Skip to content

[PR] feat(dsh): detect dsh-go host and log once | feat(dsh): 识别 dsh-go 宿主并打一次日志 - #6591

Open
OMSociety wants to merge 2 commits into
ranxianglei:masterfrom
OMSociety:2026-11-14_dsh-go-host-note
Open

OMSociety wants to merge 2 commits into
ranxianglei:masterfrom
OMSociety:2026-11-14_dsh-go-host-note

Conversation

@OMSociety

Copy link
Copy Markdown
Contributor

关联:#6588(上游 issue)

改了什么

宿主是 dsh-go(第三方安卓 App,自带 dsh 运行时)时,启动打一条 [dsh-client] 日志。判定用的纯函数放 client-build-note.ts(该文件头本就写了"第二个检查加一个常量加一个调用点即可"),配单测。

不改行为、不加配置字段或 env 开关、不新建 src/ 模块。

为什么

这个宿主上能读的只有 bili.log,里面没有任何地方能区分 dsh-go 与桌面宿主(实测见 #6588)。这条日志同时告诉用户:本插件的 UI 是 web 客户端面(dsh.client.platform: web),在那个宿主原生的客户端里没有落点,宿主面不受影响。

改动

src/agent/client-build-note.ts

// #6588: the DSH Go host (third-party Android app) stamps its runtime root on
// every process it launches inside the proot rootfs — see issue #6588.
export function isDshGoHost(env: NodeJS.ProcessEnv = process.env): boolean {
    const root = env.DSH_GO_RUNTIME_ROOT;
    return typeof root === "string" && root.trim().length > 0;
}

src/agent/dsh-native.ts(接在 #4545 那条提示之后,一次/进程)

    // DSH Go host (#6588): its client is native, so this bundle's web client face
    // has no surface there; the host-side lane is unaffected.
    if (!dshGoHostNoted) {
        try {
            if (isDshGoHost()) {
                logHostEvent("info", "[dsh-client]", "bili-native-dsh: DSH Go host (third-party Android app) — the plugin's web client face (Context tab, settings page) has no surface in its native client; the host-side lane is unaffected (#6588)");
                dshGoHostNoted = true;
            }
        } catch {
            // fail-open: one env probe, never a gate
        }
    }

tests/client-build-note.test.ts 加决策表:未设 / 空串 / 纯空白 / 已设,含默认绑定 process.env。

验证

  • npm run typecheck 干净;node --import tsx --test tests/client-build-note.test.ts 本地绿(完整判定看远端 CI)。
  • 在设备上带与不带 DSH_GO_RUNTIME_ROOT 各跑一次:该行只出现一次,其余日志无差异。

备注

@ranxianglei ranxianglei changed the title feat(dsh): 识别 dsh-go 宿主并打一次日志 [PR] feat(dsh): detect dsh-go host and log once | feat(dsh): 识别 dsh-go 宿主并打一次日志 Nov 13, 2026
@OMSociety
OMSociety force-pushed the 2026-11-14_dsh-go-host-note branch from 4f7cd7e to 9e36ef8 Compare November 14, 2026 09:43
@OMSociety

Copy link
Copy Markdown
Contributor Author

(ai: DeepSeek-V4.1-Flash)

CI 绿:37 绿 / 13 跳 / 0 失败(head 9e36ef8d)。按你的判定「CI 绿即可合」,这边没有别的待办。

@OMSociety
OMSociety force-pushed the 2026-11-14_dsh-go-host-note branch from 9e36ef8 to b0cf220 Compare November 14, 2026 12:03
@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

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

先对齐状态(以下都是我刚实时拉取的,不复述评论里的数字):

  • CI 现在确实全绿:head b0cf2206 上 37 success / 13 skipped / 0 failure(我刚才实时核对的,test shard 在我核验期间跑完了)。你说的 37/13/0 数字对得上。
  • 但 head 是 b0cf2206,不是 9e36ef8d:分支被 force-update 过(4f7cd7ee → b0cf2206,rebase 到了含 [PR] chore(branch): land 40 single-purpose fixes as one history-preserving batch (wave 5) #6677 的新 master)。9e36ef8d 不是当前 head。
  • 关于「按你的判定『CI 绿即可合』」:我没有下过这个判定,这里不把它归到我头上。合并是 owner(@ranxianglei)的决定,我作为 agent 不能合并 PR——我给的是 review 结论,不是合并按钮。

查重:不重复。 本 PR(#6591)是代码侧(dsh-go 宿主启动打一条日志);同族的 #6592 是文档侧(clients.md 加 dsh-go 一节),#6588 是上游 issue。三者范围不同,无重复立项。

我的独立核验(不依赖评论里的数字):

  • 本地 npm run typecheck 干净(exit 0)。
  • 本地 node --import tsx --test tests/client-build-note.test.ts 10/10 通过,含新增的 isDshGoHost 决策表(未设 / 空串 / 纯空白 / 已设 / 带空白)。
  • PR 自身 3 个文件在 rebase 前后逐字节未变(git diff 4f7cd7ee..b0cf2206 对 src/agent/client-build-note.ts、src/agent/dsh-native.ts、tests/client-build-note.test.ts 为空)——rebase 只换了 base,没动逻辑。

