Fix LangGraph content-block serialization in gen_ai.output.messages (#189)

[JacksonWeber]

Summary

Fixes #189.

LangChain/LangGraph AIMessage.content may be either a plain string or a list of content-block dicts (e.g. [{"type": "text", "text": "...", "phase": "final_answer", "id": "msg_..."}]). The previous _langchain_content helper called str(c) on this value, producing a Python-repr blob with single quotes and leaked phase/index/id keys inside what the GenAI semconv requires to be a plain TextPart.content string. This broke Foundry's cloud trace evaluators (builtin.coherence, builtin.fluency, etc.) which read the assistant text out of gen_ai.output.messages[*].parts[*].content.

Verification

Serialized a real LangGraph-shaped output and validated against the upstream gen-ai-output-messages.json schema:

[
  {
    "role": "assistant",
    "parts": [
      { "type": "text", "content": "# One-Day Food Walk in Vancouver\n## Assumptions" },
      { "type": "tool_call", "id": "tool_1", "name": "search", "arguments": "{\"q\":\"food\"}" }
    ],
    "finish_reason": "stop"
  }
]

All spec constraints satisfied: role ∈ allowed enum, TextPart.content is a plain string with no repr blob or leaked metadata, ToolCallRequestPart has const type: "tool_call" + required name, finish_reason in allowed enum.

tests/langchain/test_utils.py: 128/128 pass (was 125; +3 new).

[github-actions]

Performance comparison

Threshold: regressions >15.0% on gating scenarios fail the build. Higher ops/s is better; positive Δ means the PR is slower.

Scenario	Gating	Baseline (ops/s)	Candidate (ops/s)	Δ %	Status
azure_monitor_log	yes	25,905.4	26,300.6	-1.50%	✅
azure_monitor_span	yes	154,035.7	155,472.6	-0.92%	✅
otel_log	no	31,259.8	32,460.2	-3.70%	✅
otel_span	no	34,042.6	34,112.2	-0.20%	✅

[Copilot]

Pull request overview

This PR fixes LangChain/LangGraph AIMessage.content serialization for GenAI semantic conventions by normalizing list-shaped “content blocks” into spec-compliant output message parts, preventing Python repr blobs from being emitted in gen_ai.output.messages.

Changes:

• Add _flatten_lc_content_blocks() to extract/concatenate text from LangGraph-style content-block lists for TextPart.content.
• Extend tool-call extraction to also harvest tool_use blocks embedded inside list-shaped content.
• Add regression tests covering content-block flattening, multi-block joining, and embedded tool_use handling; document the fix in CHANGELOG.md.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
tests/langchain/test_utils.py	Adds regression tests to ensure LangGraph content blocks serialize to spec-compliant text/tool-call parts.
src/microsoft/opentelemetry/_genai/_langchain/_utils.py	Implements content-block flattening and harvesting of embedded tool_use blocks into tool-call parts.
CHANGELOG.md	Notes the fix under “Unreleased”.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated no new comments.

[rads-1996]

LGTM