Skip to content

[PR] fix: converge inflated out-of-band dsh profile/desktop copies instead of freezing (#3160) - #3164

Closed
ranxianglei wants to merge 8 commits into
masterfrom
2026-10-14_dsh-oob-copy-self-heal
Closed

ranxianglei wants to merge 8 commits into
masterfrom
2026-10-14_dsh-oob-copy-self-heal

Conversation

@ranxianglei

@ranxianglei ranxianglei commented Oct 14, 2026 •

Copy link
Copy Markdown
Owner

What & why

#3160: a dsh profile/desktop copy whose version was hand-inflated above the registry (real case: package.json edited to 1.2.9 while registry latest is 0.1.197) froze the #1196 self-heal forever — every self-heal gate compared one-directionally (local < authority ⇒ converge, local ≥ authority ⇒ "up to date", skip), so an inflated copy read as "newer" and was never touched; and nothing carried a provenance check, so the copy loaded silently with no warning.

This makes host-managed dsh lanes treat the registry as authoritative for the equality of the installed copy (same principle as KDD "the bundled registry snapshot must not outrank the live registry") instead of trusting any on-disk version ≥ the registry. An ahead copy is no longer read as "up to date" — it is surfaced loudly (self-heal log + bili doctor) so it can never again be a silent, permanent wedge.

Owner-feedback refinements (this head): two rounds tightened when bili may act on an ahead copy:

  1. converging back must never silently downgrade a genuinely-published build that merely leads the tracked tag → the ahead path consults the full registry packument (fetchPublishedVersions).
  2. going further — being absent from the public packument is not safe proof of tampering: a user who maintains their own build at a higher version is also absent from the public registry, and auto-replacing that copy would overwrite their maintenance. So an ahead registry-form copy is now never auto-downgraded in this PR — only surfaced loudly and left untouched. The safe auto-quarantine (replace only a copy whose existence no install record explains) is deliberately deferred to [Detection] No signal when dsh profile manifest/lockfile/on-disk copy are in three-way conflict (follow-up to #3160, #3169 closed without merge) | [Detection] dsh profile 的 manifest/lockfile/盘上副本三方矛盾时无任何信号(#3160 后续,#3169 已关闭未合并) #3224, gated on positive unexplained-evidence rather than packument-absence alone.

Changes

  • fetchPublishedVersions (src/update.ts) — GETs the npm packument and returns the set of all published versions; undefined on any failure / unexpected shape. Used here purely for diagnostic wording (telling the operator which situation they're seeing: published-other-tag vs unverifiable vs absent); it does not gate any downgrade.
  • refreshDshDesktopCopy (src/update.ts) — skip only when exactly in step. Any ahead registry-form desktop copy → loud warn (naming the situation) + left untouched; no auto-downgrade, no tarball reinstall. A locally-pinned copy falls through so its siblings refresh.
  • refreshDshProfileCopy (src/update.ts) — same. A locally-pinned running copy stays a deliberate dev pin (info log; sibling registry-pinned profiles still refreshed via dsh's plugin channel). A registry-form ahead copy → loud warn + left untouched.
  • bili doctor (src/doctor.ts) — dshOutOfBandCopyWarning advisory surfaces any registry-form profile copy ahead of the tracked latest, worded to name both possibilities (deliberate other-tag/local build vs hand-placed/tampered) and point at the fix (pin via link:/file: or a private registry, or remove it); verdict unchanged, mirroring dshBundleShadowWarning.
  • Regression tests: the frozen inflated-copy case (profile, desktop) is now asserted left untouched + warned, never auto-downgraded; a published-but-ahead case likewise; the local-pin case asserts siblings still refresh while the pin stays put.

Behavior change (disclosed)

Case Old New
dsh copy local > registry-latest, version absent from registry (hand-placed fake) silent no-op ("up to date") — frozen forever loud warn + left untouched (never auto-downgraded)
dsh copy local > registry-latest, version present in registry (other-tag build) OR packument unverifiable silent no-op — frozen forever loud warn + left untouched (never auto-downgraded)
local == registry silent silent (unchanged)
local < registry converge up converge up (unchanged)
link: / file: / git+ pin untouched untouched (unchanged; siblings still refresh)

Net effect: the #3160 wedge can no longer be silent or read as "up to date" — it always yields a loud warning plus a bili doctor finding. What this PR deliberately does not do is auto-replace it: replacing an ahead copy on the strength of "not in the public registry" would clobber a legitimate self-maintained build, so that action waits for #3224's stronger evidence. A user protecting their own higher version should pin it (link:/file: or a private registry) — pinned copies are never touched.