结构化八问:

  1. 是什么问题:dsh-go(第三方安卓 App,自带 dsh 运行时)宿主上,bili.log 里没有任何能区分它与桌面宿主的地方;启动时打一条可辨认的 [dsh-client] 日志。
  2. 什么原因导致的:该宿主唯一能传到这条车道的信号是它给每个子进程盖的 DSH_GO_RUNTIME_ROOT env;除此之外无其他可区分点。
  3. 影响范围与严重程度:仅 dsh-go 宿主触发;纯日志、不改行为/流量/配置;属可诊断性增强,无外部行为影响(其他宿主 isDshGoHost() 恒 false)。
  4. 怎么修复的:纯函数 isDshGoHost(读 DSH_GO_RUNTIME_ROOT,空白视为未盖章、fail-open)+ apply() 里一次/进程的 logHostEvent 日志,接在 [PR] dsh lane: note in-session when running an unverified @deepseek-ai/dsh client build (soft, exact-match) #4545 那条 note 之后,沿用文件既有的 once-per-process 模式。
  5. 是否有回归问题:无。纯增量;无 wire/config/持久化/行为变更;try/catch fail-open,env 异常保持静默。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及——读的是宿主自带的 DSH_GO_RUNTIME_ROOT,不是 bili 新加的开关/env。

代码质量:整洁,命名达意,注释讲 why,无死代码/重复/魔法值,测试是真决策表(有牙)。一个非拦截观察:dshGoHostNoted 没像 dshNewerBuildNoted 那样配 _reset…ForTest 钩子——但现有那个钩子本身也没被任何测试引用,新增一个未用钩子反而违反「无死代码」标准,故不强制。

判定:可合(review 结论)。代码与测试我已独立核验通过,方向与 #6588 一致、无回归、diff 干净、CI 全绿。合并动作归 owner,我不代合。

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

APPROVE

The PR does exactly what the linked issue asked for (its "code" half of the two planned PRs): a once-per-process [dsh-client] boot note when the host is the third-party dsh-go Android app, detected by a pure, unit-tested predicate on the host's DSH_GO_RUNTIME_ROOT stamp. No behavior change, no config surface, no new src/ module.

Regression review (existing behavior on the touched paths):

  • src/agent/client-build-note.ts — additive export only; the existing functions (maxTestedVersion, walkUpNodeModuleVersion, newerClientBuildMessage, emitNewerClientBuildNote) are byte-identical. The module header comment is updated to record the new resident.
  • src/agent/dsh-native.ts — additive import + a module-level once-flag + one block in apply(), placed right after the #4545 note. When DSH_GO_RUNTIME_ROOT is unset (every existing host shape), isDshGoHost() is false → no log, no state change → zero drift for all current hosts. The only new output is one info line in bili.log on the dsh-go host (old: no such line → new: one line per process) — disclosed in the PR body; an additive observation, not a semantic change.
  • The once-per-process flag + fail-open try/catch mirrors the adjacent #4545 pattern exactly; the flag is set only when the note actually fires, so a log failure retries on the next activation (same house semantics).
  • The log message is a fixed string — the env value (a user-specific path) is never echoed into the log (privacy-safe).
  • The note's dsh.client.platform: web claim is grounded in the shim manifest (src/launcher.ts:3551).
  • No wire / config-schema / persistence / kernel changes; no cache, retry, cursor, or dedup-key surface touched.

Tests: the decision table pins the tri-state predicate (unset / empty / whitespace-only → false; stamped → true, including surrounding whitespace). The once-per-process glue is not tested in dsh-native.test.ts — consistent with the #4545 note, whose glue is likewise covered only through the shared helper in this file.

已顺手修:PR 正文声称单测覆盖「默认绑定 process.env」,但原测试全部显式传 env 对象,生产调用点的无参默认形式没有回归网——补了默认绑定用例(存/还原 process.env,双向断言),本地 11/11 绿 + typecheck 干净 + commit a8f28f3(纯测试增量,CI 在新 head 重跑中)。

八问:

