fix(planning): handle empty LLM response in macro plan parsing (#162)#171
fix(planning): handle empty LLM response in macro plan parsing (#162)#171liaoyl830 wants to merge 1 commit into
Conversation
…inglinyi#162) When using glm-4.5-air or other models that occasionally return empty responses, _parse_llm_response would throw a cryptic JSONDecodeError instead of a clear error message. Add a check for empty cleaned content before attempting json.loads(), raising a descriptive ValueError that autopilot_daemon can handle gracefully via its existing error handling.
📝 WalkthroughWalkthroughThis PR adds defensive error handling and improves EPUB export robustness. LLM response parsing now validates and fails early with a user-friendly error when content is empty. EPUB export adds diagnostic logging for missing chapter content and explicitly generates proper EPUB 3 navigation documents. ChangesError Handling and Export Improvements
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes Possibly related PRs
Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
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. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
🧹 Nitpick comments (1)
application/core/services/export_service.py (1)
200-203: ⚡ Quick winInclude nav in spine for EPUB 3 reader compatibility.
At Line 203,
EpubNav()is added, but the spine still excludes"nav". Some readers rely on nav being in reading order; adding it to spine improves compatibility.Proposed change
- book.spine = spine_items + book.spine = ["nav", *spine_items]🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@application/core/services/export_service.py` around lines 200 - 203, The EPUB spine currently assigns book.spine = spine_items after adding book.add_item(epub.EpubNav()), but it never includes the "nav" item id in the spine; modify the spine construction so the nav id is included in the reading order (e.g., insert or append "nav" into the spine_items list before assigning book.spine) so the epub.EpubNav() item added by book.add_item(epub.EpubNav()) is present in the spine for EPUB 3 compatibility.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Nitpick comments:
In `@application/core/services/export_service.py`:
- Around line 200-203: The EPUB spine currently assigns book.spine = spine_items
after adding book.add_item(epub.EpubNav()), but it never includes the "nav" item
id in the spine; modify the spine construction so the nav id is included in the
reading order (e.g., insert or append "nav" into the spine_items list before
assigning book.spine) so the epub.EpubNav() item added by
book.add_item(epub.EpubNav()) is present in the spine for EPUB 3 compatibility.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro Plus
Run ID: f5b19975-a32b-4796-90ec-19ad7b0969cd
📒 Files selected for processing (2)
application/blueprint/services/continuous_planning_service.pyapplication/core/services/export_service.py
|
感谢修复，这个问题当时确实会让宏观规划在模型空响应时直接掉进 JSONDecodeError，体验很差。 这块后续主线已经加入了空响应判断、输出清洗和 JSON repair 兜底，并且相关规划流程也有更大调整。当前 PR 基于旧上下文，继续合并会有比较高的冲突和重复实现风险，所以我先关闭。 如果后面还想继续参与，欢迎联系我加入 dev 仓库。我们可以先在 dev 分支/仓库里把具体模型兼容问题修稳、跑通，再合进开源 master。感谢你帮忙补这个坑。 |
|
再补充说明一下：这次关闭 PR 只是基于当前 master 稳定性、冲突风险和合并策略做的整理，不代表否定大家的贡献。 大家已经投入的工作我都看到了，也会统一计入项目贡献者名单/贡献记录里。很多思路后续仍然会被吸收，只是现在为了尽量避免开源 master 出问题，我会更倾向于先在 dev 仓库里拆小块开发、验证、跑通，再合并到 master。 如果后续还愿意一起推进，欢迎直接联系我，我可以开放 dev 仓库权限。我们按模块拆 PR，一块一块稳着来。再次感谢大家愿意在这么早期、不成熟的阶段帮忙共建，真的感谢。 |
2 participants
变更类型
fixBug 修复变更说明
修复使用 glm-4.5-air 等模型时，LLM 偶尔返回空响应导致的
JSONDecodeError: Expecting value: line 1 column 1 (char 0)错误。根因
_parse_llm_response()方法在清理 LLM 返回内容后，直接调用json.loads()。当模型返回空字符串时（可能由于 API 限流、网络问题或模型自身问题），json.loads("")会抛出难以理解的 JSONDecodeError。修复
在
json.loads()之前增加空内容检查，返回描述性ValueError，便于autopilot_daemon的现有错误处理逻辑识别并优雅降级。架构影响
applicationSummary by CodeRabbit