Scope notes

  • Gate 3 (staleDshProfileCopies / convergeDshProfileBundles, dsh boot crash on mixed copies: global >=0.1.176 bundle yml (bare name) + profile copy <=0.1.175 (exports["."]=CLI) — window opened by manual npm i -g, self-heal blocked by the crash itself #1803) deliberately unchanged: its mandate is the older-copy boot-crash. Ahead-copies are surfaced by the self-heal gates (from the copy's own process) plus the doctor advisory (offline, covering dormant profiles too).
  • When the running copy is ahead we early-return, so sibling profiles aren't opportunistically refreshed in that same cycle (they still refresh on the next in-step cycle); this trades one opportunistic pass for the guarantee that no build — real or not — is downgraded.
  • Config surface: none — no new config fields, env vars, or CLI flags.
  • Gating: touches src/update.ts (load-bearing infra — human review required, AGENTS.md §7.4). Shipping additionally requires the §5 no-op validation release first; this PR is ready to review now and sequences behind that release.

Closes #3160
Related: #3224 (auto-quarantine of provably-unexplained copies — the follow-up that safely restores automatic replacement)

… of freezing (#3160)

A hand-placed copy whose version is inflated above the registry (e.g. 1.2.9 vs
registry 0.1.197) froze the #1196 self-heal forever: every gate compared
one-directionally (local < authority => converge; local >= authority => "up to
date", skip) and none carried a provenance check.

- refreshDshDesktopCopy: skip only when exactly in step; an ahead/out-of-band
  registry-form copy is loudly warned and converged back through the verified
  registry tarball (the installer also re-syncs the profile manifest spec, #2894).
- refreshDshProfileCopy: same divergence classification; a locally-pinned running
  copy stays a deliberate dev pin (info, siblings still refreshed), a registry-form
  ahead copy is warned and converged back through dsh's plugin channel.
- bili doctor: new dshOutOfBandCopyWarning advisory surfaces any registry-form
  profile copy ahead of the registry latest (it reads as "ok" under laneVerdict);
  verdict unchanged, mirroring dshBundleShadowWarning.

Regression tests reproduce the frozen inflated-copy case and assert it now
converges + warns (profile, desktop, and doctor lanes).

Fixes #3160
@github-actions

github-actions Bot commented Oct 14, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-10-14_dsh-oob-copy-self-heal (ac562af)

Option A — Install this exact build (recommended)

npm install -g billion-context@0.0.3803-pr.3164.504e95e

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

Option B — Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr3164.tgz
npm install -g package

Install instructions are refreshed on each push.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

独立审核完成:在 PR head 上逐段通读了完整 diff,并独立核对了相关既有代码路径(gate 逻辑、收敛通道、doctor verdict、backoff),未复用 issue 侧结论。Review 发现 3 个小问题,已直接修在分支上(3 个带 #3164 review 标注的 commit,push 至 7f9243201):

  1. readDiskVersion(src/update.ts:778)无 typeof 守卫——package.json 的 version 非字符串时(如写成数字,恰是 Profile 副本自愈可被本地高版本号永久卡死,且 out-of-band 副本无来源校验(#1196 变体) #3160 篡改形态之一),下游 parseSemVer 的 .match 会抛 TypeError,整个自愈周期被 catch 吞掉、以原始报错替代设计好的 warn+收敛。属既有暴露(旧代码同样无守卫,非本 PR 引入),但正落在本 PR 的篡改-threat-model 内 → 非字符串一律按不可读处理。
  2. doctor 告警措辞与代码事实不符——原文写「converges it back … through dsh's plugin channel」,但 doctor 的 dsh 行包含 desktop profile,desktop 实际走 verified-tarball 安装收敛(bili 是唯一 writer,dsh 渠道把桌面应用独占的 profiles/desktop 当普通 profile 驱动 —— 第二写者冲突 #1575),不是 plugin channel → 改为 lane-neutral 措辞。
  3. 新增的 ahead + locally-pinned → info 分支无测试覆盖——补了测试:pinned profile 保持不动、兄弟 registry-pinned profile 仍被刷新、不误报 out-of-band warn。