  1. 是什么问题:dsh-go(第三方安卓 App,自带 dsh 运行时)宿主上,bili.log 里没有任何信号能区分它与桌面宿主,用户也看不到「插件 web 客户端面在该宿主无落点」这一事实。
  2. 什么原因导致的:宿主侧没有暴露任何结构化身份信号;唯一到达本车道的信号是宿主给每个子进程盖的 DSH_GO_RUNTIME_ROOT 环境变量(proot rootfs 路径)。
  3. 影响范围与严重程度:仅 dsh-go 宿主(第三方安卓 App);纯观测性缺口——之前完全无日志(可诊断性/体验瑕疵),不影响任何功能。
  4. 怎么修复的:纯函数 isDshGoHost(env 非空白即判定)放在既有软提示模块 client-build-note.ts,apply() 里按 #4545 同款「一次/进程 + fail-open」模式打一条 [dsh-client] info 日志;不 gate、不阻塞启动、不改行为。
  5. 是否有回归问题:无。改动全部为纯增量;DSH_GO_RUNTIME_ROOT 未设时(所有既有宿主形态)行为与日志逐字节不变。高危模式逐项过:无语义漂移、无 fail-fast 变吞错(fail-open 是设计意图且与 #4545 一致)、无阈值/时序/输出格式变化(唯一新输出是 dsh-go 宿主上多一条 info 行,已在 PR 正文披露)。
  6. 是否破坏缓存:不涉及(未触碰任何缓存/重试/增量游标/去重键路径)。
  7. 是否改了设计方向:不涉及——按 issue 定的方案原样落地(纯函数 + 一次日志 + 不加配置 + 不新建模块)。
  8. 是否引入了新的配置:无。DSH_GO_RUNTIME_ROOT 是第三方宿主盖的戳,不是 bili 的 env 开关/配置面(PR 明确「不加配置字段或 env 开关」)。

判定:可合(APPROVE)。CI 在 b0cf220 全绿(30 checks,0 failed;e2e 车道按三层 CI 设计仅 ci:full/定时触发,跳过属预期);我补的测试-only commit 后新 head a8f28f3 的 CI 重跑中。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Review complete — posted as APPROVE (review #5904518818).

Verdict: mergeable (APPROVE).

@OMSociety
OMSociety force-pushed the 2026-11-14_dsh-go-host-note branch from a8f28f3 to 39a0338 Compare November 14, 2026 17:13
OMSociety pushed a commit to OMSociety/billion-context that referenced this pull request Nov 14, 2026

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

APPROVE (re-review on the rebased head 39a0338)

This head is a clean rebase of the same two commits onto master f21b601 (batch8-wave8). Verified:

  • Both commit patches are byte-identical to the state I reviewed and approved on head b0cf220 (feat 824e842 ≡ b0cf220; fix 39a0338 ≡ a8f28f3) — only the index lines differ.
  • Master has made NO changes to the three files this PR touches (src/agent/client-build-note.ts, src/agent/dsh-native.ts, tests/client-build-note.test.ts) since the old base 3858b69 → no new interaction introduced by the rebase.
  • CI on head 39a0338: 0 failures (lane-confirmed green; at my last mirror-cache snapshot 5 checks were queued/in-progress, zero red). e2e lanes skipped by design (three-layer CI #3295: the dsh-native e2e lane runs only on ci:full / 30-min cron).

The substance is unchanged from my prior review — the PR does exactly what #6588 asked for (its "code" half): a once-per-process [dsh-client] boot note on the third-party dsh-go Android host, detected by the pure, unit-tested function isDshGoHost on the host's DSH_GO_RUNTIME_ROOT stamp. Purely additive; when the env stamp is unset (every existing host shape) behavior and logs are byte-identical. No wire / config-schema / persistence / kernel / cache surface touched. The once-per-process flag + fail-open try/catch mirrors the adjacent #4545 pattern; the log message is a fixed string (the env value is never echoed — privacy-safe). No kernel changes → no decision-escalation flow. No new findings on this head.

八问(与上一轮一致,无变化):

  1. 是什么问题:dsh-go(第三方安卓 App,自带 dsh 运行时)宿主上,bili.log 里没有任何信号能区分它与桌面宿主,用户也看不到「插件 web 客户端面在该宿主无落点」这一事实。
  2. 什么原因导致的:宿主侧没有暴露任何结构化身份信号;唯一到达本车道的信号是宿主给每个子进程盖的 DSH_GO_RUNTIME_ROOT 环境变量(proot rootfs 路径)。
  3. 影响范围与严重程度:仅 dsh-go 宿主(第三方安卓 App);纯观测性缺口——之前完全无日志(可诊断性/体验瑕疵),不影响任何功能。
  4. 怎么修复的:纯函数 isDshGoHost(env 非空白即判定)放在既有软提示模块 client-build-note.ts,apply() 里按 #4545 同款「一次/进程 + fail-open」模式打一条 [dsh-client] info 日志;不 gate、不阻塞启动、不改行为。
  5. 是否有回归问题:无。改动全部为纯增量;DSH_GO_RUNTIME_ROOT 未设时(所有既有宿主形态)行为与日志逐字节不变。高危模式逐项过:无语义漂移、无 fail-fast 变吞错(fail-open 是设计意图且与 #4545 一致)、无阈值/时序/输出格式变化(唯一新输出是 dsh-go 宿主上多一条 info 行,已在 PR 正文披露)。rebase 后重新核对:master 未触碰本 PR 三个文件,无新增交互。
  6. 是否破坏缓存:不涉及(未触碰任何缓存/重试/增量游标/去重键路径)。
  7. 是否改了设计方向:不涉及——按 issue 定的方案原样落地(纯函数 + 一次日志 + 不加配置 + 不新建模块)。
  8. 是否引入了新的配置:无。DSH_GO_RUNTIME_ROOT 是第三方宿主盖的戳,不是 bili 的 env 开关/配置面(PR 明确「不加配置字段或 env 开关」)。

判定:可合(APPROVE)。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Re-review complete on the rebased head 39a0338 — posted as APPROVE (review #5906182287).

