Skip to content

[PR] fix: preserve signed ordinary text parts on the wire and in tool-loop rebuilds - #6859

Open
ranxianglei wants to merge 1 commit into
masterfrom
2026-11-15_signed-ordinary-text
Open

ranxianglei wants to merge 1 commit into
masterfrom
2026-11-15_signed-ordinary-text

Conversation

@ranxianglei

Copy link
Copy Markdown
Owner

Closes #6842

What

Gemini can attach a thoughtSignature to an ordinary text part (no thought: true). Per KDD #10, a signed block is verified byte-for-byte on replay and must never be rewritten. Two paths violated that for signed ordinary text:

  1. Client wire — the Google SSE adapter and the non-streaming JSON rewriter only exempted thought: true parts from prose filtering, so a signed ordinary text part containing marker-like text (render tags / [ACP] markers / bili internal artifacts) was mutated before reaching the client.
  2. Tool-loop re-request — the streaming loop recorded ordinary text as a merged string and dropped its signature when rebuilding the next upstream request, even when the text contained no markers. The continuation then disagreed with the client's persisted history.

How

  • src/loop/adapter-google.ts: a text part carrying a non-empty thoughtSignature is now treated like a thought part — never prose-filtered, and the signature rides the loop event. The complete frame was already forwarded verbatim via the carrier mechanism, so the client now gets the text + signature byte-exact (SSE and JSON-array forms).
  • src/loop/core.ts: the text RoundPart / ParsedStreamEvent carry an optional signature. Accumulation treats a part that already carries its signature as complete (Gemini computes the signature over the full part text, so it arrives in that part's last chunk; any later text delta starts a new part). The proxy-tool rebuild puts the signature back on the re-request as googleThoughtSignature at the part's position, so two adjacent signed parts stay two signed parts with their boundary intact.
  • src/stream-google.ts: the non-streaming JSON rewriter exempts signed parts from stripAcpTags (and from the requestText pre-scan).

Behavior changes (old → new)

  • Signed ordinary text parts containing marker-like text: were prose-filtered before reaching the client → now forwarded byte-exact. (Unsigned prose filtering is unchanged.)
  • Tool-loop re-request for a signed ordinary text part: had no thoughtSignature → now carries googleThoughtSignature. (Unsigned text still gets no signature — nothing is fabricated.)

Both changes are the intended fix for the reported corruption. Existing thought / functionCall / standalone-signature behavior is unchanged (gated by negative test rows).

Modes

Reasoned in both compression modes. The client-wire fix is mode-independent (adapter + JSON rewriter run on every response). The re-request fix lives in the proxy-tool rebuild, which is exercised in proxy mode; the adapter's signature-carrying event is also what plugin mode's re-sent history relies on. The regression suite covers both SSE and non-streaming JSON responses.

Tests

New tests/issue6842-google-signed-ordinary-text.test.ts (11 tests, both directions per gate discipline):

  • signed ordinary text with marker-like text forwarded byte-exact (SSE + JSON-array)
  • signed ordinary text without markers (baseline)
  • gate: unsigned text with marker-like text is still filtered
  • gate: thought parts + standalone signatures keep existing behavior
  • the issue's repro: two adjacent signed text parts keep both signatures and their part boundary on the re-request
  • multi-chunk signed part: deltas merge, the last-chunk signature lands on the merged part
  • gate: unsigned text in a proxy-tool round gets no fabricated signature
  • JSON rewriter: signed not rewritten / gate: unsigned still stripped / gate: thought + real functionCall unchanged

Targeted suites green on the rebased head (71 tests incl. issue3158, json-rewrite-history, adapter-google, e2e-google-compress); typecheck + build clean. Full-suite verdict is the CI test lanes.

No new config surface. No wire-contract ledger entry — this establishes local wire corruption / metadata loss, not a new upstream rejection constraint (per the issue, no claim that every model rejects missing text-part signatures with HTTP 400).

… rebuilds (#6842)

Gemini can attach a thoughtSignature to an ORDINARY text part (without
thought: true). The signature is verified byte-for-byte over the part's
text on replay (KDD #10), so such a part must be treated like a thought
part — never rewritten.

