fix: close Claude stream when final usage is estimated

[mercy719]

📝 变更描述 / Description

修复 OpenAI-compatible stream 转 Anthropic /v1/messages 时的结束事件丢失问题。

此前当上游最后一个 chunk 只有 finish_reason: "stop"、没有 usage 时，转换器会提前返回等待 usage-only chunk。对于已经在最终处理阶段写入本地估算 usage 的请求，这会跳过 content_block_stop、message_delta 和 message_stop，导致 Claude Code 等客户端一直等待流结束。

本次改动只在当前 chunk 和 ClaudeConvertInfo.Usage 都没有 usage 时继续等待；如果已有估算 usage，则正常关闭当前 content block 并输出 Anthropic 结束事件。

🚀 变更类型 / Type of change

• 🐛 Bug 修复 (Bug fix) - 请关联对应 Issue，避免将设计取舍、理解偏差或预期不一致直接归类为 bug
• ✨ 新功能 (New feature) - 重大特性建议先通过 Issue 沟通
• ⚡ 性能优化 / 重构 (Refactor)
• 📝 文档更新 (Documentation)

🔗 关联任务 / Related Issue

• Closes Bug: /v1/messages streaming for qwen3.6-plus lacks content_block_stop/message_delta/message_stop, causing Claude Code to hang #4697

✅ 提交前检查项 / Checklist

• 人工确认: 我已亲自整理并撰写此描述，没有直接粘贴未经处理的 AI 输出。
• 非重复提交: 我已搜索现有的 Issues 与 PRs，确认不是重复提交。
• Bug fix 说明: 若此 PR 标记为 Bug fix，我已提交或关联对应 Issue，且不会将设计取舍、预期不一致或理解偏差直接归类为 bug。
• 变更理解: 我已理解这些更改的工作原理及可能影响。
• 范围聚焦: 本 PR 未包含任何与当前任务无关的代码改动。
• 本地验证: 已在本地运行并通过测试或手动验证，维护者可以据此复核结果。
• 安全合规: 代码中无敏感凭据，且符合项目代码规范。

📸 运行证明 / Proof of Work

docker run --rm -v /Users/yimingshi/Documents/new-api:/workspace -w /workspace golang:1.25 go test ./service -run TestStreamResponseOpenAI2ClaudeClosesStopChunkWithoutUpstreamUsage -count=1
docker run --rm -v /Users/yimingshi/Documents/new-api:/workspace -w /workspace golang:1.25 go test ./service -count=1
docker run --rm -v /Users/yimingshi/Documents/new-api:/workspace -w /workspace golang:1.25 go test ./relay/channel/openai -run TestNonexistent -count=1

Summary by CodeRabbit

• Bug Fixes

  • Prevented incorrect completion/usage emission when upstream usage data is missing, ensuring stop reasons and final message signals are handled correctly during stream conversion.

• Tests

  • Added unit tests covering streaming behavior across reasoning, content, and finish-reason scenarios to validate correct stop-reason mapping, usage propagation, and finalization.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 089fc861-fda9-411d-8118-7930127509fc

📥 Commits

Reviewing files that changed from the base of the PR and between 29d2466 and acb5574.

📒 Files selected for processing (2)

• service/convert.go
• service/convert_test.go

🚧 Files skipped from review as they are similar to previous changes (2)

• service/convert_test.go
• service/convert.go

---

Walkthrough

Adds a nil-usage guard in StreamResponseOpenAI2Claude's done-chunk path to avoid emitting a final message_delta when no usage exists, and adds a unit test that verifies chunk ordering, stop-reason mapping, usage propagation, and Claude conversion completion state.

Changes

Stream Response Completion Handling

Layer / File(s)

Summary

Guard for nil usage in done-chunk handling and validation test service/convert.go, service/convert_test.go

Adds a conditional in the doneChunk path of StreamResponseOpenAI2Claude that detects when both openAIResponse.Usage and info.ClaudeConvertInfo.Usage are nil and prevents emitting the final message_delta/stop-reason. Adds TestStreamResponseOpenAI2ClaudeClosesStopChunkWithoutUpstreamUsage verifying chunk ordering (reasoning/content/stop), index expectations, mapping of FinishReasonStop → "end_turn", propagation of usage from ClaudeConvertInfo, and setting info.ClaudeConvertInfo.Done=true on completion.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• QuantumNous/new-api#2887: Addresses nil-usage and patching missing usage fields before downstream message_delta processing.
• QuantumNous/new-api#4090: Modifies the same StreamResponseOpenAI2Claude done-chunk handling and adds tests for missing upstream usage.
• QuantumNous/new-api#1522: Adjusts finish-stop handling and stop-reason propagation in StreamResponseOpenAI2Claude.

Poem

🐰 I bounced through streams to find the end,
Guarded the gap where usage might bend,
Mapped stop to "end_turn" with a careful hop,
Now chunks close cleanly — no more drop-stop! ✨

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately summarizes the main change: fixing the stream closure by handling estimated usage in the OpenAI-to-Claude conversion logic when final chunks lack upstream usage data.
Linked Issues check	✅ Passed	The PR implements the solution to issue #4697 by adding a guard that ensures the converter continues processing and emits required Anthropic SSE end events (content_block_stop, message_delta, message_stop) when estimated usage exists, preventing clients from hanging.
Out of Scope Changes check	✅ Passed	All changes are directly scoped to fixing the identified issue: the production code change adds a usage check guard in the conversion logic, and the test case validates the specific scenario of closing a stop chunk without upstream usage.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}