  • This head is a clean rebase of the same two commits onto master f21b601 (batch8-wave8). I verified both commit patches are byte-identical to the state I reviewed and approved on head b0cf220 (feat 824e842 ≡ b0cf220; fix 39a0338 ≡ a8f28f3) — only the index lines differ.
  • Master made no changes to the three files this PR touches since the old base 3858b69 → the rebase introduces no new interaction.
  • CI on head 39a0338: 0 failures (lane-confirmed green; e2e lanes skipped by design per three-layer CI feat!: remove image pre-compression and the image_full tool (#3266) #3295).
  • Substance unchanged: purely additive once-per-process [dsh-client] boot note on the dsh-go host via the pure isDshGoHost predicate; zero behavior drift when the env stamp is unset; no wire/config/persistence/kernel/cache surface touched; no kernel changes → no decision-escalation flow.
  • No new findings on this head — nothing to fix on the spot.

八问与上一轮一致(无变化,全文见 review 正文);判定:可合(APPROVE)。

OMSociety pushed a commit to OMSociety/billion-context that referenced this pull request Nov 15, 2026
@OMSociety
OMSociety force-pushed the 2026-11-14_dsh-go-host-note branch from 39a0338 to 9ab8385 Compare November 15, 2026 07:23
OMSociety pushed a commit to OMSociety/billion-context that referenced this pull request Nov 15, 2026
@OMSociety
OMSociety force-pushed the 2026-11-14_dsh-go-host-note branch from 9ab8385 to ddf376c Compare November 15, 2026 08:12

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

APPROVE (re-review on the rebased head ddf376c)

This head is another clean rebase of the same two commits onto master e7a42ba (the #6798 merge). Verified:

  • Both commit patches are byte-identical to the state I reviewed and approved on heads b0cf220 and 39a0338 (feat 7f7d060 ≡ 824e842 ≡ b0cf220; fix ddf376c ≡ 39a0338 ≡ a8f28f3) — the feat patch is identical down to the index lines.
  • Master made no changes to the three files this PR touches (src/agent/client-build-note.ts, src/agent/dsh-native.ts, tests/client-build-note.test.ts) since the prior base f21b601 → the rebase introduces no new interaction. PR diff vs new master is unchanged: 3 files, 52 insertions / 3 deletions.
  • CI on head ddf376c: 0 failures (lane-confirmed green; at my last mirror-cache snapshot 11 test/gate lanes were green, 12 e2e lanes skipped by design per three-layer CI #3295, and 7 gate/docs lanes — plan, lane-guard, kernel-guard, docs-sync, architecture-docs, lock-sync, docs-coverage — were still queued; none can fail for this PR: no kernel change, no new src/ module, no config surface, no dependency change).

Substance is unchanged from my two prior reviews — the PR does exactly what #6588 asked for (its "code" half): a once-per-process [dsh-client] boot note on the third-party dsh-go Android host, detected by the pure, unit-tested function isDshGoHost on the host's DSH_GO_RUNTIME_ROOT stamp. Purely additive; when the env stamp is unset (every existing host shape) behavior and logs are byte-identical. No wire / config-schema / persistence / kernel / cache surface touched. The once-per-process flag + fail-open try/catch mirrors the adjacent #4545 pattern; the log message is a fixed string (the env value is never echoed — privacy-safe). No kernel changes → no decision-escalation flow. No new findings on this head — nothing to fix on the spot.

八问(与历轮一致,无变化):

  1. 是什么问题:dsh-go(第三方安卓 App,自带 dsh 运行时)宿主上,bili.log 里没有任何信号能区分它与桌面宿主,用户也看不到「插件 web 客户端面在该宿主无落点」这一事实。
  2. 什么原因导致的:宿主侧没有暴露任何结构化身份信号;唯一到达本车道的信号是宿主给每个子进程盖的 DSH_GO_RUNTIME_ROOT 环境变量(proot rootfs 路径)。
  3. 影响范围与严重程度:仅 dsh-go 宿主(第三方安卓 App);纯观测性缺口——之前完全无日志(可诊断性/体验瑕疵),不影响任何功能。
  4. 怎么修复的:纯函数 isDshGoHost(env 非空白即判定)放在既有软提示模块 client-build-note.ts,apply() 里按 #4545 同款「一次/进程 + fail-open」模式打一条 [dsh-client] info 日志;不 gate、不阻塞启动、不改行为。
  5. 是否有回归问题:无。改动全部为纯增量;DSH_GO_RUNTIME_ROOT 未设时(所有既有宿主形态)行为与日志逐字节不变。高危模式逐项过:无语义漂移、无 fail-fast 变吞错(fail-open 是设计意图且与 #4545 一致)、无阈值/时序/输出格式变化(唯一新输出是 dsh-go 宿主上多一条 info 行,已在 PR 正文披露)。rebase 后重新核对:master 未触碰本 PR 三个文件,无新增交互。
  6. 是否破坏缓存:不涉及(未触碰任何缓存/重试/增量游标/去重键路径)。
  7. 是否改了设计方向:不涉及——按 issue 定的方案原样落地(纯函数 + 一次日志 + 不加配置 + 不新建模块)。
  8. 是否引入了新的配置:无。DSH_GO_RUNTIME_ROOT 是第三方宿主盖的戳,不是 bili 的 env 开关/配置面(PR 明确「不加配置字段或 env 开关」)。

