sequence with TypeGuard[Sequence[Any]] and introduced _as_message_sequence to normalize QwenMessage | Sequence[QwenMessage] | None inputs without false-branch narrowing issues. Verified with pyright across the package with zero errors.

eternalcuriouslearner · 2026-10-19T01:36:08Z

+            text = _field_value(item, "text")
            if text is not None:
-                texts.append(text)
+                texts.append(str(text))


question(non-blocking): Is this needed?

Kept str(text) as defensive conversion in commit acfe76f so non-string text values from raw dictionary items are handled consistently without raising in str.join, verified with pyright with zero errors.

eternalcuriouslearner · 2026-10-19T01:36:51Z

@@ -0,0 +1 @@
+Capture qwen-agent messages, content parts, and tool calls provided as tuples.


can you please rename the file to 913 instead of 904?

Renamed to instrumentation/opentelemetry-instrumentation-genai-qwen-agent/.changelog/913.fixed in commit acfe76f.

eternalcuriouslearner · 2026-10-19T01:38:08Z

    assert child_spans[0].parent.span_id == parent_spans[0].context.span_id
+
+
+def test_convert_to_input_messages_handles_tuples():


question(blocking): Could we make the integration tests pass tuple inputs to Agent.run(), have _run() yield normal lists, and assert both captured input and output? For tuple/UserList content coverage, could we use raw dictionary messages so the content reaches instrumentation unchanged, since Message(...) converts those containers to lists during validation?

Updated in commit acfe76f: integration tests now pass tuple and UserList inputs with raw dictionary messages to Agent.run(), ensuring container shapes reach instrumentation intact without being converted by Message(...) validation. _run() yields standard lists, and assertions verify both captured input messages and final output messages.

somuai · 2026-10-19T03:11:08Z

All review feedback from @eternalcuriouslearner has been addressed in commit acfe76f:

Scoped tuple inputs to input messages and content extraction, while keeping output converters and tool call lookup scoped to standard lists.
Implemented _as_message_sequence and annotated _is_sequence with TypeGuard[Sequence[Any]], achieving zero typecheck errors with pyright.
Renamed changelog entry to .changelog/913.fixed.
Updated integration tests to pass tuple and UserList inputs with raw dictionary messages so content containers reach instrumentation intact, with _run() yielding normal lists.
All 28 unit and integration tests, ruff check, and pyright pass cleanly.

123liuziming · 2026-10-20T08:11:00Z

+    """Normalize input messages into a sequence without copying."""
+    if messages is None:
+        return ()
+    if isinstance(messages, Sequence) and not isinstance(


Should we use if _is_sequence(messages): directly here?

Good catch, updated to use _is_sequence(messages) directly in commit 766a21e.

Update qwen-agent conversion helpers to accept tuples for messages and content parts. Fixes open-telemetry#904 Signed-off-by: Soumyajit Ghosh <jobsoumyajit6124@gmail.com>

Address review feedback: - Use _is_sequence across conversion helpers to support general Sequence types like UserList. - Use _field_value in _extract_content_text to read text from dictionary content items. - Add regression tests covering dict content items and UserList sequences. Signed-off-by: Soumyajit Ghosh <jobsoumyajit6124@gmail.com>

@eternalcuriouslearner

…name changelog to 913 Address review feedback from @eternalcuriouslearner: - Implement _as_message_sequence helper and TypeGuard[Sequence[Any]] on _is_sequence to resolve pyright union narrowing errors cleanly. - Scope tuple inputs to input messages and content, keeping output converters and tool call lookup scoped to standard lists. - Update integration and unit tests to pass tuple/UserList inputs with raw dictionary messages so content reaches instrumentation unchanged, with _run() yielding normal lists. - Rename changelog entry from 904.fixed to 913.fixed. Signed-off-by: Soumyajit Ghosh <jobsoumyajit6124@gmail.com>

Signed-off-by: Soumyajit Ghosh <jobsoumyajit6124@gmail.com>

somuai · 2026-10-21T13:58:14Z

/dashboard route:reviewers

opentelemetry-pr-dashboard · 2026-10-21T14:00:46Z

@somuai, your reviewer-routing request was recorded; this pull request has the approvals it needs and is now waiting on maintainers. The handoff remains active across pushes until newer actionable human feedback arrives. Top-level feedback through this request will not return; unresolved review threads remain open.

addressed

somuai requested a review from a team as a code owner October 17, 2026 20:57

Copilot AI balanced review requested due to automatic review settings October 17, 2026 20:57

github-project-automation Bot added this to GenAI Semantic Conventions and Instrumentation libraries Oct 17, 2026

Copilot started reviewing on behalf of somuai October 17, 2026 20:58

View session

opentelemetry-pr-dashboard Bot mentioned this pull request Oct 17, 2026

Pull Request Dashboard #171

Open

Copilot AI reviewed Oct 17, 2026

View reviewed changes

eternalcuriouslearner previously requested changes Oct 19, 2026

View reviewed changes

123liuziming reviewed Oct 20, 2026

View reviewed changes

lmolkova approved these changes Oct 21, 2026

View reviewed changes

123liuziming approved these changes Oct 21, 2026

View reviewed changes

somuai added 4 commits October 21, 2026 19:24

fix(qwen-agent): accept tuple inputs for messages and content

b521a45

Update qwen-agent conversion helpers to accept tuples for messages and content parts. Fixes open-telemetry#904 Signed-off-by: Soumyajit Ghosh <jobsoumyajit6124@gmail.com>

refactor(qwen-agent): use _is_sequence helper in _as_message_sequence

07a6e45

Signed-off-by: Soumyajit Ghosh <jobsoumyajit6124@gmail.com>

somuai force-pushed the fix/qwen-agent-tuple-inputs branch from 27fa2d3 to 07a6e45 Compare October 21, 2026 14:02

lmolkova enabled auto-merge October 22, 2026 01:36

lmolkova added this pull request to the merge queue Oct 22, 2026

Merged via the queue into open-telemetry:main with commit 9b8af3a Oct 22, 2026
82 checks passed

github-project-automation Bot moved this to Done in GenAI Semantic Conventions and Instrumentation libraries Oct 22, 2026



		def _is_sequence(val: object) -> bool:
		"""Check if a value is a non-string, non-mapping sequence."""

		@@ -0,0 +1 @@
		Capture qwen-agent messages, content parts, and tool calls provided as tuples.

		assert child_spans[0].parent.span_id == parent_spans[0].context.span_id


		def test_convert_to_input_messages_handles_tuples():

Conversation

somuai commented Oct 17, 2026

Description

Changes

Testing

Uh oh!

opentelemetry-pr-dashboard Bot commented Oct 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Pull request dashboard status

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

somuai commented Oct 19, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

somuai commented Oct 21, 2026

Uh oh!

opentelemetry-pr-dashboard Bot commented Oct 21, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

opentelemetry-pr-dashboard Bot commented Oct 17, 2026 •

edited

Loading