定向验证:三个相关测试文件本地 42/42 通过,typecheck 干净;全量回归看远端 CI(7f9243201 上运行中)。


结构化八问

  1. 是什么问题:版本号被手工抬高到 registry 之上的 dsh profile/desktop 副本(真实案例 1.2.9 vs 0.1.197)让 dsh plugin-market installs never auto-update: the refresh chain's only trigger is a global bili self-update #1196 自愈永久冻结,且副本静默加载、全程无任何告警。
  2. 什么原因导致的:三道自愈 gate 全部单向比较(local < registry ⇒ 收敛,否则当最新跳过);host-managed lane 本应守「盘上 == 权威源」不变量,且没有任何 provenance 校验。分层:根因 = 等值 gate 缺失 + provenance 缺失;用户报告的单机冻结是表面现象。
  3. 影响范围与严重程度:任何被塞入抬高版 registry-form 副本的 host-managed dsh 机器;命中即完全无法自愈 + 零告警(严重)。另发现连带效应:旧代码顶层 early-return 会把同一 home 下其他 profile 的正常刷新也一起冻住。
  4. 怎么修复的:把「≤ 权威源就算新」收紧为「恰好 in-step 才静默」。ahead 的 registry-form 副本 → 响亮 warn + 经宿主通道收敛回(profile 走 dsh plugin add / pnpm registry 解析,desktop 走 integrity-verified tarball);local pin(link:/file:/git+)尊重为有意 dev pin(info),且兄弟 profile 刷新不再被连坐;doctor 只读 advisory、verdict 不变。Gate 3(dsh boot crash on mixed copies: global >=0.1.176 bundle yml (bare name) + profile copy <=0.1.175 (exports["."]=CLI) — window opened by manual npm i -g, self-heal blocked by the crash itself #1803 旧副本启动崩溃)故意不动,职责不重叠。
  5. 是否有回归问题:逐项清点过——in-step 静默 ✓、stale 升级 ✓、pin 不动 ✓、gate 3 不变 ✓、InstallBackoff(Owner-managed update lanes (dsh desktop / pi / profile converge) retry every 3 min with no InstallBackoff — desktop lane re-downloads the tarball on every retry #2192)键与路径原样 ✓、wire/persistence/config/version 均无变化(静态 gate 全绿佐证)。披露的语义翻转:ahead → 收敛回(旧行为 = 永久冻结);追 prerelease 的用户把 updateTag 指向该渠道后目标解析到 prerelease、in-step 成立,不会误判 ahead。极窄边界(已知、不修):manifest 不再声明 billion-context 依赖但经 bundles 配置挂载的副本,外层会响亮告警但内层刷新不含它——罕见形态,loud-detection 目标已达成。
  6. 是否破坏缓存:不涉及(InstallBackoff/重试策略未触碰)。
  7. 是否改了设计方向:「盘上 ≤ 权威源」收紧为「盘上 == 权威源才静默」——这是对 dsh plugin-market installs never auto-update: the refresh chain's only trigger is a global bili self-update #1196「live registry 为权威」语义的补完整(同 KDD「bundled snapshot 不得压过 live registry」原则),不是转向。
  8. 是否引入了新的配置:无(无新 config 字段/env/CLI flag,config-surface-guard 绿)。

合并前置条件(PR 正文已列,重申):

  • src/update.ts 属 load-bearing infra(§7.4),必须人工 review;
  • 发布前需先做 §5 no-op 验证 release;
  • ahead→收敛回 是本 PR 的核心产品决策,请 owner 在合并时确认(我独立核过安全性论证成立:收敛只走 integrity-verified 渠道、单 writer 安全、deliberate 高版本安装应自行 pin)。

判定:可合(待 7f9243201 CI 转绿 + 上述前置条件满足)。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