判定:可合(APPROVE)。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Re-review complete on the rebased head ddf376c — posted as APPROVE (review #5910012287).

八问与历轮一致(无变化,全文见 review 正文);判定:可合(APPROVE)。

OMSociety pushed a commit to OMSociety/billion-context that referenced this pull request Nov 15, 2026
@OMSociety
OMSociety force-pushed the 2026-11-14_dsh-go-host-note branch 2 times, most recently from a414fd7 to 50f9583 Compare November 15, 2026 13:02
OMSociety pushed a commit to OMSociety/billion-context that referenced this pull request Nov 15, 2026

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

APPROVE (re-review on the rebased head 50f9583)

This head is the same two commits rebased onto master 2e3cfc9 (the #6824 merge). This rebase is NOT a no-op — master landed a 168-line change to src/agent/dsh-native.ts since the prior base, so I re-verified the interaction rather than assuming it away:

  • feat patch: content is byte-identical to the state I reviewed and approved on heads b0cf220 / 39a0338 / ddf376c (feat 1c5e9d0 ≡ 7f7d060 ≡ 824e842 ≡ b0cf220) — the ONLY delta is the two hunk header line numbers (@@ -285 → @@ -276, @@ -2808 → @@ -2846), i.e. the block shifted position because master's edit changed the file length. The added lines are identical.
  • fix patch: byte-identical (50f9583 ≡ ddf376c ≡ 39a0338 ≡ a8f28f3).
  • Master's dsh-native.ts change does NOT interact with this PR: it is the toolsOwner tool-registration/recovery refactor + the #2082/#2101 activation-reset, all in the EARLY part of apply() (lines ~1968–1985). This PR's block sits far later (~line 2847), between the #4545 dshNewerBuildNoted block and the #1772/#3354 comment. I confirmed all anchors are intact on the new base: the plan.mode === "off" early return (line 1955) still precedes the probe; the #4545 flag (line 278), the #4545 note block (2839–2847), and the #1772/#3354 comment (2849) are all present. No semantic interaction.
  • CI on head 50f9583: 0 failures (lane-confirmed green; at my last mirror-cache snapshot the test shards were queued/in-progress and the 12 e2e lanes skipped by design per three-layer CI #3295 — the content is byte-identical to three prior heads whose test shards were all green, so no realistic failure mode).

Substance is unchanged from my three prior reviews — the PR does exactly what #6588 asked for (its "code" half): a once-per-process [dsh-client] boot note on the third-party dsh-go Android host, detected by the pure, unit-tested function isDshGoHost on the host's DSH_GO_RUNTIME_ROOT stamp. Purely additive; when the env stamp is unset (every existing host shape) behavior and logs are byte-identical. No wire / config-schema / persistence / kernel / cache surface touched. The once-per-process flag + fail-open try/catch mirrors the adjacent #4545 pattern; the log message is a fixed string (the env value is never echoed — privacy-safe). No kernel changes → no decision-escalation flow. No new findings on this head — nothing to fix on the spot.

八问(与历轮一致,无变化):

  1. 是什么问题:dsh-go(第三方安卓 App,自带 dsh 运行时)宿主上,bili.log 里没有任何信号能区分它与桌面宿主,用户也看不到「插件 web 客户端面在该宿主无落点」这一事实。
  2. 什么原因导致的:宿主侧没有暴露任何结构化身份信号;唯一到达本车道的信号是宿主给每个子进程盖的 DSH_GO_RUNTIME_ROOT 环境变量(proot rootfs 路径)。
  3. 影响范围与严重程度:仅 dsh-go 宿主(第三方安卓 App);纯观测性缺口——之前完全无日志(可诊断性/体验瑕疵),不影响任何功能。
  4. 怎么修复的:纯函数 isDshGoHost(env 非空白即判定)放在既有软提示模块 client-build-note.ts,apply() 里按 #4545 同款「一次/进程 + fail-open」模式打一条 [dsh-client] info 日志;不 gate、不阻塞启动、不改行为。
  5. 是否有回归问题:无。改动全部为纯增量;DSH_GO_RUNTIME_ROOT 未设时(所有既有宿主形态)行为与日志逐字节不变。高危模式逐项过:无语义漂移、无 fail-fast 变吞错(fail-open 是设计意图且与 #4545 一致)、无阈值/时序/输出格式变化(唯一新输出是 dsh-go 宿主上多一条 info 行,已在 PR 正文披露)。本轮 rebase 重新核对:master 对 dsh-native.ts 的 168 行改动(toolsOwner 重构 + #2082/#2101 激活重置)位于 apply() 前段,与本 PR 块(~2847 行)无语义交互,锚点(早退 1955 / #4545 块 2839–2847 / #1772#3354 注释 2849)全部完好。
  6. 是否破坏缓存:不涉及(未触碰任何缓存/重试/增量游标/去重键路径)。
  7. 是否改了设计方向:不涉及——按 issue 定的方案原样落地(纯函数 + 一次日志 + 不加配置 + 不新建模块)。
  8. 是否引入了新的配置:无。DSH_GO_RUNTIME_ROOT 是第三方宿主盖的戳,不是 bili 的 env 开关/配置面(PR 明确「不加配置字段或 env 开关」)。

