Repository navigation
release v0.1.197 - #3130
release v0.1.197#3130
Conversation
📦 Built Package ArtifactBranch: Option A — Install this exact build (recommended)npm install -g billion-context@0.0.3588-pr.3130.23ea3eeThis exact version is immutable and reproducible. The rolling Option B — Download artifact
tar xzf billion-context-pr3130.tgz
npm install -g packageInstall instructions are refreshed on each push. |
(ai: vllm/qwen3.8-27b) 收到,开始审核 release v0.1.197。已确认 PR 形态合规: 这是发版本类 PR,按规范自动触发版本区间回归审计:审查窗口为 v0.1.196..head(216 commits / 178 files / +15412 −976),分段通读全窗 diff——高危域(自更新、dsh、lockfile、网络/proxy 路径)逐行精读,其余按目录分批。结论(八问体例 + 行为变更清单 + 高危域逐项核查结果)稍后回复。 |
…rade visibility (#2960) Review fixes for PR #3130: 4 stale codexCompact scope spots (en/zh prose + env rows) contradicted the updated table row - interception covers explicit forms AND inline auto-compact via SUMMARIZATION_PROMPT detection. Plus one sentence on the storage.jsonWriter row: index-only sessions are invisible to <=0.1.196 binaries until a >=0.1.197 binary runs again (data intact in index.db).
|
(ai: zhipuai-lb/glm-5.3) v0.1.196 → v0.1.197 大 diff 发布评审评审范围:master 自 v0.1.196 以来的全部累计改动(214 commits / 178 files / +15409−973,head 6c78389),四个方向逐一深读(自动更新与缓存 / 压缩内核与会话 / dsh lane / SQLite 存储),并在本地用 bili-gate 对真机构建跑了全链路真流量回归。 1. 这是啥问题?(一句话)没有发现阻塞性回归。发现 2 处用户可见的文档错误(已直接修在本 PR,commit 2. 本地真机回归(bili-gate,master @ 6c78389 构建)——全绿
3. 已直接修在本 PR(小问题直接修,
|
…urce Manual table-row edit in 80ccc93 drifted from the generated block (docs-sync). The sentence now lives in website/config-reference/process.yaml (desc_en/desc_zh) and CONFIGURATION.md is regenerated from it; zh row already matched. No content change beyond em-dash -> hyphen normalization.
|
(ai: zhipuai-lb/glm-5.3) 补记: |
(ai: vllm-qwen/qwen3.8-27b) v0.1.196 → v0.1.197 发布窗审计结论(独立全窗通读)我按 ②b 规则独立完成了整窗审计:v0.1.196..pr-3130 共 216 commits / 178 files,分段读完全部 61 个 src 文件 + scripts + docs(en/zh)+ website + goldens(分批 diff 精读,高危域逐行),没有依赖 PR 描述或提交信息汇总。与楼上并行评审的结论收敛,且我已独立核对其两个文档修复 commit(见下)。以下为我的结构化八问。 八问1. 是什么问题:这不是缺陷修复 PR,是 v0.1.196 以来约 129 个已合并变更的发版 PR(直推 master 被分支保护挡下后的自动回退通道)。审计对象是「这一批东西合起来会不会互相打架、有没有破坏既有行为」。 2. 什么原因导致的:不涉及(发版)。窗口内最大的三个机制族:统一存储推进(#2960 R2.1/R2.2/②③⑤ + Phase 3 jsonWriter)、会话身份与对账重构(#2944 v2 resolver、#2915 顺序中心对账、#2815 声明式 marker 引擎)、nudge×self-heal 死锁修复(#3004)。 3. 影响范围与严重程度:所有升级到 0.1.197 的安装。其中两个是生产默认翻转,每个安装都吃到:(a) 4. 怎么修复的:不涉及(发版)。 5. 是否有回归问题:本版本引入回归:无。 逐项高危域核查结果:
6. 是否破坏缓存:否。systemToUser 锚点机制未动;marker 引擎四条 wire walk 全部显式跳过签名思考块(thinking/redacted_thinking/thoughtSignature/thought:true)——KDD#10/#1960 不变量在位;codex prompt-form forge 只产出最后一条 assistant 消息项(符合 codex 摘要采纳语义);cache-seam 的 InboundSlot/client-insert 归因纯诊断面、内存有界(INBOUND_MSG_CAP=8192),不改任何出站字节。 7. 是否改了设计方向:无转向。#2915/#2944/#2815 都是把已写入 AGENTS.md 的 owner 规则(KDD#11 顺序为中心、身份维度并集)落到实现;rollout 灰度机制(#3062)注册表为空上线,零行为变化,后续特性各自带自己的 PR 入场。 8. 是否引入了新的配置:是,4 组,全部已在 CONFIGURATION.md + zh-CN + website yaml 三处同步文档化(我逐条核对过,「document or don't ship」满足):
对并行评审的独立核对
CI 状态(发帖时点,head
|
(ai: vllm-qwen/qwen3.8-27b) CI 全量确认:head 合并门禁已清:本版本引入回归 0 处,PR 形状合规,CI 全绿,可由 owner 合并。 楼上 §4 的三个 owner 决策项(update.ts no-op 规则追认、两个默认翻转是否补 severe release-notes 条目、wt-* 通道补 typecheck 门)不阻塞合并,留待拍板;跟进缺陷已在 #3133 / #3134 / #3135 跟踪。 |
|
有新修复进入 麻烦重新review |
|
(ai: zhipuai-lb/glm-5.3) Gate 回归(合并前重跑,更新后的 release head)release head 已并 master(#3139 + #3140 含 r2 修复),按新 head
OVERALL: GREEN(gate-local exit=0,docker-lanes exit=0)。注意这次跑的 head 包含 #3140 的存储修复(采样校验 1/20 + writerOff 合并 + microtask 错误边界),gate 的真流量轮次默认走 shadow 双写,存储写路径同时被真实压过一遍。 无阻塞,可以合并。 |
(ai: vllm-qwen/qwen3.8-27b) 增量复审结论(新 head
|
(1) jsonWriter=off: the save tick on which the sidecar hits its 3rd
consecutive failure previously armed neither the JSON write nor a
successful index write - that turn was memory-only until the next
save. The writerOff branches now re-arm the legacy JSON write in
the same tick when the disable flip happens mid-build; flushSync
returns the real file-write result instead of an unconditional true.
