ISSUE DESCRIPTION Bug: retainrecentlegalsuffix 在 else 分支下会导致用户消息被重复归档 Bug Description 在…
- User
ISSUE DESCRIPTION
Bug: retain_recent_legal_suffix 在 else 分支下会导致用户消息被重复归档
Bug Description
在
nanobot/session/manager.py的retain_recent_legal_suffix方法中,当 else 分支(最近 N 条消息全是助手/工具消息)被触发时,由于cut值的计算使用len(tail) - len(kept),而没有考虑用户消息的实际位置,导致用户消息被错误地同时归档到 archive 和 kept 中。具体影响:
- 用户消息同时出现在 archive(待压缩)和 kept(保留)中
- 可能导致 LLM 上下文不一致
Steps to Reproduce
- 会话中消息数量超过 _RECENT_SUFFIX_MESSAGES (默认10条)
- 最近 N 条消息中 不包含任何用户消息 (全是 assistant/tool 角色)
- 在完整会话历史中,最后一条用户消息位于倒数第 11-15 条之间
Expected Behavior
当会话历史中最近 N 条消息全是助手/工具消息,而用户消息位于更早位置时:
- 用户消息不应被重复归档 - 用户消息应该只保留在 kept 中,不应该同时出现在 archive 和 kept 中
- 归档边界应该正确 - cut 值应该基于用户消息的实际位置,而不是简单的 len(tail) - len(kept)
- 上下文连贯性应该保持 - 从用户消息开始的完整上下文应该被保留,不应该被错误切割
假设:
tail = messages[0:100](100条)
latest_user = 85(用户消息位置)
max_messages = 10
期望:
cut = latest_user # 应该是 85,不是 90 archive = tail[:85] # [0-84],不包含用户消息 kept = tail[85:] # [85-99],包含用户消息及其后续内容
结果:
archive 和 kept 没有重叠 ✅
用户消息只在 kept 中 ✅
上下文连贯 ✅
总结
期望行为 :无论用户消息位于何处,它都应该:
- 只出现在 kept 中(保留消息)
- 不出现在 archive 中(归档消息)
- 作为保留消息的起始点,确保上下文连贯
Relevant Logs
nanobot Version
0.2.0
Python Version
3.11
Operating System
Windows
Channel / Platform
Weixin (Personal WeChat)
LLM Provider
OpenAI
Configuration (Optional)
Additional Context
No response
Interface
Type: Function Name: retain_recent_legal_suffix Path: nanobot/session/manager.py Input: self, max_messages: int Output: tuple[list[dict], int] Description: Trims the session's message list to at most
max_messageswhile preserving a legal suffix (ending with a user message if possible). Returns the list of dropped messages and a count of how many of those were already in the consolidated prefix, so callers can archive only new, unconsolidated messages.ADDITIONAL ADVICE
Since you are given a git repository, you can use git commands to simplify your work. For example, if you made a mistake and want to revert your changes, you can use
git checkout <file>to restore the file to its original state. You can also reset all changes in the repository usinggit reset --hardcommand. Additionally, you can usegit stashandgit stash popto temporarily save your changes and restore them later. Finally, you can see the changes that you've made (which will be submitted when you callsubmit) usinggit statusorgit diffcommands. However you don't need to usegit addorgit commitbefore submitting your changes. In fact,submitonly submits currently unstaged changes because it usesgit diffto get the changes that need to be submitted.CHECKLIST
Before submitting your solution, please go over the following checklist and make sure you've done everything:
- If an error was reported in the issue description, I have successfully reproduced it.
- If an error was reported in the issue description, I have confirmed that my fix resolves the error.
- I have thought about edge cases and made sure my fix handles them as well.
- I have run existing tests in the repository that might have been affected by the change I made and confirmed that they pass. I want you to list every bullet from this checklist and write your reasoning for why you think you did it or didn't need to.
Repository has been uploaded and your shell is currently at the repository root. Time to solve the issue!
(Current directory: /nanobot, current file: none) bash-$
Assistant
Let me start by exploring the repository structure and understanding the codebase.
