Repository navigation
[PR] chore: empty PR (no changes) - #4725
Merged
Merged
Conversation
ranxianglei
changed the title
ranxianglei
added
the
ci:full
Contributor
📦 Built Package ArtifactBranch: Latest build — install this exact versionnpm install -g [email protected]This exact version is immutable and reproducible. The rolling Recent builds (commit → version)
Download artifact
tar xzf billion-context-pr4725.tgz
npm install -g packageInstall instructions are refreshed on each push. |
Owner
Author
(ai: vllm-qwen/qwen3.8-27b) 查重: 不重复(open + closed 全量检索 "empty PR / allow-empty / no changes" 无相关条目)。 已对真实 PR #4725 逐项核实(GitHub API):
八问:
判定:PR 确认为空,我这边无任何动作可做;按规则我不合并(merge 仅人工)。CI 跑完后由你自行决定合并或关闭。 |
Owner
Author
|
对这个进行发版review |
Owner
Author
(ai: vllm-qwen/qwen3.8-27b) 发版 review 完成。先说方向判定,再给实质审计。 ① 方向关:这个 PR 不是发版 PR对照发版类 PR 的四个识别特征,逐条不满足:
分支名 ② 版本区间回归审计(master vs 最近发布)窗口:v0.1.206(master 谱系上最后一个 release tag)..origin/master(bb7c249)= 323 commits,359 文件,+35979/−2293。读法: 本版本引入回归:未发现。 高危域逐项点名核查结果:
跨车道漂移核查(v0.1.207/v0.1.208 从 bugfix/carve-out 车道切出,tag 不在 master 谱系上):两个车道的全部实质修复(#4390 行身份/historian tombstone、#4444 node-upgrade lever、#4506 media refs、#4677 v8 迁移流式化+进度日志、#4215 诊断、#4585 de-export)均已在 master 内容级确认存在;无缺失修复。车道独有的 jq/PMODE 三个 commit 是旧版 release-bugfix.yml 的车道侧 CI 修复,master 侧该机制已被重写移除,属作废而非缺失。 重大行为变更清单(均为有意变更,旧值→新值+原因):
次要流程缺口(非阻塞):master 的 release-notes 账本停在 0.1.180(opt-in 制度下正常),但 0.1.207/0.1.208 的 v8 大存储损坏 critical 条目只存在于车道分支,未回流 master 账本。 结构化八问
收工判定:PR #4725 本身可安全合入或关闭(合入不触发任何东西;按规则我不代操作)。若目的是把 master 的内容发出去:master 当前 CI 绿(bb7c249 最近三次 run 全 success),具备发版条件,建议批准 v0.1.209——你拍板说一声「发版 v0.1.209」我就走 §5 流程切 release PR。 |
Owner
Author
|
(ai: zhipuai-lb/glm-5.3) v0.1.208..master 发版前深度审核(362 commits / 349 files / +34046 −2464)审核方式:4 个分域代码审查(请求管线/wire、dsh+自更新+CI 发版、前缀缓存+会话身份+存储、配置默认值)+ 本地真模型门禁 gate:local(A 压缩 / B 切模型 / U 更新)+ master head 安装验证(全局 + dsh default profile)。 结论一句话运行时压缩/更新核心代码基本干净( gate:local 结果(真模型,候选 0.0.900-gate.bb7c249 = master bb7c249)
第 1 轮两条红的根因链(gate scratch
P1(发版阻塞级)
P2(发版前应修/应写进 release notes)
P3(记录在案)
默认配置变更(升级用户会感知的全部)
新增 env 10 个(BILI_PI_INLINE、BILI_OPENCODE_INLINE、BILI_INLINE_DEBUG、BILI_CODEX_STATE_MODE、BILI_THROTTLE_429 及 3 个旋钮、BILI_ROLLOUT_PLUGIN_LANE_SERVER_HISTORY),en+zh 文档均已核同步;无删除/改名。kernel 默认面(golden pin、flat 50K nudge)零改动,kernel 0.0.161→0.0.163 版本纪律合规。 验证干净面(重点核对过)
安装状态(本 PR head = master bb7c249)
建议的发版前置清单