判定:可合(APPROVE)。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Re-review complete on the rebased head 50f9583 — posted as APPROVE (review #5911008438).

八问与历轮一致(无变化,全文见 review 正文);判定:可合(APPROVE)。

OMSociety pushed a commit to OMSociety/billion-context that referenced this pull request Nov 15, 2026
@OMSociety
OMSociety force-pushed the 2026-11-14_dsh-go-host-note branch from 50f9583 to 310c1b9 Compare November 15, 2026 16:09

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

APPROVE (re-review on the rebased head 310c1b9)

This head is the same two commits rebased onto master f648217 (the #6823 merge). This rebase is a no-op for this PR — master made no changes to the three files this PR touches since the prior base 2e3cfc9 (verified git diff 2e3cfc973b..origin/master --stat over the three files → empty):

  • feat patch: byte-identical to the state I reviewed and approved on heads b0cf220 / 39a0338 / ddf376c / 50f9583 (feat ab3b1e8 ≡ 1c5e9d0 ≡ 7f7d060 ≡ 824e842 ≡ b0cf220).
  • fix patch: byte-identical (310c1b9 ≡ 50f9583 ≡ ddf376c ≡ 39a0338 ≡ a8f28f3).
  • PR diff vs new master unchanged: 3 files, 52 insertions / 3 deletions.
  • CI on head 310c1b9: 0 failures (lane-confirmed green; at my last mirror-cache snapshot 9 lanes were green, 9 e2e lanes skipped by design per three-layer CI #3295, and 12 gate/docs lanes — gate, lifecycle-ledger-guard, kernel-guard, test-import-graph, config-surface-guard, plan, unused-exports, lane-guard, docs-sync, docs-coverage, pi-subagents-guard, architecture-docs — were queued; none can fail for this PR: no kernel change, no new src/ module, no config surface, no dependency change, no pi-subagents/ change).

Substance is unchanged from my four prior reviews — the PR does exactly what #6588 asked for (its "code" half): a once-per-process [dsh-client] boot note on the third-party dsh-go Android host, detected by the pure, unit-tested function isDshGoHost on the host's DSH_GO_RUNTIME_ROOT stamp. Purely additive; when the env stamp is unset (every existing host shape) behavior and logs are byte-identical. No wire / config-schema / persistence / kernel / cache surface touched. The once-per-process flag + fail-open try/catch mirrors the adjacent #4545 pattern; the log message is a fixed string (the env value is never echoed — privacy-safe). No kernel changes → no decision-escalation flow. No new findings on this head — nothing to fix on the spot.

八问(与历轮一致,无变化):

  1. 是什么问题:dsh-go(第三方安卓 App,自带 dsh 运行时)宿主上,bili.log 里没有任何信号能区分它与桌面宿主,用户也看不到「插件 web 客户端面在该宿主无落点」这一事实。
  2. 什么原因导致的:宿主侧没有暴露任何结构化身份信号;唯一到达本车道的信号是宿主给每个子进程盖的 DSH_GO_RUNTIME_ROOT 环境变量(proot rootfs 路径)。
  3. 影响范围与严重程度:仅 dsh-go 宿主(第三方安卓 App);纯观测性缺口——之前完全无日志(可诊断性/体验瑕疵),不影响任何功能。
  4. 怎么修复的:纯函数 isDshGoHost(env 非空白即判定)放在既有软提示模块 client-build-note.ts,apply() 里按 #4545 同款「一次/进程 + fail-open」模式打一条 [dsh-client] info 日志;不 gate、不阻塞启动、不改行为。
  5. 是否有回归问题:无。改动全部为纯增量;DSH_GO_RUNTIME_ROOT 未设时(所有既有宿主形态)行为与日志逐字节不变。高危模式逐项过:无语义漂移、无 fail-fast 变吞错(fail-open 是设计意图且与 #4545 一致)、无阈值/时序/输出格式变化(唯一新输出是 dsh-go 宿主上多一条 info 行,已在 PR 正文披露)。本轮 rebase 重新核对:master 自旧 base 起未触碰本 PR 三个文件,纯 no-op rebase,无新增交互。
  6. 是否破坏缓存:不涉及(未触碰任何缓存/重试/增量游标/去重键路径)。
  7. 是否改了设计方向:不涉及——按 issue 定的方案原样落地(纯函数 + 一次日志 + 不加配置 + 不新建模块)。
  8. 是否引入了新的配置:无。DSH_GO_RUNTIME_ROOT 是第三方宿主盖的戳,不是 bili 的 env 开关/配置面(PR 明确「不加配置字段或 env 开关」)。

判定:可合(APPROVE)。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Re-review complete on the rebased head 310c1b9 — posted as APPROVE (review #5911824266).

