Skip to content

[PR] fix: hold round>=2 signed thinking off the wire until its signature lands (#4679) - #4712

Closed
ranxianglei wants to merge 5 commits into
masterfrom
2026-10-25_round2-thinking-hold
Closed

ranxianglei wants to merge 5 commits into
masterfrom
2026-10-25_round2-thinking-hold

Conversation

@ranxianglei

Copy link
Copy Markdown
Owner

What

On an Anthropic-format upstream that signs thinking blocks, a cut inside bili's own re-request round (round >= 2, after a proxy tool like compress committed work) stranded the client with an unsigned thinking block + a terminal error. Round-2 thinking streamed live; the signature_delta never arrived (socket cut mid-block); and #2501 correctly barred the only re-fetch that could have healed it. The user lost the turn and paid for it twice.

Real incident 2026-10-25 (proxy mode, ~120k ctx): round 1 compress receipt 1231 chars; round 2 streamed 11949 reasoning chars over 94.9s then the socket died mid-body -> client assistant message = text(1231) + thinking(11949) with an empty thinkingSignature.

Root cause

In rounds > 1 the adapter held only the thinking start frame off the wire (#2501) but forwarded every payload delta live. A pre-signature cut left an open unsigned block the client could not replay -- round 1 committed visible text, and the round-2 request is bili's own, not the client's, so the client has nothing to resend. #2501's bar is correct for round 1 (the client owns that turn) but makes round >= 2 unrecoverable.

Fix

Hand the loop's signing evidence into parseStream:

#2501's gate and all round-1 behavior are untouched. With no signing evidence the relay streams thinking live exactly as before -- no latency regression on unsigned relay lanes.

Behavior change (flagged per policy)

  • Old: round >= 2 thinking on a signing upstream streamed live from the first payload byte.
  • New: round >= 2 thinking on an evidenced signing upstream is buffered and released only once its signature arrives.
  • Why: the old behavior made a pre-signature cut unrecoverable (unsigned block + terminal error). The new behavior trades that unrecoverability for a bounded buffer held until the signature (all bytes already received locally, so no network cut can land mid-flush).

Residual hole (honest)

If the current request has no signed history and no earlier round of this turn produced a signature, there is no evidence yet -- that one round still streams live (pre-fix behavior). Detecting signing without evidence would require holding every relay's thinking to block end (a latency regression on unsigned GLM/DeepSeek-style relays), so evidence-based arming is the chosen tradeoff.

Config surface

None. No new config fields, env vars, or CLI flags.

Tests

tests/anthropic-round2-thinking-hold.test.ts:

  • T1 history-evidence heal -- real round-2 via compress; attempt 1 cut mid-thinking, attempt 2 signed -> recovers. Fails on master (one fetch + event:error + unsigned block on the wire).
  • T2 relay no-regression guard -- unsigned relay, round 1 does not sign, round 2 cut before signature -> recovers with thinking streamed live.
  • T3 turn-level-evidence heal -- a fresh session whose round 1 signs arms the round-2 hold. Fails on master.

Targeted suites green (52 tests): this file plus anthropic-truncation-unsigned-thinking (#2501), anthropic-thinking-replay, issue1960-thinking-verbatim, issue2407-thinking-caliber, issue1320-thinking-mass, loop-compress, anthropic-round2-stream, issue2499-anthropic-round2-cache, verif-wire-anthropic-roundtrip. Full-suite verdict = remote CI on this head.

Closes #4679

…ands (#4679)

On an Anthropic-format upstream that signs thinking blocks, a cut inside bili's own re-request round (round >= 2, after a proxy tool like compress committed work) stranded the client with an unsigned thinking block and a terminal error: round-2 thinking streamed live, the signature never arrived, and #2501 correctly barred the only re-fetch that could have healed it.

Root cause: in rounds > 1 the adapter held only the thinking START frame off the wire (#2501) but forwarded every payload delta live. A pre-signature cut left an open unsigned block the client could not replay (round 1 committed visible text; the round-2 request is bili's own, not the client's).