CI 最终状态(head 7f9243201):全绿。 39 个 check 全部 success。过程中有 1 次无关的 flaky:test (windows-latest, 24/3) shard 里 tests/claude-native.test.ts 的「hook e2e: an occupied stable port fails loud」抛了 read ECONNRESET(uncaughtException,该文件其余端口类 e2e 全过,与本 PR 触碰的 dsh 自愈/doctor/readDiskVersion 路径无任何交集;本 PR 相关测试在该 shard 及所有其他 shard 均通过)。已触发该 job 重跑,重跑通过,workflow 结论 success。

结合上一条八问结论:判定维持「可合」,前置条件不变——人工 review(§7.4,src/update.ts)、发布前先做 §5 no-op 验证 release、owner 在合并时确认 ahead→收敛回 的语义翻转。

@YunzhouLi-hub

Copy link
Copy Markdown
Contributor

以 #3160 issue 作者的身份确认:语义翻转(local > registry-latest → loud warn + 收敛回 configured-tag 的 registry 版本)正是本 issue 的修复意图,认可,请按此实现。

一、对语义翻转的确认

host-managed lane 该守的不变量是「盘上版本 == 权威目标版本」,不是「盘上 ≤ 权威就安全」。registry-form 副本高于 latest 在物理上不可能是一次真实解析——registry 上不存在那个版本的产物——所以它只能是 out-of-band 放置。旧行为把它当「已最新」静默跳过,等于把「真·更新的 prerelease」和「手工伪造的高版本号」混为一谈,这正是 #3160 的病灶。把 ahead 从「静默 no-op」改成「loud warn + 经 integrity-verified 通道收敛回」是把 #1196 的自愈语义补完整(原来只覆盖「落后」这一半),不是行为退化。

二、披露的边界全部认可

场景 处理 作者确认
local == latest 静默 ✔ 不变
local < latest 向上收敛 ✔ 不变
local > latest(registry-form) warn + 收敛回 ✔ 即本 issue 诉求
link: / file: / git+ pin 不动(dev pin,info) ✔ 白名单正确
追 prerelease updateTag 指向对应 channel → in-step 成立 ✔ 无误伤
diskVersion 读不到 固定走刷新分支 ✔ 对运行中副本不可达,且刷新缺失/损坏副本是 #1196 的正确方向

收敛只走 integrity-verified 通道(pnpm 从 registry 解析 / verified registry tarball),单写者安全,这点也让翻转可接受。

三、与 stage 2 检测 PR(#3169)的衔接

#3169(检测 only)与本 PR 的 doctor advisory 有一处重叠:本 PR 的 dshOutOfBandCopyWarning 与 #3169 的 spec-above-registry rung 判断的是同一个条件(registry-form 高于 registry latest),同一台机器上会弹出两条等义告警。

建议这样分工,避免 doctor 重复告警:

  • 本 PR 合入后,我把 feat: loud provenance detection for wedged dsh profile copies (#3160) #3169 rebase 并去掉 spec-above-registry rung(连同它在 doctor 侧的重复面),只保留本 PR 不覆盖的 rung:copy-vs-lock(盘上版本 vs lockfile 解析)、manifest-vs-lock(声明 vs lock 记录)、copy-vs-manifest(无 lockfile 时副本无法被 manifest 解释)、build-divergence(同版本不同 build commit)。
  • 二者覆盖面互补:本 PR 的 advisory 管「高于 registry」;feat: loud provenance detection for wedged dsh profile copies (#3160) #3169 的 rung 管「lockfile/manifest 与盘上副本三方矛盾」——这类矛盾在本 PR 合入后依然有独立价值(dormant profile、gate 尚未跑到的窗口、以及收敛后 manifest 被改回正确版本的验证面)。

四、CI 备注

本 PR 当前唯一红灯 test (windows-latest, 22, 3) 是已知端口争用 flake(hook e2e: an occupied stable port fails loud — never port-hops),与本 PR 无关。另注意:你的分支切于 09:46,早于 #3088(10:00 合入,引入了 #3174 的 #2185 确定性失败)和 #3112(test-suite-lock,#3175 的 Windows 确定性失败)——rebase 到当前 master 后会撞上这两个既有失败,均已另立 issue 跟踪,不是本 PR 的问题。