(2) BILI_PERSIST=0 no longer reports a permanent 'storage degraded'
banner with the bogus 'no SQLite engine available' reason:
persistence-disabled is recorded as an intentional opt-out, exactly
like the encryption block (new recordPersistDisabled seam).
(3) The per-file / live-session storageMode stamp now reflects the mode
that ACTUALLY opened, frozen once at store construction (stampMode):
a full request whose sidecar failed to open stamps 'legacy' instead
of claiming an index.db copy that does not exist. The requested-mode
resolver effectiveFileStorageMode() is gone - every buildRecord call
site passes the instance stamp explicitly.
(4) Digest drift is wire-visible: verifyOk/verifyDrift ride the
/__bili/status + overview storage payload and the diagnostics blob.
A drifted session in dual-write full mode serves the legacy-file
view again until a clean ingest resets it; under jsonWriter=off the
index view stays (sole copy), drift surfaced via stats.
(5) An unloadable-but-complete cached engine copy is removed and
re-downloaded once (sha512-pinned, so safe) instead of returning
'cache-hit' and degrading forever; a freshly downloaded copy that
fails to load arms the 6h cooldown marker instead of re-failing on
every boot.
(6) Bounded boot runs the same index union under its own limits
(count + decoded-byte budget from summary rows), so jsonWriter=off
index-only sessions are visible under warm-start limits; lazy reload
re-imports the envelope when the JSON copy is newer than the index
row before serving the unified view (mirrors buildForLoadAll).
Docs: CONFIGURATION en/zh jsonWriter row - fail-safe wording updated to
the same-tick guarantee, plus a WAL durability note for the off lane
(synchronous=NORMAL: crash-safe, power loss can drop the last checkpoint
interval). No new config surface.
(1) jsonWriter=off: the save tick on which the sidecar hits its 3rd
consecutive failure previously armed neither the JSON write nor a
successful index write - that turn was memory-only until the next
save. The writerOff branches now re-arm the legacy JSON write in
the same tick when the disable flip happens mid-build; flushSync
returns the real file-write result instead of an unconditional true.
(2) BILI_PERSIST=0 no longer reports a permanent 'storage degraded'
banner with the bogus 'no SQLite engine available' reason:
persistence-disabled is recorded as an intentional opt-out, exactly
like the encryption block (new recordPersistDisabled seam).
(3) The per-file / live-session storageMode stamp now reflects the mode
that ACTUALLY opened, frozen once at store construction (stampMode):
a full request whose sidecar failed to open stamps 'legacy' instead
of claiming an index.db copy that does not exist. The requested-mode
resolver effectiveFileStorageMode() is gone - every buildRecord call
site passes the instance stamp explicitly.
(4) Digest drift is wire-visible: verifyOk/verifyDrift ride the
/__bili/status + overview storage payload and the diagnostics blob.
A drifted session in dual-write full mode serves the legacy-file
view again until a clean ingest resets it; under jsonWriter=off the
index view stays (sole copy), drift surfaced via stats.
(5) An unloadable-but-complete cached engine copy is removed and
re-downloaded once (sha512-pinned, so safe) instead of returning
'cache-hit' and degrading forever; a freshly downloaded copy that
fails to load arms the 6h cooldown marker instead of re-failing on
every boot.
(6) Bounded boot runs the same index union under its own limits
(count + decoded-byte budget from summary rows), so jsonWriter=off
index-only sessions are visible under warm-start limits; lazy reload
re-imports the envelope when the JSON copy is newer than the index
row before serving the unified view (mirrors buildForLoadAll).
Docs: CONFIGURATION en/zh jsonWriter row - fail-safe wording updated to
the same-tick guarantee, plus a WAL durability note for the off lane
(synchronous=NORMAL: crash-safe, power loss can drop the last checkpoint
interval). No new config surface.
Auto-fallback from the one-click release workflow: direct push to master was blocked by branch protection. Merging this PR publishes via the standard release.yml flow.
Changes since v0.1.196