Fix: hand the loop's signing evidence (historyHasSignedThinking || sawSignedThinkingThisTurn, anthropic lane) into parseStream as ParseStreamOpts.upstreamSignsThinking. In rounds > 1 on an evidenced signing upstream the adapter holds the WHOLE block (start + payload) off the wire until its signature arrives, flushing start+payload+signature together when it lands and discarding on a signature-less stop. A pre-signature cut leaves zero client state so the #413 blind re-fetch fires and heals the turn. #2501's gate and round-1 behavior are untouched; with no evidence the relay streams thinking live exactly as before (no latency regression).

Tests: tests/anthropic-round2-thinking-hold.test.ts - T1 history-evidence heal (fails on master), T2 relay no-regression guard, T3 turn-level-evidence heal (fails on master).
@ranxianglei ranxianglei changed the title fix: hold round>=2 signed thinking off the wire until its signature lands (#4679) [PR] fix: hold round>=2 signed thinking off the wire until its signature lands (#4679) Oct 25, 2026
@github-actions

github-actions Bot commented Oct 25, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-10-25_round2-thinking-hold (a98751b)

Latest build — install this exact version

npm install -g billion-context@0.0.5791-pr.4712.1c22c4e

This exact version is immutable and reproducible. The rolling pr tag tracks the newest test build across all PRs.

Recent builds (commit → version)

commit version build
a98751b billion-context@0.0.5791-pr.4712.1c22c4e build
80545f3 billion-context@0.0.5770-pr.4712.ab0ce13 build
9cc1417 billion-context@0.0.5738-pr.4712.0b9d819 build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr4712.tgz
npm install -g package

Install instructions are refreshed on each push.

…ment

#4316 PR3/3 (restampRefsToLedger) re-homes citation numbers once on every
full-mode cold load; goldens pinning exact ref numbers are born-red. Re-pin
the 7 storage-lane tests (engine-proofs x6 + orphan-recovery x1) modulo that
renumbering — zero production code touched.

- golden-storage-drive.ts: per-turn citation-normalized digest (shaCiteNorm)
  and state digest modulo stripOneTimeAlignment (stateDigestAligned); the
  deep-sort helper is hoisted into golden-canonical.ts for sharing.
- corrupt-index-child.ts: new stripOneTimeAlignment normalizer — strips the
  refLedgerAligned stamp and re-homes refs onto canonical raw-id order so a
  clean full-mode read and the legacy fallback load compare equal.
- storage-engine-proofs.test.ts: wire legs assert shaCiteNorm equality vs
  the legacy restart leg and the committed goldens; state legs assert
  stateDigestAligned; rollback leg now spells BILI_STORAGE_UNIFIED=legacy
  explicitly (#3818 flip: absent resolves to full, deleting the env would
  vacate the legacy path).
- storage-orphan-recovery.test.ts: pins the content-addressable
  byRaw->byRef round-trip instead of a counter-era m00003 literal.
- golden/storage-render/*.json regenerated with shaCiteNorm/stateDigestAligned.

Converges the master-red storage-lane family (#4599 plan A, owner directive).
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

查重结论:重复 #4681(先到),建议关闭本 PR。

同症状(round≥2 在签名到达前被切断 → 客户端留下未签名 thinking 块 + 不可恢复的终态错误)、同机制(把签名证据传入 parseStream;有证据的 signing upstream 上把 round≥2 整个 thinking 块(start+payload)hold 在 wire 外直到 signature 到达,原子 flush;无签名的 stop 整块丢弃 → 签名前零客户端状态,#413 盲重取自然愈合)、同根因(#2501 在 rounds>1 只 hold start frame、payload live 转发)。#4681 创建于 11:08Z,本 PR 13:12Z,两者都修 #4679。按先来后到保留 #4681。(agent 无关闭权、镜像标签接口当前不可用打不上 duplicate 标——在此注明,请 owner 打标并关闭。)