Before: the Google SSE adapter and the non-streaming JSON rewriter only
exempted thought: true parts from prose filtering, so a signed ordinary
text part containing marker-like text was mutated before reaching the
client. Separately, the streaming tool loop recorded ordinary text as a
merged string and dropped its signature when rebuilding the next upstream
request, even when the text contained no markers.

Now: signed ordinary text parts are never prose-filtered (adapter + JSON
rewriter), keeping the text, signature, and part boundary byte-exact on
the client wire. The loop carries the signature on the text part and puts
it back on the re-request (googleThoughtSignature) at its position, so the
continuation agrees with the client's persisted history.

A signed text part is a complete part: Gemini computes the signature over
the full part text, so it arrives in the part's last chunk; any later text
delta starts a new part.

Closes #6842
@ranxianglei ranxianglei added bug Something isn't working gemini p2 scope/wiring labels Nov 15, 2026
@github-actions

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-11-15_signed-ordinary-text (79b929b)

Latest build — install this exact version

npm install -g billion-context@0.0.10227-pr.6859.8f0947b

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
79b929b billion-context@0.0.10227-pr.6859.8f0947b build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr6859.tgz
npm install -g package

Install instructions are refreshed on each push.

@ranxianglei ranxianglei changed the title fix: preserve signed ordinary text parts on the wire and in tool-loop rebuilds [PR] fix: preserve signed ordinary text parts on the wire and in tool-loop rebuilds Nov 15, 2026
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

查重: 不重复。检索 open+closed 的 issue 与 PR,与 #6842 相关的既有工作:#3162(standalone signature)、#6103(promote sig to surviving functionCall)、#6623(duplicate whole-frame forwarding)——三者都仍把 signed ordinary text 当 unsigned prose 处理,与本 PR 的修复面(把 signed ordinary text 从 prose filter 豁免 + 在 rebuild 里带回 signature)不重叠。本 thread 无并列的第二条 fix PR(#6859 是唯一)。

Review of #6859 — preserve signed ordinary text parts

Verdict: MERGEABLE (pending CI green — 30 checks, 0 failed, 13 still running at review time).

Direction (① 原 issue)

#6842 reports that Gemini attaches thoughtSignature to ordinary text parts (no thought: true), and two paths violate KDD#10 (signed blocks replayed byte-for-byte, never rewritten):

  1. Client wire — the Google SSE adapter + non-streaming JSON rewriter only exempted thought === true from prose filtering, so signed ordinary text containing marker-like content (render tags / ACP markers) was mutated before reaching the client.
  2. Tool-loop re-request — the streaming loop recorded ordinary text as a merged string and dropped the signature, so the rebuild lost the thoughtSignature even without markers.

The PR fixes both root causes, not the symptom: it changes the filter-exemption condition from "is thought" to "is thought OR carries a non-empty thoughtSignature", and carries the signature through accumulation + rebuild. This is 疏 not 堵 — it fixes the producer (the exemption decision + the signature carrier), it does not filter payload in transit.

Regression audit (② 既有行为清点)

The diff touches 4 files (+389/−8), all on-topic, no unrelated churn. Branch is fresh (0 behind master @ f648217cb8). Existing behaviors touched, each checked:

Path Existing behavior Verdict
Prose filter (adapter-google.ts) unsigned text filtered Kept — gate tests d/j confirm unsigned still filtered
Prose filter thought parts exempt Kept — gate tests e/k confirm thought + standalone sig unchanged
RoundPart/ParsedStreamEvent accumulation (core.ts) unsigned text merged Kept — gate test h confirms no fabricated signature on unsigned
Proxy-tool rebuild text part (core.ts) unsigned text, no sig field Kept — googleThoughtSignature added only when protocol === "google" && part.signature
JSON rewriter (stream-google.ts) unsigned stripped Kept — gate test j
Degenerate-turn warn / strip summary counts filtered text chars Consistent — signed chars no longer counted, same as thought chars (already not counted, comment lines 369-371)
recordReplyText for signed parts was filtered Consistent — now raw, same as reasoning parts (already raw, core.ts:1016)

No semantic drift, no fail-fast→swallow, no threshold/timing/output-format drift. No kernel/ changes — the kernel already maps googleThoughtSignature → part.thoughtSignature in both contentParts() branches (kernel/src/wire/google.ts:520,541), and coreToGoogle preserves part boundaries (two adjacent signed parts stay two parts). So the rebuild serializes correctly with no kernel gate triggered.