五、门禁

认可「排在 no-op release 之后合入」的排序。src/update.ts 的改动我不催 merge;检测 PR(#3169)不碰 updater,可以先落地止血「静默」窗口。

@YunzhouLi-hub

Copy link
Copy Markdown
Contributor

勘误:本条上面确认评论里「mirrors the bundled-snapshot-must-not-outrank-live-registry KDD (#2439)」的引用是错的——#2439 是 Windows PowerShell 黑框 issue,与版本不变量无关,抱歉。

正确的依据是仓内已文档化的同向不变量:CONFIGURATION.md:1406「Live registry wins when reachable, bundled snapshot is the offline floor (#282)」——profile/desktop 副本属 local copy,live registry 必须 outrank;never-lower 判据在 registry-form 场景把这条倒转了。另一个现存先例是 src/agent/pi-native.ts:94(两份副本共存时宿主 lane 胜出 + 大声提示移除另一份,而非「版本号高的静默留着」)。已在 #3160 回复里更正并展开论证。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

逐条回复:

1. 勘误——已核实:#2439 确实是「Windows: 重建 combined-ca.pem 时闪出 PowerShell 黑框」,与版本不变量无关,更正成立。一点澄清:本线程我发的两条 review 评论均未引用 #2439(第 7 问只以文字表述了「bundled snapshot 不得压过 live registry」原则、未给编号),被更正的引用应来自线程里的其他评论。你给出的两个正确依据我都查证过、成立:CONFIGURATION.md 的「Live registry wins when reachable, bundled snapshot is the offline floor (#282)」(约 1399 行,原文写在计费 price-profile 语境,方向一致);src/agent/pi-native.ts:94 的 warnLegacyBcpCoResident(双安装共存 → error 级大声告警 + 明确移除指引,是「异常副本响亮检测而非静默接受」的好先例)。

2. 语义翻转确认——收到。作为 #3160 作者,你的确认是这个产品决策的重要背书;但按仓库规则最终拍板仍在 owner(@ranxianglei):人工 review(§7.4,src/update.ts 属 load-bearing infra)、发布前的 §5 no-op 验证 release、以及合并时对 ahead→收敛回 翻转的签认。实现已与你确认的语义完全一致,无需改动;你列的边界表与 PR 正文披露的行为矩阵逐项吻合。

3. 与 #3169 的衔接——注意:#3169 当前状态是 closed(不是 open)。你截断的「本 PR 合入后我把 #3169 rebase 并去…」计划以它仍开放为前提,请先核对现状与意图——是已被有意关闭(被本 PR 取代 / 待重开)?分工建议本身我认同方向:本 PR 的 doctor 告警是只读、离线(覆盖 dormant profiles),若 #3169 的 spec-above-registry rung 以后回归,「doctor 告警归本 PR、#3169 保留其余 rung」可避免同机双告警。具体怎么拆由你与 owner 定夺,我不会代为改动 #3169。

@YunzhouLi-hub

Copy link
Copy Markdown
Contributor

确认该分支已过时(mergeable_state=behind),并做了两件核实,供 refresh 时参考:

一、落后幅度与 CI 现状

  • 分支最后一次并入 master 是 701cb49a(10/15 01:44),master 现至 effd5442(Merge PR Per-route k̂ calibration policy + pressure-band floor plumbing (#3211) #3218,10/15 02:39),落后 13 个 commit。
  • 但 head 916990f7 的 CI 是全绿(36/36 checks success,含 ubuntu/windows × 22/24 全部 test shards 与 static-gates),即当前内容本身是健康的,纯粹是新鲜度问题。

二、两个此前登记的 master 级失败不会复活

我此前在本 PR 与 #3160 评论区提过「rebase 后会撞两个既有失败」——这两个失败的修复已经落在本分支里了:

两者均早于本分支 01:44 的 master 合入,已在分支内(这也是本轮 test lanes 全绿的原因)。两个 issue 均已 closed/completed。

三、本地试合结果:零冲突

我把当前 origin/master(effd5442)试合进本分支 HEAD(916990f7):automatic merge went well,无冲突文件。落后的 13 个 commit(#3192/#3200/#3210、#3204/#3213、#3211/#3218 等 dead-ref 提示、attach 诊断、校准策略系列)与本 PR 的改动区域(src/update.ts、src/doctor.ts 及其测试)互不相交。