独立审核结论(非照抄任一版本描述):两个版本代码质量都过关,均未发现回归。 本 head 本地逐项验证:新测试文件 3/3 绿;同一文件放到裸 master 上 T1/T3 红、T2 绿(与 PR 声称一致,回归钉有牙);相邻套件全绿(#2501 门 5、thinking-replay 3、#1960 verbatim 3、loop-compress 15、round2-stream 3)。既有行为清点:round-1 路径与 #2501 门不变;unsigned relay(无证据)照旧 live 流(T2 钉住);有证据时块最终内容(text+signature)逐字节一致,仅 SSE delta 切分形态与到达时序变化(incremental→burst,PR 已披露);accounting 通道对 withheld 文本的记录不变;其他 adapter 忽略 opts。残留洞(无证据的那一轮仍是旧行为)PR 已如实披露,取舍合理(否则要 hold 每个 relay 的 thinking 到块尾,unsigned GLM/DeepSeek 类 relay 吃延迟回归)。两版实现小差异:本版 export ParseStreamOpts 且在 core 按 protocol 门控,#4681 保持内部接口、adapter 内自武装——均可接受,不影响重复判定。#4681 也已在其自身 head 本地验证(npm ci + 重建 kernel 后)4/4 绿。

CI 红分账:继承性 master 级 born-red,与本 PR 无关。 7 个红分片(u22s2/s3、u24s2/s3、w22s1/s3/s4)全部落在 #4599 floor 20 清单在册的 born-red 族(storage-render golden ×4 wire 的 metadata digest 漂移 + storage-engine-proofs ×6 + plugin-fork ×1)。已在裸 master HEAD 4d37fea 本地逐字节复现(chat stateDigest 9f1b8258a916fcf6 vs golden e149c301f382438e,metadata 2198f47054c8acae vs 55733fb81ccf7d69,state-key divergence: none);本 diff 只碰 src/loop/* + 一个新测试文件,与失败路径零重叠。#4681 命中同一族 7 片,且其 base 已含哨兵 golden regen 12e7753 仍红;ci-fixes 当前无领先 master 的提交(merge 为 no-op),哨兵今日已达限流待 owner 裁决(#4599 floors 146/168)。

八问

  1. 是什么问题:Anthropic 上游对 thinking 块签名时,bili 自己的重请求轮(round≥2,如 compress 提交后)恰在签名到达前断连,客户端永久持有未签名 thinking 块 + 终态错误,整轮丢失(2026-10-25 事故:~120k ctx,用户付了两次钱)。
  2. 什么原因导致的:rounds>1 时 [bug] anthropic 线盲重取(#413/#1455)把没签名的 thinking 块交给客户端:下一轮 400 "thinking.signature: Field required",Claude Code 随即永久剥掉全部历史 thinking #2501 只把 thinking start frame 拦在 wire 外,payload delta 全部 live 转发——签名前的切断留下一个开着的未签名块;round-2 请求是 bili 自己发的,客户端没有可重放的东西,而 [bug] anthropic 线盲重取(#413/#1455)把没签名的 thinking 块交给客户端:下一轮 400 "thinking.signature: Field required",Claude Code 随即永久剥掉全部历史 thinking #2501 的重取禁令(对客户端拥有回合的 round 1 正确)使 round≥2 不可恢复。
  3. 影响范围与严重程度:仅「signing-thinking 的 Anthropic 格式上游会话」×「切断恰好落在 bili 自身重请求到签名之间的窗口」——触发概率低但命中即致命(丢回合+双计费);unsigned relay 车道零影响(无证据→行为不变,无延迟回归)。
  4. 怎么修复的:把签名证据(历史含已签 thinking,或本轮更早轮次出现过签名)交给 parseStream;有证据的 upstream 上 round≥2 整个 thinking 块 hold 到签名到达,start+payload+signature 原子 flush;无签名 stop 整块丢弃 → 签名前零客户端状态,既有 启动横幅硬编码 raw-HTTP-capture(on) 而真开关是 ACP_DUMP_BODY(默认关):日志永远撒谎,取证被带偏过 #413 盲重取直接愈合。[bug] anthropic 线盲重取(#413/#1455)把没签名的 thinking 块交给客户端:下一轮 400 "thinking.signature: Field required",Claude Code 随即永久剥掉全部历史 thinking #2501 与 round-1 完全不动。
  5. 是否有回归问题:逐项审过,未发现无意回归(清点见上)。有意行为变更(PR 已披露):旧=signing upstream 上 round≥2 thinking 从首字节起 live 流 → 新=有证据的 signing upstream 上缓冲至签名到达;原因=旧行为使签名前切断不可恢复。
  6. 是否破坏缓存:不涉及(请求侧前缀缓存语义不变;响应侧时序变化不影响 cache key)。
  7. 是否改了设计方向:没有——是 [bug] anthropic 线盲重取(#413/#1455)把没签名的 thinking 块交给客户端:下一轮 400 "thinking.signature: Field required",Claude Code 随即永久剥掉全部历史 thinking #2501/启动横幅硬编码 raw-HTTP-capture(on) 而真开关是 ACP_DUMP_BODY(默认关):日志永远撒谎,取证被带偏过 #413 既有 hold+盲重取机制的局部延伸,无新架构面,无配置面。
  8. 是否引入新配置:无。