Test validity (④)

tests/issue6842-google-signed-ordinary-text.test.ts — 11 tests, both gate directions per #3617, with teeth (deepEqual on re-request body AND client-visible stream, exact byte counts, signature counts):

  • a/b: signed ordinary text w/ marker-like text forwarded byte-exact (SSE + JSON-array)
  • c: baseline no-markers
  • d (gate): unsigned w/ markers still filtered
  • e (gate): thought parts + standalone sigs unchanged
  • f: the issue repro — two adjacent signed parts keep both sigs AND part boundary on re-request
  • g: multi-chunk signed part — deltas merge, last-chunk sig lands on merged part
  • h (gate): unsigned gets NO fabricated signature
  • i: JSON rewriter signed not rewritten
  • j (gate): JSON rewriter unsigned still stripped
  • k (gate): JSON rewriter thought + signed functionCall unchanged

Loop-level byte-equality on the re-request body (f/g/h) + client-visible stream (a/b/c) — satisfies KDD#10's requirement for both signed wires.

Local targeted verification (on PR head 79b929bf9f)

  • New test file: 11/11 pass
  • Neighbor Google/loop (adapter-google, google-complete-frame, issue1960, issue3158, json-rewrite-history, issue6103, issue6623): 75/75 pass
  • Core loop (loop-core, issue539, issue1960-loop-redacted, loop-adapters): 73/73 pass
  • npm run typecheck: clean
  • Total: 159 targeted tests, 0 failures. Full-suite verdict = remote CI (running).

One noted assumption (not a blocker)

Test g pins delta-merge semantics: each chunk's text is a delta, the signature arrives in the part's last chunk. If Gemini instead re-sent the full part text in the final chunk, the accumulated text would double. Evidence it does not: (1) the existing reasoning path uses identical append-only accumulation and works in production for signed thought parts — a full-text resend would visibly corrupt production thought parts (no such report); (2) the issue's repro fixtures were built from observed traffic + Google docs; (3) test g pins the shape. I could not re-verify against Google docs (firewalled env), so this is verified-by-consistency, not by doc.

Labels

Independently labeled on the PR's own diff: bug / p2 / scope/wiring / gemini — correct (Gemini-wire defect, narrow trigger, session-brick when hit).


结构化八问(中文)

  1. 是什么问题:Gemini 把 thoughtSignature 挂在普通文本 part(非 thought:true)上时,客户端线路上该文本会被 prose filter 改写、tool-loop 重建时 signature 丢失,违反 KDD#10「签名块必须逐字节回放」——下一轮 replay 时签名校验失败,会话 brick。
  2. 什么原因导致的:根因是两条路径的「签名 part 判定」只看 thought === true 标志、不看「是否携带 signature」。① 客户端线(SSE adapter + 非流式 JSON rewriter)的 filter 豁免条件只认 thought,导致带 marker 样内容的 signed 普通文本在发给客户端前被改写;② 流式 loop 把普通文本记成合并字符串、丢弃 signature,rebuild 无法还原。表面现象是「签名丢失」,根因是「判定条件漏了 signature 这一维」。
  3. 影响范围与严重程度:仅 Gemini(google 协议)会话,且需模型把 signature 挂在普通文本(非 thought)上、并满足「文本含 marker 样内容(客户端线)」或「发生 tool continuation(rebuild)」之一。触发时会话 brick(致命),但触发条件窄、非普适故障,故 p2(个体功能异常、特定条件下、可绕行)。
  4. 怎么修复的:机制级修复,非治标——① filter 豁免条件从「is thought」改为「is thought 或携带非空 thoughtSignature」,signed 普通文本逐字节发给客户端(与 thought part 同待遇,符合 [acp-loop] In-turn compress re-request edits the latest assistant message → Anthropic 400 "thinking blocks … cannot be modified" (session stays broken) #1960 裁决);② signature 在 accumulation(把 signed part 视为完整、后续 delta 起新 part)与 rebuild(text part 加 googleThoughtSignature)全程携带,使 re-request 与客户端持久化历史逐字节一致。kernel 已映射 googleThoughtSignature → part.thoughtSignature,无需改 kernel。
  5. 是否有回归问题:逐项审过,无回归——unsigned 文本在四条路径仍被 filter(gate 测试 d/j/h)、thought part 与 standalone signature 不变(gate e/k)、degenerate-turn 警告与 strip 摘要不计 signed 字符(与 thought 字符一致)、recordReplyText 对 signed part 记 raw(与 reasoning part 一致)。无语义漂移、无 fail-fast 变吞错、无阈值/时序/输出格式漂移。
  6. 是否破坏缓存:不涉及——未触碰 cache、retry、增量游标、去重键。
  7. 是否改了设计方向:无架构转向。是把既有「signed part 永不 filter」原则([acp-loop] In-turn compress re-request edits the latest assistant message → Anthropic 400 "thinking blocks … cannot be modified" (session stays broken) #1960)从 thought part 扩展到 signed 普通文本 part,数据流不变。
  8. 是否引入了新的配置:无新配置。