建议

merge master 或 rebase 刷新一下分支即可,刷新后按原计划走 §5 no-op 验证 release 与 §7.4 人工 review。语义侧我作为 issue 作者已在 #3160 确认(含 packument 判据这版),无进一步意见,不催 merge。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

你列的事实我都独立核实过,然后按 §7.2(合并前对齐当前 master)刷新了分支。逐条:

核实结果

已完成的刷新

  • origin/master(effd5442,含 Per-route k̂ calibration policy + pressure-band floor plumbing (#3211) #3218)合入 PR 分支,merge commit 60a20adea,已推送。选 merge 而非 rebase:分支里已有 master 合入历史,且 rebase 会重写他人已推的提交。
  • 本地验证:typecheck 干净;定向测试 update-dsh-self-refresh / update-desktop-inplace / doctor 共 44/44 通过(比刷新前多 2 个,即 packument gating 的新用例)。
  • GitHub 侧 mergeable_state 已从 behind 转为 blocked(= 不再落后,只等 required checks 变绿)。

顺带 review 了 916990f7c(packument gating):设计自洽——ahead 副本只有在 packument 证明「该版本不存在」时才收敛回;已发布或无法验证的一律 defer + loud warn + 明确 hint(set updateTag / pin / remove)。defer 分支必须 return 而不能落到尾部刷新:runProfileBundlesRefresh 对 registry-form 非 in-step 副本会直接重装,落下去就等于绕过「永不自动降级已发布构建」这条新不变量。fail-open 方向安全(查不了就不降级),#3160 真实案例(1.2.9 未发布)仍走收敛路径,PR 描述也已同步更新披露。

当前状态:60a20adea 的 CI 在跑(24 checks 刚启动)。全绿后本 PR 侧无事可做,剩原计划的三件人工事项:§7.4 人工 review(src/update.ts)、§5 no-op 验证 release、owner 对 ahead→条件收敛 语义的最终签认。不催 merge。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

刷新后的 head 60a20adea CI 全绿:40/40 checks success(含 ubuntu/windows × 22/24 全部 test shards、static-gates、kernel-guard、e2e 各 lane),无 flake、无需 rerun。分支已对齐当前 master,agent 侧工作到此收工;剩余即原计划的三件人工事项:§7.4 人工 review、§5 no-op 验证 release、owner 语义签认后合并。

…ly (#3160 owner-feedback)

Being absent from the public packument is not safe proof of tampering: a user's own maintained build is also absent. An ahead registry-form copy is now surfaced loudly (self-heal log + bili doctor) and left untouched instead of being converged back; local pins still fall through so their siblings refresh. Auto-quarantine of provably-unexplained copies moves to #3224.
ranxianglei pushed a commit that referenced this pull request Oct 17, 2026
The #3224 author's in-thread decision (floor 5): the desktop profile is
the one dsh lane bili writes IN PLACE — after every self-update its
manifest spec is synced to the new version while the lockfile stays stale
by design until dsh's in-app manager reconciles it (#2894), so post-update
manifest/lock/copy disagreement there is expected and locally
indistinguishable from a true wedge. Detecting it would mean a permanent
false positive on every machine with an updated desktop profile, with a
remediation command the CLI refuses (--profile desktop, #1575). The ahead
face of that lane rides #3164's registry advisory instead.

One name-based guard in dshProfileCopyFindings covers both detection exits
(startup audit + doctor lane); no path guessing. One new test pins the
skip: identical wedge shape on desktop + web, only web reported.
@ranxianglei

Copy link
Copy Markdown
Owner Author

(ai: zhipuai-lb/glm-5.3)

陈旧票清理(第三批动作①):目标 issue #3160 已关。为避免重复劳动关闭本 PR;如判断有误请直接 reopen 并留言。

@ranxianglei ranxianglei changed the title fix: converge inflated out-of-band dsh profile/desktop copies instead of freezing (#3160) [PR] fix: converge inflated out-of-band dsh profile/desktop copies instead of freezing (#3160) Oct 24, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Profile 副本自愈可被本地高版本号永久卡死,且 out-of-band 副本无来源校验(#1196 变体)

2 participants