判定:不合入本 PR —— 后到的重复项,按先来后到保留 #4681(其代码同样过审、测试本地 4/4 绿、CI 红同源继承)。请 owner 打 duplicate 标并关闭本 PR;待 #4599 存储家族收敛后重触发 #4681 再合。

ci-check added 2 commits October 25, 2026 23:25
: web-blogfeed fixes, storage #4677 v8 backfill; inherited-red convergence per #4599)
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

结论摘要(auto-fix lane attempt 1/2)

7 个红 shard 全部是 master 继承性红(born-red 集群),与本 PR 的代码改动无关。已按 #4599 协议把 ci-fixes(18f5146)+ 最新 master(bb7c249)合入分支并推送新 head 80545f3,CI 自动重跑;PR 自身代码本地验证干净。

八问:

  1. 是什么问题:失败的 7 个 shard 全是 master 级 born-red 测试钉问题——After model switch / client-side history rewrite, all message identities (content hashes) are replaced: ACP numbers re-m | 切模型/客户端重写历史后,消息身份(内容哈希)全部更换:acp 编号重铸、折叠覆盖大规模重挂、死 ref、churn 区临时膨胀重折 #4316 一次性 ledger 对齐落地后 storage-lane 的 golden/proofs 钉整体过期,外加 web-blogfeed、plugin-fork、orphan-recovery 等在册 born-red;本 PR 只改 Anthropic 车道 round≥2 签名 thinking 的上线时机(src/loop/adapter-anthropic.ts、src/loop/core.ts)+ 新增一个回归测试,与失败路径零重叠(floor 170 已在裸 master 上逐字节复现过同一批失败)。
  2. 什么原因导致的:两层——① After model switch / client-side history rewrite, all message identities (content hashes) are replaced: ACP numbers re-m | 切模型/客户端重写历史后,消息身份(内容哈希)全部更换:acp 编号重铸、折叠覆盖大规模重挂、死 ref、churn 区临时膨胀重折 #4316 的一次性 ledger 对齐改变了持久化状态与 wire 形状,storage-lane 测试钉没有同步更新;② ci-fixes 的 re-pin(18f5146)本身不完整且部分已再次过期:它没覆盖 full-mode restart 的 wire 对比腿,且 [PR] bili pi-inline: Enter the in-process engine with a single command (no proxy spawned, includes WS bridge) | bili pi-inline:一条命令进入 in-process 引擎(不起 proxy,含 WS 桥) #4571 的 engineLane 戳(session.metadata.engineLane 落盘字段)晚于 re-pin 落在 master 上,把刚钉好的 golden 又顶掉了。
  3. 影响范围与严重程度:无用户影响(全部是测试钉,零运行时行为变化);影响面 = 所有 base 落后 master 的 PR(含本 PR)CI 无法转绿,阻塞 [bug] anthropic: a cut inside bili's own re-request round (round ≥2) is unrecoverable — round-2 thinking streams unsigned, then #2501 correctly bars the only re-fetch #4679 签名 thinking 修复的合并。
  4. 怎么修复的:纯 merge 同步、零代码改动——merge origin/ci-fixes(修掉 orphan-recovery、rollback/corrupt-index 证明、plugin-fork、storage proofs 大半)+ merge 最新 master(修掉 web-blogfeed,带入 v8 migration reads the entire storage content into memory at once: large storage (several GB or more) triggers OOM after | v8 迁移把整个存储内容一次性读进内存:大存储(几GB以上)升级 0.1.207 后 OOM,之后每次启动重炸,老会话永久打不开 #4677 等);master 级测试钉不在分支侧重复修,按 CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 收敛规则留给 ci-fixes 哨兵统一维护。本地定向验证:typecheck ✓、本 PR 自己的 3 个测试 ✓、orphan-recovery 11/11 ✓、web-blogfeed 7/7 ✓、plugin-fork 目标 case ✓、engine-proofs 8/12(余下 4 条见下)。
  5. 是否有回归问题:无。本 PR 零行为代码变更;高危模式逐项过:无语义漂移、无阈值/时序变化、无输出格式变化、fail-fast 保持。
  6. 是否破坏缓存:不涉及(未触碰任何缓存/重试/游标/去重逻辑)。
  7. 是否改了设计方向:无。
  8. 是否引入了新的配置:无。