八问与历轮一致(无变化,全文见 review 正文);判定:可合(APPROVE)。

OMSociety pushed a commit to OMSociety/billion-context that referenced this pull request Nov 16, 2026
@OMSociety
OMSociety force-pushed the 2026-11-14_dsh-go-host-note branch from 310c1b9 to a29c487 Compare November 16, 2026 07:27

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

Re-review on the rebased head a29c487 — APPROVE。

  • 该 head 是同样的两个 commit 干净 rebase 到当前 master 尖端 de895aa(#6889 合并)之上:feat 1c9f521、fix a29c487。
  • 两个 commit 补丁与我在历轮 head(b0cf22066 / 39a0338 / ddf376c / 50f9583 / 310c1b9)上审过并批准的状态逐字节一致(feat 1c9f521 ≡ ab3b1e8;fix a29c487 ≡ 310c1b9)——仅 index 行不同。
  • master 自上一 base f648217 以来对本 PR 触碰的 3 个文件零改动(git diff f648217cb8..origin/master -- src/agent/client-build-note.ts src/agent/dsh-native.ts tests/client-build-note.test.ts 为空)→ rebase 不引入新交互。
  • 新 base 上锚点全部复核:plan.mode === "off" 早退(src/agent/dsh-native.ts:1956)仍先于新块(约 2847 行);新块位于 #4545 dshNewerBuildNoted 块与 #1772/#3354 注释之间;once-per-process 标志(279 行)沿用既有模式;logHostEvent("info", "[dsh-client]", …) 与 src/agent/host-log.ts:26 的签名一致。
  • CI on head a29c487:37 success / 13 skipped / 0 failure(50 个 check run 全部完成;e2e 车道按三层 CI #3295 设计跳过)。
  • 实质不变:纯增量的 dsh-go 宿主一次/进程 [dsh-client] 启动注记,经纯函数、有单测的 isDshGoHost 判定;env 未盖章时(所有既有宿主形态)行为与日志零漂移;不碰 wire/config/持久化/kernel/缓存面;无 kernel 改动 → 无需决策升级流程。
  • 本 head 无新发现——无需顺手修。

结构化八问:

  1. 是什么问题:dsh-go(第三方安卓 App,自带 dsh 运行时)宿主上,bili.log 里没有任何能区分它与桌面宿主的地方;启动时打一条可辨认的 [dsh-client] 日志。
  2. 什么原因导致的:该宿主唯一能传到这条车道的信号是它给每个子进程盖的 DSH_GO_RUNTIME_ROOT env;除此之外无其他可区分点。
  3. 影响范围与严重程度:仅 dsh-go 宿主触发;纯日志、不改行为/流量/配置;属可诊断性增强,无外部行为影响(其他宿主 isDshGoHost() 恒 false)。
  4. 怎么修复的:纯函数 isDshGoHost(读 DSH_GO_RUNTIME_ROOT,空白视为未盖章、fail-open)+ apply() 里一次/进程的 logHostEvent 日志,接在 #4545 note 之后,沿用文件既有的 once-per-process 模式。
  5. 是否有回归问题:无。纯增量;无 wire/config/持久化/行为变更;try/catch fail-open,env 异常保持静默。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及——读的是宿主自带的 DSH_GO_RUNTIME_ROOT,不是 bili 新加的开关/env。

判定:可合(APPROVE)。合并动作归 owner,我不代合。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

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

Re-review complete on the rebased head a29c487 — posted as APPROVE (review #5916584850)。

八问与历轮一致(无变化,全文见 review 正文);判定:可合(APPROVE)。合并动作归 owner,我不代合。

OMSociety pushed a commit to OMSociety/billion-context that referenced this pull request Nov 17, 2026
@OMSociety
OMSociety force-pushed the 2026-11-14_dsh-go-host-note branch from a29c487 to 2e05b7a Compare November 17, 2026 10:17