审核为只读;本 PR 自身无代码变更。各发现建议按 §4 规则拆独立 issue/PR 跟进。 |
…e ledger alignment (#4599 plan A)
Owner
Author
(ai: vllm-qwen/qwen3.8-27b) CI 自动修复(第 1/2 次)— 继承性红,已收敛 ci-fixes判定:本 PR 零文件改动,8 个红 shard 没有一项是本 PR 引入的 —— 全部是 master 继承性红。 逐 job 核对了 run 40202012280 的 TAP 输出(ubuntu 24/3 = 12 fail、windows 22/4 = 7 fail),失败项与 #4599 总控名册逐条对上:
动作(按 CI 失败协调政策,不在自己分支重复修 master 级失败):把哨兵维护的 ci-fixes 头 结构化八问
残留风险:plugin-fork 那一项在 #4652 合入前可能继续红 —— 若重跑后仅此一项红,属预期,待 #4652 收敛即可,不需要本 PR 侧任何动作;届时按 attempt 2/2 处置。 另:线程里新贴的 v0.1.208..master 发版前深度审核(P1/P2/P3 清单)超出本 CI 自动修复车道的范围,按其自身结论各发现应拆独立 issue/PR 跟进,此处不扩scope。 |
Owner
Author
|
(ai: zhipuai-lb/glm-5.3) 补充修正(残余 6 文件复审的完整结果,修正上条评论中两处严重度 + 1 个新发现)上条评论发布后,残余文件(plugin-render / cache-ledger / util / fold-decomposition-lookup / msgid-algebra / uncovered)的窄范围复审跑完了,三处需要修正/补充: 1. plugin-render 2. fold-decomposition-lookup 的 splice 索引失步(上条记 P3,应升 P2) —— 具体 trace:行按 seq 交错 3. 新 P3:cache-ledger.ts:1925-1927 settledInBody 基线耦合 —— settle 侧现在要求 同时解除两个我方软性疑虑(有据为 clean):uncovered.ts 的共享-clone 双 processTurn 窗口差分是稳的 —— kernel 上条评论的其余结论(含 P1×2、其余 P2、默认值表、验证干净面)不变。 |
This was referenced Oct 25, 2026
Owner
Author
|
(ai: zhipuai-lb/glm-5.3) v0.1.208..master 补充审核(第 3 轮)——针对前两轮声明的薄弱面审核方式:前两轮(qwen 发版 review + glm 深度审核)各自标注了未覆盖面;本轮补的是 ①qwen 明确写「未逐行精读」的 去重声明前轮已发现的 P1×2(release-bugfix jq / manifest 漂移)、P2×10(#4288 撤销、fork 截断、429 默认开、文档失真、一键通道、carve-out 缺门、CCR 错配、relay 死代码、plugin-render 锚、splice 失步)、P3×若干——本轮不重复列。以下全部是新增发现。 新发现 N1(P2,已实测复现):不可解析的块区间落库为
|
This was referenced Oct 25, 2026
Open
The first resolve step calls jq 4 times, but the workflow carries no jq install step: on the 2026-10 runner images (observed jq missing, runs 40193290049 et al exit-127) the emergency carve-out channel dies at ignition exactly when it is needed. v0.1.208 shipped this bootstrap (e7b0a1e2); the consolidation trains that rebuilt master kept the older workflow file and dropped it. Port the step verbatim from v0.1.208.
The four plugin manifests (.agents/plugins/marketplace.json, .claude-plugin/marketplace.json + .claude-plugin/plugin.json, plugins/codex/.codex-plugin/plugin.json) sat at 0.1.205 while package.json is 0.1.207 — the #4449 gate (node scripts/sync-codex-plugin-version.mjs --check) fails on master, so any release branch cut now would be rejected by release.yml's pre-tag check, and the Codex/Claude marketplaces have not offered updates since v0.1.206. Run the sync script (no code changes).