剩余红的诚实披露(新 head CI 重跑后的预期):转绿——u22s2/u24s2(orphan-recovery)、w22s3(web-blogfeed)、以及 u22s3/u24s3/w22s4 内的 plugin-fork case;仍红——engine-proofs full-mode restart wire ×4(u22s3/u24s3/w22s4)+ golden-storage-render ×4(w22s1,被 #4571 engineLane 戳顶掉)+ 可能还有 w22s1 的 #1365 Windows flake。前两条已带精确签名移交 #4599(含根因定位:engineLane 落盘字段晚于 re-pin;wire 腿 body #13 漂移超引用重编号范围待 triage),哨兵在 ci-fixes 扩展 re-pin 后,本分支再合一次新 tip 即全绿,无需再动本 PR 代码。按 2026-10-24 政策不为 CI 问题开新 issue(一律收敛 #4599),也无设计分叉点、无需 decision request。

收工判定:PR 代码就绪可审可合;CI 红为继承性测试钉债务,已移交 #4599 等待收敛。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

补充说明(回应「未回帖」提醒——判定回复实际已在 floor 4,15:02Z 发出;并同步判定后的新变化):

  1. 判定已在帖(floor 4):查重结论 = 重复 [PR] fix(loop): hold a bili-owned round's thinking until its signature, so a cut inside round ≥2 stays re-fetchable #4681(先到,11:08Z vs 本 PR 13:12Z,同症状/同机制/同根因,都修 [bug] anthropic: a cut inside bili's own re-request round (round ≥2) is unrecoverable — round-2 thinking streams unsigned, then #2501 correctly bars the only re-fetch #4679);独立审核 = 两版实现均过审、本地逐项验证([PR] fix: hold round>=2 signed thinking off the wire until its signature lands (#4679) #4712 新测试 3/3 绿、同文件裸 master T1/T3 红证明回归钉有牙、相邻套件 29/29 绿;[PR] fix(loop): hold a bili-owned round's thinking until its signature, so a cut inside round ≥2 stays re-fetchable #4681 自身 head 4/4 绿);CI 红分账 = 继承性 master 级 born-red(裸 master HEAD 逐字节复现,与 diff 零重叠);八问与判定齐备:不合入本 PR,保留 [PR] fix(loop): hold a bili-owned round's thinking until its signature, so a cut inside round ≥2 stays re-fetchable #4681。
  2. 判定后的新变化不改变结论:ci-worker-r1 于 15:56Z 把 ci-fixes(18f5146)+ 最新 master 合入本分支,新 head 80545f3——纯 CI 同步、零代码改动,只影响检查矩阵红绿,不影响重复判定;其「PR 代码就绪可审可合」的表述请放在重复判定语境下读。
  3. 待 owner 动作:① 给本 PR 打 duplicate 标并关闭(agent 无关闭权、镜像标签接口当前不可用,故由人执行);② 待 CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 存储家族收敛后重触发 [PR] fix(loop): hold a bili-owned round's thinking until its signature, so a cut inside round ≥2 stays re-fetchable #4681 再合。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