收工判定:MERGEABLE(待 CI 全绿)。修复正确、打到根因、无回归、测试有牙且双向、独立打标;唯一 noted assumption(multi-chunk delta-merge)经一致性论证成立,非 blocker。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

🔶 需要决策

The #6859 ↔ master conflict is not a textual merge conflict — it is two competing fixes for the same issue (#6842) colliding on three files. Resolving it means choosing/combining implementations, which is a design call I will not make unilaterally. I did NOT force-push and did NOT touch any branch.

What I verified (evidence, not guessing)

The key finding (I ran the tests)

I ran the PR's own test file tests/issue6842-google-signed-ordinary-text.test.ts (11 cases, with positive+negative gates) against both implementations:

  • Against the PR's implementation: 11/11 pass.
  • Against master's Afloat implementation: 10/11 — case #6842 g fails.

#6842 g = "multi-chunk signed part — text deltas merge, the signature (last chunk) lands on the merged part". Afloat's accumulation logic splits a signed part that spans several SSE chunks into two parts (signature attached only to the last chunk's part). Gemini computes the signature over the full part text and delivers it on the last chunk, so the correct re-request is one merged part carrying the signature. Split, the signature no longer matches the text; on the client's echo-back Gemini's verification fails → the continuation bricks.

Bottom line: the PR is not redundant — its test catches a real (edge-condition) defect in the already-merged Afloat fix. But Afloat's fix has wider coverage (full-part preservation via rawGoogleParts + plugin.ts). How to reconcile is the decision.

八问(决策请求)

  1. 是什么问题: 同一 Google signed ordinary text is rewritten and loses signatures in tool continuations #6842 的两个竞争性修复在 3 个文件上撞车;冲突解决 = 选实现或合并实现,属设计决策,非机械 rebase。
  2. 什么原因导致的: 两个修复并行落地——Afloat 的先合入 master,PR 基于更早 base(f648217cb8)用不同思路开发,base 落后后冲突。根因是「同 issue 双 PR」,不是代码 bug。
  3. 影响范围与严重程度(按选项):
    • A 关 PR(留 master): 多 chunk 签名 part 被拆分 → 签名与文本错位 → 特定条件(签名 part 跨多 chunk 且后续有 proxy-tool 轮)下 replay 验签失败、会话 brick。边缘条件,但触发即 p1(签名/会话不可用类)。
    • B 全用 PR 实现: 修好多 chunk;但丢 rawGoogleParts 全 part 保留——签名 part 若同时带图片,图片会丢(回退 Afloat 已覆盖的行为)。
    • C/D 混合: 保留 Afloat 全 part 保留 + plugin.ts 覆盖,同时修好多 chunk。是 Afloat 的超集,无已知回归。
  4. 怎么修复的(推荐 C/D): 保留 Afloat 的 rawGooglePart 承载(整 part + rawGoogleParts 回写,与现有 Gemini 图片/媒体机制一致)+ plugin.ts 两处;只把 src/loop/core.ts 的累积逻辑改成 PR 的签名感知合并(签名 part 的最后一个 chunk 并入前面的无签名 text part,而非另起新 part)。PR 的测试文件 cherry-pick 进 master 作回归测试。
  5. 是否有回归问题(按选项): A=无新回归,但保留 Afloat 的多 chunk bug;B=有回归风险(丢 rawGoogleParts,破坏签名 part 携带图片的场景);C/D=无回归(超集)。
  6. 是否破坏缓存: 不涉及缓存/重试/增量游标。影响的是「签名是否正确挂回 re-request」(KDD release v0.1.3 #10 逐字节契约),不触及 prefix cache 键。
  7. 是否改了设计方向: 不改架构/边界/数据流。只在「签名普通 text part 如何在 loop 里承载与回写」这一个机制上二选一(或取并)。
  8. 是否引入了新的配置: 不涉及。无新 env/config/默认值。