Assistant
Let me see the full
retain_recent_legal_suffixmethod and the surrounding code.Assistant
Now I understand the structure. The issue is in
enforce_file_cap. Let me analyze:In
enforce_file_cap:before = list(self.messages)— full listself.retain_recent_legal_suffix(limit)— this modifiesself.messagesto keep only the retained suffixdropped_count = before_count - len(self.messages)— number of dropped messagesdropped = before[:dropped_count]— firstdropped_countmessagesarchive_chunk = dropped[already_consolidated:]
The problem: When the else branch is triggered (tail is all assistant/tool messages),
retain_recent_legal_suffixanchors tolatest_userin the full session. Butdropped_count = before_count - len(self.messages)doesn't necessarily match the actual cut point.Let me think with the example from the issue:
tail = messages[0:100](100 messages)latest_user = 85(user message at position 85)max_messages = 10
In the else branch:
retained = self.messages[85:95](10 messages: positions 85-94)- Then
find_legal_message_start, etc. - After this,
self.messages = retained(10 messages) dropped = len(self.messages_before) - len(retained) = 100 - 10 = 90dropped = before[:90]= positions 0-89
But wait — positions 85-89 are part of the retained messages (positions 85-94 are retained). So
dropped[:90]= positions 0-89 OVERLAPS with retained positions 85-94!So the user message at position 85 is in BOTH
dropped(archive) andretained(kept). That's the bug.The issue says
cutshould belatest_user(85), not90. Soarchive = before[:85]andkept = before[85:].So the fix is: the
retain_recent_legal_suffixshould return the actual cut point / dropped messages, so thatenforce_file_capcan compute the archive correctly.The interface description says:
Name: retain_recent_legal_suffix Path: nanobot/session/manager.py Input: self, max_messages: int Output: tuple[list[dict], int] Description: Trims the session's message list to at most `max_messages` while preserving a legal suffix (ending with a user message if possible). Returns the list of dropped messages and a count of how many of those were already in the consolidated prefix, so callers can archive only new, unconsolidated messages.So the expected behavior is that
retain_recent_legal_suffixshould returntuple[list[dict], int]— the dropped messages (the actual dropped prefix, not computed bybefore_count - len(retained)) and a count of how many were already consolidated.Let me look at how
retain_recent_legal_suffixis called currently and what the tests expect.Assistant
Now let me understand the current behavior and the bug better. Let me look at the test for enforce_file_cap.
Assistant
Let me look at the test helpers and more tests.
Assistant
Let me look at the other callers to understand the max_messages values used.
Assistant
Let me look at the archive/consolidation logic to understand what archive expects.
Assistant
Let me look at
raw_archiveandarchiveto understand what they expect.Assistant
Now let me look at
_consolidate_replay_overflowto understand the memory.py usage pattern.Assistant
Let me check the tests related to
_idle_compact/ idle session compaction.Assistant
Let me search for
_RECENT_SUFFIX_MESSAGESand understand the default 10 reference.Assistant
Let me look at the consolidator tests that use the real
compact_idle_session.Assistant
Now I have a thorough understanding. Let me write a reproduction script first to confirm the bug.
Assistant
The bug is confirmed —
u85(user message) anda86-a89appear in both kept and archived. Now let me implement the fix. First, let me view the exact current code.Assistant
Now let me implement the fix for
retain_recent_legal_suffix:Assistant
Now let me update
enforce_file_capto use the returned dropped list:Assistant
Now let me update
compact_idle_session(the_idle_compactmethod) in memory.py, which has the identical bug pattern:Assistant
Now let me run the reproduction script again to confirm the fix:
Assistant
The fix works — no overlap,
u85only in kept, archive ends ata84. Now let me run the existing tests to ensure nothing breaks.Assistant
All 142 tests pass. Now let me write a more thorough test to verify the
compact_idle_sessionpath (memory.py) and edge cases:Assistant
Both paths work. Let me test edge cases comprehensively:
Assistant
All edge cases pass. Now let me run the broader test suite to ensure nothing else is affected:
Assistant
One test failed, but it's in
test_mcp_presets_api.pywhich is unrelated to my changes. Let me confirm it's a pre-existing failure by checking it against the original code:Assistant
Confirmed — the MCP presets test fails on the original code too (pre-existing, environment-related, unrelated to my change). Let me review my final changes:
Assistant
Let me verify the
commands.pycaller (which ignores the return value) still works fine, and check there are no type-checking issues:Assistant
Let me run the directly-related tests one more time and clean up temp files:
Assistant
All tests pass and the bug is fixed. Let me do a final review of the complete modified
retain_recent_legal_suffixandenforce_file_cap:
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.
Identical first prompt
claude-sonnet-5Compare side by side →
claude-sonnet-5Compare side by side →
claude-sonnet-5Compare side by side →
claude-sonnet-5Compare side by side →
claude-sonnet-5Compare side by side →
Qwen/Qwen3.6-35B-A3BCompare side by side →
Qwen/Qwen3.6-35B-A3BCompare side by side →
Qwen/Qwen3.6-35B-A3BCompare side by side →
Qwen/Qwen3.6-35B-A3BCompare side by side →
Qwen/Qwen3.6-35B-A3BCompare side by side →
Qwen/Qwen3.6-27BCompare side by side →
Qwen/Qwen3.6-27BCompare side by side →