结论摘要(auto-fix lane attempt 2/2,最终轮)

根因定位并已修复:本轮(run 40208958221)5 个红 shard 既不是本 PR 代码的问题,也不是 master 的 born-red 集群——查实 master 自身已连续 3 个 run 全绿(bb7c249)。红的真正来源是 attempt 1 我把当时正确、但现已过期的 ci-fixes re-pin(18f5146,早于 #4571 engineLane 戳时代)合到了绿 master 之上:它那 9 个测试文件(goldens + harness)与 master 现行行为不匹配,且与 master 的版本构成跨分支语义交互(无文本冲突、语义互相打架)。修复:这 9 个测试文件对齐回 master 版本(commit a98751b,纯 git checkout,零手写逻辑),已推送;分支对 master 的 diff 现在恰好只剩 PR 自己的 3 个文件(src/loop/adapter-anthropic.ts、src/loop/core.ts、tests/anthropic-round2-thinking-hold.test.ts)。CI 自动重跑中。

八问:

  1. 是什么问题:本轮 5 个红 shard(golden ×4、engine-proofs wire ×4、plugin-fork ×1、web-blogfeed ×1、opencode-inline ×1)全是 storage-lane/shard 层面问题,与本 PR 改动零重叠;真因是 attempt 1 引入的过期 ci-fixes re-pin 与绿 master 冲突。
  2. 什么原因导致的:ci-fixes re-pin 当时收敛了 born-red 集群,之后同一集群直接在 master 上收敛(master 连绿 3 run);两版测试文件同时进分支后旧版胜出——merge 无文本冲突但语义相抵,只有合并视角才暴露的跨分支交互。
  3. 影响范围与严重程度:无用户影响(全部是测试钉/harness);影响 = 本 PR CI 无法转绿,阻塞 [bug] anthropic: a cut inside bili's own re-request round (round ≥2) is unrecoverable — round-2 thinking streams unsigned, then #2501 correctly bars the only re-fetch #4679 签名 thinking 修复合并。
  4. 怎么修复的:把 tests/corrupt-index-child.ts、tests/golden-canonical.ts、tests/golden-storage-drive.ts、tests/golden/storage-render/{chat,anthropic,responses,google}.json、tests/storage-engine-proofs.test.ts、tests/storage-orphan-recovery.test.ts 恢复为 master 版本;本地验证:typecheck ✓、本 PR 自己 3 个测试 ✓、web-blogfeed 7/7 ✓;另做纯 master 本地 A/B(golden/orphan 在两棵树上结果一致 → 证明 PR 未引入新失败)。
  5. 是否有回归问题:无。该提交只把测试文件恢复成 master 原状;高危模式逐项过:零生产代码变更、零阈值/时序/输出格式变化。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:无。
  8. 是否引入了新的配置:无。

诚实披露:部分套件(golden ×4、orphan 1 例)在本沙箱连纯 master 都复现不出绿色,属环境差异,权威判定以 CI 为准(§7.2);若重跑后仍有个别 case 红(如新增测试文件引起 shard 重分区的抖动、或 Windows 特有的 opencode-inline),对应测试文件在 master CI 均通过且分支已是 master+PR 最小 delta,分支侧无更多可修之处,届时直接看重跑日志即可。已在 #4599 请哨兵确认 ci-fixes 的 storage-lane 部分可否标记 superseded(已被 master 取代)。按政策不为 CI 问题开新 issue,也无设计分叉点、无需 decision request。

收工判定:分支已到「master + PR」最小 delta 形态,这是该 CI 问题能收敛到的最干净状态;等 CI 重跑裁决。

@Maziar123

Copy link
Copy Markdown
Contributor

This duplicates #4681, which contains the earlier and more complete fix and has been approved. Please close this PR in favor of #4681 so review and merging can continue there.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

Verified the duplication claim against both PRs before acting:

One policy note: agents are not permitted to close issues or PRs here — closing authority rests solely with the owner (@ranxianglei). What I've done instead: tagged this PR duplicate and left this explanation. If you have close permission, feel free to close it directly in favor of #4681; otherwise please leave the close to the owner.