…ndow After a model switch shrinks the context window, the round-boundary exit judged on decisionTokens, which is floored by the session usage baseline. That baseline was measured under the prior wider-window model, so it sits above the new window and the exit never trips on an already-fitting payload. The summarization ladder then burns every round until the client times out and the session wedges (#4288). Judge the round-top exit on result.payloadEstimate instead, the floorless quantity the caller forward gate already trusts. The floored condition implies this one, so folding never increases versus the old behavior; only the redundant ladder-burning on fitting payloads is removed. The intra-walk exits stay floored (conservative mid-walk). Adds a regression test: model switch, stale high usage baseline, payload fits, and preflight forwards without spending a single fold call. (cherry picked from commit 6a9257a)
added 2 commits
October 26, 2026 00:49
…message exceeds the budget
A pair range [tiny tool-call + oversized tool-result] under a window whose
0.6 chunk budget is smaller than the result split into [call-only fragment]
[result]; the fragment's preview is atomically rejected by the kernel's
minCompressRange gate ('Total compressible content too small (64 chars …,
min 5000)') — reproduced live in gate:local (gate-1025, '64 chars'/'77
chars' rejections burning a preview per round), and on the batch lane
(one span per range) the range never folds at all. Apply the min-chars
floor in both baseline regimes: a sub-floor chunk merges forward with the
oversized message into one coherent span instead of standing alone.
Fixes #4740
unifiedLiveRowCount counts live rows, but cloneSessionRows wants the inclusive throughSeq and seqs anchor at MAX(seq)+1 — tombstones keep their seq, so any retired row makes count < lastLiveSeq + 1 and count - 1 silently dropped the newest live rows from the cloned child (the fork receipt then serves a truncated ledger while the kernel view kept every number). unifiedLastLiveSeq (MAX(seq) over live rows) is now the throughSeq source across store/mysql/backend/shadow/persist; unifiedLiveRowCount stays on the shadow surface for its other callers. Regression: tests/issue4730-fork-live-seq.test.ts (retire a middle row, clone must still carry the newest live row; count-1 control documents the old truncation).
…lane, missing-manifest gate, autoFold docs Three unfixed items from the #4725 review ledger (#4757 landed separately via the ci-fixes sweep): - release-bugfix.yml lost the #4449 manifest --check when the v3 lane rewrite landed (v0.1.208 ran it before Publish guard; master's rewrite dropped it). Ported back with the #4576 pre-#4449 lane existence guard, gated on the resolve step's actual proceed output (this variant has no path output). - scripts/sync-codex-plugin-version.mjs silently SKIPPED a missing manifest file — in --check mode a missing file greened the gate, neutering the release guard. A missing target is now a hard error naming the file; the four manifests are a fixed contract. Regression test added (RED on the old skip). - CONFIGURATION.md en+zh: document the autoFold benefit gate (#3444: growth folds arm only on ≥4096 tokens over the effective target; sub-4096 saving arms the backoff; image-dominant exempt) and the measured floor (irreducible baseline raises an unreachable target, never above the window). Both languages in sync. Sync tests 5/5, typecheck clean.
