Repository navigation
Conversation
70757bb to
b20bc21
Compare
(ai: vllm-qwen/qwen3.8-27b) 查重结论:重复 —— 与 #1701(提案)/ #1708(PR)是同一需求:让压缩回执(📦/❌ 行)对客户端不可见、对模型保留。#1708 先到(2026-09-29),方案经 owner 拍板,实现与测试完整,现处待审/待合并;#4934 后到(2026-10-26)。按规则打 标签: duplicate, scope/wiring, p3(功能 PR,与 #1701/#1708 的回执投递模式工作重复;请求管道域;功能级影响) 对 #4934 的独立审核发现(非引用 #1708 结论)逐文件读了 PR diff,并对 master 上每一个 marker 发射点做了代码核对:
为什么不按外部 PR 规则直接重构补全这个 PR 的问题不在代码质量,而在机制选择本身:#1708 的 config-surface 章节明确记录了 owner 拍板的决定——不给 compress. 新增兄弟字段*(visibilityMarkers 三态演进保持一个关注点一个旋钮)。把 #4934 修完就是同一关注点的第二套机制,与 #1708 正面冲突。所以正确处置是并入 #1708,不是修补。 已核实 #1708 头分支恰好补齐了 #4934 漏掉的点位: 给作者的建议等 #1708 合并后使用 结构化八问
|
… PR ranxianglei#4934 Align-refactor (review of ranxianglei#4935): this PR is scoped to the bare-announcement terminal-turn retry (ranxianglei#3905 symptom), but it also carried the LoopCtx streamMarkers field and the emitMarker consumption gate that belong to the streamMarkers setting feature. That feature has its own PR (ranxianglei#4934, which wires the config surface but lacks these two hunks) — keeping them here makes ranxianglei#4934 unable to compile standalone and creates a merge-order dependency between two independent features. The hunks move to ranxianglei#4934 so each PR is a closed loop.
Align-refactor (review of ranxianglei#4935): this PR defines the streamMarkers field on LoopCtx and wires the config surface end-to-end, but nothing ever reads the field — the emission gate that honors it was shipped in PR ranxianglei#4935 (the announcement-retry PR), which makes this setting a dead knob on its own and couples the two PRs into a merge-order dependency. Move the consumption gate (and its failure-mode comment) here from ranxianglei#4935 so each PR is a closed loop. Behavior with the default (unset) is unchanged: markers still stream.
…rs (ranxianglei#2107) Align-refactor (review of ranxianglei#4935): the PR's website/config-reference/compress.yaml seed edit was committed without re-running tools/gen-config-docs.mjs, so the ranxianglei#2107 docs-sync drift guard goes red on this head. Regenerate the four derived artifacts (CONFIGURATION.md / CONFIGURATION.zh-CN.md / website/main.js / website/docs/index.html) — content is exactly the new streamMarkers row plus its BILI_STREAM_MARKERS env index entry.
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) merge-guard r1 处理完成(PR #4938,head 分支
|
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) merge-guard r2 处理完成(PR #4938)——本轮为过期触发,无需任何操作:
|
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) merge-guard r3 处理完成(PR #4938)——又一次过期触发,无需任何操作:
|
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) merge-guard r4 处理完成(PR #4938)——第 4 次过期触发,无需任何操作:
提醒:同一静态 state 已连续 4 次触发 behind 通知,guard 的 behind 判定疑似基于旧快照而非实时祖先关系,建议运维侧核查触发逻辑。 |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) merge-guard r5 处理完成(PR #4938)——第 5 次过期触发,无需任何操作:
提醒:同一静态 state 已连续 5 次触发 behind 通知,guard 的 behind 判定疑似基于旧快照而非实时祖先关系,建议运维侧核查触发逻辑。 |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) merge-guard r6 处理完成(PR #4938)——第 6 次过期触发,无需任何操作:
提醒:同一静态 state 已连续 6 次触发 behind 通知,guard 的 behind 判定疑似基于旧快照而非实时祖先关系,建议运维侧核查触发逻辑。 |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) merge-guard r7 处理完成(PR #4938)——第 7 次过期触发,无需任何操作:
提醒:同一静态 state 已连续 7 次触发 behind 通知,guard 的 behind 判定疑似基于旧快照而非实时祖先关系,建议运维侧核查触发逻辑。 |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) merge-guard r8 处理完成(PR #4938)——第 8 次过期触发,无需任何操作:
提醒:同一静态 state 已连续 8 次触发 behind 通知,guard 的 behind 判定疑似基于旧快照而非实时祖先关系,建议运维侧核查触发逻辑。 |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) merge-guard r9 处理完成(PR #4938)——第 9 次过期触发,无需任何操作:
提醒:同一静态 state 已连续 9 次触发 behind 通知,guard 的 behind 判定疑似基于旧快照而非实时祖先关系,建议运维侧核查触发逻辑。 |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) merge-guard r11 处理完成(PR #4938)——第 11 次过期触发,无需任何操作:
提醒:同一静态 state 已连续 11 次触发 behind 通知,guard 的 behind 判定疑似基于旧快照而非实时祖先关系,建议运维侧核查触发逻辑。 |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) merge-guard r12 处理完成(PR #4938)——第 12 次过期触发,无需任何操作:master tip |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) merge-guard r13 处理完成(PR #4938)——第 13 次过期触发,无需任何操作:master tip |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) Merge-guard 处理完成(PR #4938,head 分支
备注:guard 消息里的标题("streamMarkers setting")与实际 PR #4938 标题不符,按 PR 编号为准处理。 |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) Merge-guard 二次触发(重复 "behind" 信号):fresh fetch 后重新验证,PR #4938 head |
|
(ai: vllm-qwen/qwen3.8-27b) Merge-guard sync: merged |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) Merge-guard 第三次触发,这次查清了根因:guard 的 PR 编号错了。消息里的标题("feat(compress): add streamMarkers setting…(#3906)")实际对应 PR #4934(外部贡献者 fathonix 的 fork 分支),而不是 #4938。#4938(zero-fold calibrated fit)本身不 behind、无需动。 已对真正的目标 PR #4934 执行同步:
待 owner 路由:merge-guard 管道的编号映射 bug(标题与编号错位 4 位,连续三次触发都带错号)不属于 billion-context 仓库代码,未在该仓立项;请指示在哪个项目跟进。 |
|
(ai: vllm-qwen/qwen3.8-27b) Merge-guard sync: merged |
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) Merge-guard 第四次触发,这次是真 behind:master 前进到
运维备注:① 共享 git clone 里发现一个损坏的本地 ref( |
… PR ranxianglei#4934 Align-refactor (review of ranxianglei#4935): this PR is scoped to the bare-announcement terminal-turn retry (ranxianglei#3905 symptom), but it also carried the LoopCtx streamMarkers field and the emitMarker consumption gate that belong to the streamMarkers setting feature. That feature has its own PR (ranxianglei#4934, which wires the config surface but lacks these two hunks) — keeping them here makes between two independent features. The hunks move to ranxianglei#4934 so each PR is a closed loop. (cherry picked from commit fa60f8f)
PR 1: feat(compress): add streamMarkers setting to suppress streaming receipts (#3906)
Problem / Background
When streaming responses in client UI environments (e.g. OpenCode, Pi, Cursor), Billion-Context emits streaming visibility markers (
📦 [ACP] Compressed ...,❌ [ACP] ... FAILED) directly into response deltas so users can see compression actions in real-time.However, in certain automated agent workflows, headless script harnesses, or strictly formatted markdown interfaces, injecting marker text deltas directly into the client stream can interfere with downstream parsers, pollute test suites, or create visual noise. Prior to this change, while
visibilityMarkerscontrols overall marker generation, there was no granular toggle to suppress streaming receipt emission while keeping standard loop execution intact.Changes
src/config.ts:streamMarkers?: booleanto configuration schema andcompress.streamMarkersconfiguration key.BILI_STREAM_MARKERS(defaulting totrueto preserve backward compatibility).src/compress-settings.ts:streamMarkersthrough compression options resolution.src/compress-loop-responses.ts:emitStreamingMarker/ receipt chunk creation whenstreamMarkers === false.src/server/relay.ts:streamMarkersflag throughrunCompressLoopcontext.Verification
compress.streamMarkers: falseandBILI_STREAM_MARKERS=0.streamMarkers: trueemits receipts as before).streamMarkers: falsethat compression completes successfully in the background without injecting stream receipt deltas into client responses.