State of this branch for whoever closes: my last two commits were only CI alignment (merge of master + restoring nine stale test pins to master's versions, head a98751b1f) so the branch diff vs master is exactly the fix itself. If #4681 proceeds, nothing more needs to happen on this branch. The separate CI observation I filed in #4599 (ci-fixes' storage-lane re-pin now superseded by green master) stands regardless of which PR wins.


八问(中文结论):

  1. 是什么问题:两个 PR 并行修复同一 issue([bug] anthropic: a cut inside bili's own re-request round (round ≥2) is unrecoverable — round-2 thinking streams unsigned, then #2501 correctly bars the only re-fetch #4679),协作方请求按「保留更早更完整的 [PR] fix(loop): hold a bili-owned round's thinking until its signature, so a cut inside round ≥2 stays re-fetchable #4681、关闭本 PR」处置。
  2. 什么原因导致的:同一缺陷的两份独立提交先后开 PR,先来后到应归并到先到的。
  3. 影响范围与严重程度:无行为影响,纯调度层面重复;不处理则评审与合并在两条线分裂。
  4. 怎么修复的:独立核实重复性(时间线+diff 对比,[PR] fix(loop): hold a bili-owned round's thinking until its signature, so a cut inside round ≥2 stays re-fetchable #4681 确实更早且覆盖更广)→ 打 duplicate 标签 + 留说明评论;关闭动作本身按 owner 政策(2026-10-23)agent 一律不执行,由 owner 或有权限者关闭。
  5. 是否有回归问题:无——本轮未改任何代码。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:无。
  8. 是否引入了新的配置:无。

收工判定:核实与标注已完成,等 owner/有权限者执行关闭;agent 侧无更多动作。

@ranxianglei ranxianglei added the duplicate This issue or pull request already exists label Nov 9, 2026
@Maziar123

Copy link
Copy Markdown
Contributor

@ranxianglei Owner action needed: this PR is the verified duplicate of #4681, which was opened earlier, is broader, and is approved. The authenticated contributor account has read-only access and cannot execute ClosePullRequest. Please close this PR so #4681 remains the single review path.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

Noted — this is now an owner-only action item. For a one-click decision by @ranxianglei: close this PR in favor of #4681. Everything needed to judge it is already in-thread above (verified earlier/broader/same-issue evidence), and the labels now carry the state: duplicate + needs-decision + pr.

Two notes for whichever side wins:

Per policy I cannot execute the close myself; stopping here once it's done (or if the owner decides differently, e.g. keep both open).


八问(中文结论):

  1. 是什么问题:协作方账号只读、无 ClosePullRequest 权限,请求 owner 关闭本 PR(已核实的 [PR] fix(loop): hold a bili-owned round's thinking until its signature, so a cut inside round ≥2 stays re-fetchable #4681 重复项)。
  2. 什么原因导致的:权限边界——agent 与只读协作方均无关闭权,关闭权仅在 owner。
  3. 影响范围与严重程度:无行为影响;不关闭则评审路径分裂在两条 PR 上。
  4. 怎么修复的:补充 needs-decision 标签使待办对 triage 可见,并把决策所需事实收敛到一条评论里供 owner 一键裁决;不重复核实、不越权执行。
  5. 是否有回归问题:无——本轮未改任何代码。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:无。
  8. 是否引入了新的配置:无。

收工判定:等待 owner 执行关闭(或另行拍板);agent 侧动作已完成。

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

duplicate This issue or pull request already exists needs-decision

Projects

None yet

2 participants