选项对比

选项 多 chunk 正确 全 part(rawGoogleParts)保留 plugin.ts 覆盖 回归风险
A 关 PR(留 master) ❌ 有 bug ✅ ✅ 无新回归,但留 bug
B 全用 PR 实现 ✅ ❌ 丢图片 ❌ 需补 中(丢媒体保留)
C 混合(推荐) ✅ ✅ ✅ 无(超集)
D 修 Afloat 累积 ✅ ✅ ✅ 无(同 C,落在 Afloat 代码上)

Recommendation: C/D — keep Afloat's wider coverage (full part + plugin.ts) and fix only its accumulation's multi-chunk defect using the PR's signature-aware merge; cherry-pick the PR's test file into master as the regression test. Whichever option is chosen, the PR's test file should be kept — it is the only test that catches the defect.

Default if undecided: leave the PR conflicted (no force-push), keep master on Afloat's fix as-is (zero new regression), and let the multi-chunk gap stay documented here for a follow-up PR once the direction is set. (Closing the PR is owner-only; I will not close it.)

needs-decision label applied. This run: no force-push, no branch modified.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

Status (re-trigger, floor 9): unchanged since floor 6 — still waiting on your A/B/C/D call. No force-push, no branch touched.

Re-verified live this run (not from memory):

Per my stated default (floor 6): holding — no force-push, master stays on Afloat's fix (zero new regression). The full analysis (options table, the #6842 g test evidence, the 八问) is at floor 6 — not re-posting it here.

I need your call on A/B/C/D to proceed. Until then, each re-trigger will find the same state. (Closing the PR is owner-only; I will not close it.)

八问(状态更新,细节见 floor 6)

  1. 是什么问题: 同 floor 6——Google signed ordinary text is rewritten and loses signatures in tool continuations #6842 两个竞争性修复在 3 个文件上撞车,属设计决策,非机械 rebase。
  2. 什么原因导致的: 同 floor 6——两个修复并行落地,Afloat 的先合入 master,PR 基于更早 base。
  3. 影响范围与严重程度: 同 floor 6(按 A/B/C/D 选项分别列于 floor 6)。
  4. 怎么修复的: 本轮未修——等 owner 拍板;默认动作 = 保持冲突态、不 force-push。
  5. 是否有回归问题: 无——本轮未改任何分支、未 force-push,master 维持 Afloat 现状。
  6. 是否破坏缓存: 不涉及。
  7. 是否改了设计方向: 不涉及(未动)。
  8. 是否引入新配置: 不涉及。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

2 similar comments
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Status (3rd re-trigger, floor 15): still unchanged — owner decision on A/B/C/D still pending. No force-push, no branch touched.

Re-verified live this run:

This is the 3rd re-trigger with no decision yet. The full analysis (options, the #6842 g test evidence, 八问) is at floor 6. I'm holding per my stated default (no force-push, master stays on Afloat's fix).

To stop the re-trigger loop, please either:

  • Decide A/B/C/D — my recommendation is C/D: keep Afloat's rawGooglePart + plugin.ts coverage, fix the multi-chunk accumulation bug in core.ts (the PR's signature-aware merge), and cherry-pick the PR's test file as the regression test; or
  • Close the PR (owner-only) if you'd rather take Afloat's fix as final and track the multi-chunk gap separately.