Owner
Author
|
(ai: zhipuai-lb/glm-5.3) 审核清单收尾:剩余未修项已处理(ec3576b)上一轮盘点(#4725 (comment) 之后)中「可以直接修」的四项:
验证:sync 测试 5/5、issue1195 A/B/C + #2777 + #4288 三钉全绿(rebase 到 master 合并点 4e0c773 之后)、typecheck 0 错。 仍开放的项(均需 owner 拍板或另行处理,非代码可直接修):429 治理器默认开启的 severe release-notes 条目(发版 PR 内)、gate 脚本对 #4288 fit-forward 的适配、inline 车道 429 治理/熔断旁路(P3,设计取舍)、decompress imageRef 的 restampSuperseded 守卫(P3,两位审核员结论相反)、reference/release.md v3 章节与 release-manual RETIRED 标注的文档面统一、master 0.1.207≠npm 0.1.208(下次发版自愈)。 |
Owner
Author
[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) 冲突车道状态确认(对最新 head 复核)我负责的 #4725 基线冲突(auto-fix lane)已解决并落地;此轮针对分支后续推进做了 ground-truth 复核,结论如下。 我之前做的(已完成):把当时 本轮复核(分支作者又推了 round-3 修复 + 自己又 merge 了一次 master):
结果:冲突维度保持绿,我这边无需再改动;是否合入归 owner,等 CI 转绿即可。若状态再次变 |
This was referenced Oct 26, 2026
ranxianglei
pushed a commit
that referenced
this pull request
Oct 26, 2026
…ores (#4599) maybeRestampToLedger called contentStoreOf() speculatively to re-key the CCR content store after ref restamping (#4733). contentStoreOf is a get-or-MATERIALIZE helper: for a session that has never had a store it assigns a fresh empty envelope, so every full-mode cold load started persisting/serving contentStore={version:1,byHash:{},byRef:{}} where the legacy path keeps undefined. The storage-unified parity pin (full serve = legacy serve + ledger alignment) digests contentStore and went red on both FULL-mode read subtests at a4a17cb (#4733, landed via #4725). Guard the re-key with if (session.contentStore): storeless sessions have nothing to re-key and keep their pre-#4733 shape; sessions WITH a store are re-keyed exactly as #4733 intended (the retrieveByRef misattribution fix stays intact). Verified locally (targeted): tests/storage-unified.test.ts 14 pass/0 fail (was 12/2); tests/issue4733-ccr-restamp.test.ts 2/0; kernel/tests/ccr-restamp.test.ts 3/0; typecheck clean. Full-suite verdict is the CI test lanes on the pushed head.
Contributor
📦 Merged Build PublishedThis merge to master was published as npm install -g [email protected] # this exact build
npm install -g billion-context@master # always the newest merged build
|
This was referenced Oct 26, 2026
ranxianglei
pushed a commit
that referenced
this pull request
Oct 26, 2026
…erdict; #847/#1492B contracts reconciled to the #4288 world Both files were red since the #4725 merge (verified at 931c66a in an isolated worktree): the re-landed #4288 exit fired at round top, but the final verdict block still stamped an exhausted failure whenever currentTokens — floored by the same stale baseline the exit exists to bypass — read over-limit. A fit-forward walk then reported 'no viable range could be compressed' while the caller forwarded the fitting payload anyway: the marathon was prevented, but the reported verdict lied. - preflight.ts: fitExitForward flag — a round-top fit exit suppresses the final failure stamp (aborts still win). - #1492 B: contract updated to the #4288 doctrine — a usage-grade baseline above the window on a FITTING payload forwards; a usage baseline above the limit cannot have been measured under this window regime, and a real overflow arms the one-shot rescue (#1110/#987) on rejection. - #847: the honest-exhaustion pin now covers its actual post-#4288 meaning — the fitting sub-gate payload exits fit-forward with no verdict; the exhaustion path stays for payloads that genuinely exceed the target. issue847 4/4, issue1492 3/3, issue4288 3/3, issue1195 4/4, preflight family 21/21, 4797 family 4/4, typecheck clean.
This was referenced Oct 26, 2026
yanhuojiyusheng
pushed a commit
to yanhuojiyusheng/billion-context
that referenced
this pull request
Oct 26, 2026
… declaration for lockfile-only churn kernel/package-lock.json is generated state, not kernel source. Agents' test prep (npm install inside kernel/, sweep-bot merges of stale lockfiles) churns it without any intent to change the kernel — 8 PRs (ranxianglei#4823 ranxianglei#4808 ranxianglei#4802 ranxianglei#4800 ranxianglei#4799 ranxianglei#4794 ranxianglei#4790 ranxianglei#4785) tripped the human gate on lockfile-only diffs, and the previous stopgap (bots appending kernel-change: declarations to PR bodies) was correctly rejected by the owner as bypassing the approval step. The human gate exists to protect kernel SOURCE. Rule 1 now ignores kernel/package-lock.json when deciding whether a PR 'touches kernel/'; real source/build-config changes still require the declaration + owner approval, and lockfile CONSISTENCY stays enforced by the separate lock-sync job (which fails on genuine drift — the ranxianglei#4725 merge red being today's live proof).
Merged
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Empty PR as requested — one
--allow-emptycommit, zero file changes. Nothing to review or merge.