八问(状态更新,细节见 floor 6)

  1. 是什么问题: 同 floor 6——Google signed ordinary text is rewritten and loses signatures in tool continuations #6842 两个竞争性修复撞车,属设计决策,非机械冲突。
  2. 什么原因导致的: 同 floor 6——两个修复并行落地,PR 未 rebase 到含 Afloat 修复的 master。
  3. 影响范围与严重程度: 同 floor 6(按选项 A/B/C/D 列于 floor 6)。
  4. 怎么修复的: 本轮未修——等 owner 拍板;默认 = 保持冲突态、不 force-push。
  5. 是否有回归问题: 无——本轮未改分支、未 force-push,PR 分支仍为 79b929bf。
  6. 是否破坏缓存: 不涉及。
  7. 是否改了设计方向: 不涉及(未动)。
  8. 是否引入新配置: 不涉及。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Status (4th re-trigger, floor 19): still unchanged — owner decision on A/B/C/D still pending. No force-push, no branch touched.

Re-verified live this run:

This is the 4th re-trigger with no decision. The full analysis (options, the #6842 g test evidence, 八问) is at floor 6. I'm holding per my stated default (no force-push, master stays on Afloat's fix).

To stop the re-trigger loop, please either:

  • Decide A/B/C/D — my recommendation is C/D: keep Afloat's rawGooglePart + plugin.ts coverage, fix the multi-chunk accumulation bug in core.ts (the PR's signature-aware merge), and cherry-pick the PR's test file as the regression test; or
  • Close the PR (owner-only) if you'd rather take Afloat's fix as final and track the multi-chunk gap separately.

八问(状态更新,细节见 floor 6)

  1. 是什么问题: 同 floor 6——Google signed ordinary text is rewritten and loses signatures in tool continuations #6842 两个竞争性修复撞车,属设计决策,非机械冲突。
  2. 什么原因导致的: 同 floor 6——两个修复并行落地,PR 未 rebase 到含 Afloat 修复的 master。
  3. 影响范围与严重程度: 同 floor 6(按选项 A/B/C/D 列于 floor 6)。
  4. 怎么修复的: 本轮未修——等 owner 拍板;默认 = 保持冲突态、不 force-push。
  5. 是否有回归问题: 无——本轮未改分支、未 force-push,PR 分支仍为 79b929bf。
  6. 是否破坏缓存: 不涉及。
  7. 是否改了设计方向: 不涉及(未动)。
  8. 是否引入新配置: 不涉及。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

1 similar comment
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Status (5th re-trigger, floor 24): still unchanged — owner decision on A/B/C/D still pending. No force-push, no branch touched.

Re-verified live this run:

This is the 5th re-trigger with no decision. The full analysis (options, the #6842 g test evidence, 八问) is at floor 6. I'm holding per my stated default (no force-push, master stays on Afloat's fix).

To stop the re-trigger loop, please either:

  • Decide A/B/C/D — my recommendation is C/D: keep Afloat's rawGooglePart + plugin.ts coverage, fix the multi-chunk accumulation bug in core.ts (the PR's signature-aware merge), and cherry-pick the PR's test file as the regression test; or
  • Close the PR (owner-only) if you'd rather take Afloat's fix as final and track the multi-chunk gap separately.

八问(状态更新,细节见 floor 6)

  1. 是什么问题: 同 floor 6——Google signed ordinary text is rewritten and loses signatures in tool continuations #6842 两个竞争性修复撞车,属设计决策,非机械冲突。
  2. 什么原因导致的: 同 floor 6——两个修复并行落地,PR 未 rebase 到含 Afloat 修复的 master。
  3. 影响范围与严重程度: 同 floor 6(按选项 A/B/C/D 列于 floor 6)。
  4. 怎么修复的: 本轮未修——等 owner 拍板;默认 = 保持冲突态、不 force-push。
  5. 是否有回归问题: 无——本轮未改分支、未 force-push,PR 分支仍为 79b929bf。
  6. 是否破坏缓存: 不涉及。
  7. 是否改了设计方向: 不涉及(未动)。
  8. 是否引入新配置: 不涉及。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Status (6th re-trigger, floor 28): still unchanged — owner decision on A/B/C/D still pending. No force-push, no branch touched.

Re-verified live this run (not from memory):

This is the 6th re-trigger with no decision. The re-trigger loop will continue until either the owner decides or the PR is closed. The full analysis (options table, the #6842 g test evidence, 八问) is at floor 6 — not re-posting it here.

To stop the re-trigger loop, please either:

  • Decide A/B/C/D — my recommendation is C/D: keep Afloat's rawGooglePart + plugin.ts coverage, fix the multi-chunk accumulation bug in core.ts (the PR's signature-aware merge), and cherry-pick the PR's test file as the regression test; or
  • Close the PR (owner-only) if you'd rather take Afloat's fix as final and track the multi-chunk gap separately.

八问(状态更新,细节见 floor 6)

  1. 是什么问题: 同 floor 6——Google signed ordinary text is rewritten and loses signatures in tool continuations #6842 两个竞争性修复撞车,属设计决策,非机械冲突。
  2. 什么原因导致的: 同 floor 6——两个修复并行落地,PR 未 rebase 到含 Afloat 修复的 master。
  3. 影响范围与严重程度: 同 floor 6(按选项 A/B/C/D 列于 floor 6)。
  4. 怎么修复的: 本轮未修——等 owner 拍板;默认 = 保持冲突态、不 force-push。
  5. 是否有回归问题: 无——本轮未改分支、未 force-push,PR 分支仍为 79b929bf。
  6. 是否破坏缓存: 不涉及。
  7. 是否改了设计方向: 不涉及(未动)。
  8. 是否引入新配置: 不涉及。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Status (7th re-trigger, floor 32): still unchanged — owner decision on A/B/C/D still pending. No force-push, no branch touched.

Re-verified live this run (not from memory):

This is the 7th re-trigger with no decision. The full analysis (options table, the #6842 g test evidence, 八问) is at floor 6 — not re-posting it here.

To stop the re-trigger loop, please either:

  • Decide A/B/C/D — my recommendation is C/D: keep Afloat's rawGooglePart + plugin.ts coverage, fix the multi-chunk accumulation bug in core.ts (the PR's signature-aware merge), and cherry-pick the PR's test file as the regression test; or
  • Close the PR (owner-only) if you'd rather take Afloat's fix as final and track the multi-chunk gap separately.

八问(状态更新,细节见 floor 6)

  1. 是什么问题: 同 floor 6——Google signed ordinary text is rewritten and loses signatures in tool continuations #6842 两个竞争性修复撞车,属设计决策,非机械冲突。
  2. 什么原因导致的: 同 floor 6——两个修复并行落地,PR 未 rebase 到含 Afloat 修复的 master。
  3. 影响范围与严重程度: 同 floor 6(按选项 A/B/C/D 列于 floor 6)。
  4. 怎么修复的: 本轮未修——等 owner 拍板;默认 = 保持冲突态、不 force-push。
  5. 是否有回归问题: 无——本轮未改分支、未 force-push,PR 分支仍为 79b929bf。
  6. 是否破坏缓存: 不涉及。
  7. 是否改了设计方向: 不涉及(未动)。
  8. 是否引入新配置: 不涉及。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Status (8th re-trigger, floor 36): still unchanged — owner decision on A/B/C/D still pending. No force-push, no branch touched.

Re-verified live this run (not from memory):

This is the 8th re-trigger with no decision. The full analysis (options table, the #6842 g test evidence, 八问) is at floor 6 — not re-posting it here.

To stop the re-trigger loop, please either:

  • Decide A/B/C/D — my recommendation is C/D: keep Afloat's rawGooglePart + plugin.ts coverage, fix the multi-chunk accumulation bug in core.ts (the PR's signature-aware merge), and cherry-pick the PR's test file as the regression test; or
  • Close the PR (owner-only) if you'd rather take Afloat's fix as final and track the multi-chunk gap separately.

八问(状态更新,细节见 floor 6)

  1. 是什么问题: 同 floor 6——Google signed ordinary text is rewritten and loses signatures in tool continuations #6842 两个竞争性修复撞车,属设计决策,非机械冲突。
  2. 什么原因导致的: 同 floor 6——两个修复并行落地,PR 未 rebase 到含 Afloat 修复的 master。
  3. 影响范围与严重程度: 同 floor 6(按选项 A/B/C/D 列于 floor 6)。
  4. 怎么修复的: 本轮未修——等 owner 拍板;默认 = 保持冲突态、不 force-push。
  5. 是否有回归问题: 无——本轮未改分支、未 force-push,PR 分支仍为 79b929bf。
  6. 是否破坏缓存: 不涉及。
  7. 是否改了设计方向: 不涉及(未动)。
  8. 是否引入新配置: 不涉及。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Google signed ordinary text is rewritten and loses signatures in tool continuations

1 participant