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

Re-review on the rebased head 2e05b7a — APPROVE。

  • 该 head 是同样的两个 commit 干净 rebase 到当前 master 尖端 6a641f3(release v0.1.216 合并 #6938)之上:feat 3c60533、fix 2e05b7a。
  • fix 补丁逐字节一致(2e05b7a7ce ≡ 310c1b9);feat 补丁内容逐字节一致(3c60533ebe ≡ 1c9f521 ≡ ab3b1e8),唯一差异是 hunk 头行号(@@ -2867,6 +2868,23 @@ vs @@ -2846,6 +2847,23 @@)——master 在本块之前加了 21 行,块整体下移,新增行本身未变。
  • master 对 dsh-native.ts 的改动(#6880 purpose 声明:GateAttribution/LlmDispatchScope 加 purpose 字段、llm/stream 捕获、x-bili-plugin-purpose 头,位于 1221–2620 行)与本 PR 块(2877 行)无交互——不同机制、不同位置,本块只读 env 打一条日志,不触碰 dispatch/headers 路径。
  • 新 base 上锚点全部复核在位:plan.mode === "off" 早退(src/agent/dsh-native.ts:1961)先于新块;新块位于 #4545 dshNewerBuildNoted 块与 #1772/#3354 注释之间;once-per-process 标志(279 行)沿用既有模式;logHostEvent("info", "[dsh-client]", …) 与 host-log.ts:26 签名一致。
  • CI on head 2e05b7a:37 success / 13 skipped / 0 failure(50 个 check run 全部完成;e2e 车道按三层 CI #3295 设计跳过)。
  • 实质不变:纯增量的 dsh-go 宿主一次/进程 [dsh-client] 启动注记,经纯函数、有单测的 isDshGoHost 判定;env 未盖章时(所有既有宿主形态)行为与日志零漂移;不碰 wire/config/持久化/kernel/缓存面;无 kernel 改动 → 无需决策升级流程。
  • 本 head 无新发现——无需顺手修。

结构化八问:

  1. 是什么问题:dsh-go(第三方安卓 App,自带 dsh 运行时)宿主上,bili.log 里没有任何能区分它与桌面宿主的地方;启动时打一条可辨认的 [dsh-client] 日志。
  2. 什么原因导致的:该宿主唯一能传到这条车道的信号是它给每个子进程盖的 DSH_GO_RUNTIME_ROOT env;除此之外无其他可区分点。
  3. 影响范围与严重程度:仅 dsh-go 宿主触发;纯日志、不改行为/流量/配置;属可诊断性增强,无外部行为影响(其他宿主 isDshGoHost() 恒 false)。
  4. 怎么修复的:纯函数 isDshGoHost(读 DSH_GO_RUNTIME_ROOT,空白视为未盖章、fail-open)+ apply() 里一次/进程的 logHostEvent 日志,接在 #4545 note 之后,沿用文件既有的 once-per-process 模式。
  5. 是否有回归问题:无。纯增量;无 wire/config/持久化/行为变更;try/catch fail-open,env 异常保持静默;master 的 #6880 改动与本块无交互(已逐行核对)。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及——读的是宿主自带的 DSH_GO_RUNTIME_ROOT,不是 bili 新加的开关/env。

判定:可合(APPROVE)。合并动作归 owner,我不代合。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

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

Re-review complete on the rebased head 2e05b7a — posted as APPROVE (review #5937290244)。

八问与历轮一致(无变化,全文见 review 正文);判定:可合(APPROVE)。合并动作归 owner,我不代合。

OMSociety pushed a commit to OMSociety/billion-context that referenced this pull request Nov 17, 2026
@OMSociety
OMSociety force-pushed the 2026-11-14_dsh-go-host-note branch from 2e05b7a to 7a3b3d1 Compare November 17, 2026 12:34

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

Re-review on the rebased head 7a3b3d1 — APPROVE。

  • 该 head 是同样的两个 commit 干净 rebase 到当前 master 尖端 77fe0b1(#6592 合并——本 PR 在 #6588 里的「文档半边」)之上:feat 01c2a26、fix 7a3b3d1。
  • 两个 commit 补丁与我历轮批准的状态逐字节一致(连 index 行都没变——base 文件状态与上一 head 相同):feat 01c2a26 ≡ 3c60533 ≡ 1c9f521;fix 7a3b3d1 ≡ 2e05b7a ≡ 310c1b9。
  • master 自上一 base 6a641f3 以来对本 PR 触碰的 3 个文件零改动(git diff 6a641f3785..origin/master 对这 3 个文件为空)→ 这次 rebase 对 PR 内容是 no-op,不引入新交互。
  • CI on head 7a3b3d1:37 success / 13 skipped / 0 failure(50 个 check run 全部完成;e2e 车道按三层 CI #3295 设计跳过)。
  • 实质不变:纯增量的 dsh-go 宿主一次/进程 [dsh-client] 启动注记,经纯函数、有单测的 isDshGoHost 判定;env 未盖章时(所有既有宿主形态)行为与日志零漂移;不碰 wire/config/持久化/kernel/缓存面;无 kernel 改动 → 无需决策升级流程。
  • 本 head 无新发现——无需顺手修。

结构化八问:

  1. 是什么问题:dsh-go(第三方安卓 App,自带 dsh 运行时)宿主上,bili.log 里没有任何能区分它与桌面宿主的地方;启动时打一条可辨认的 [dsh-client] 日志。
  2. 什么原因导致的:该宿主唯一能传到这条车道的信号是它给每个子进程盖的 DSH_GO_RUNTIME_ROOT env;除此之外无其他可区分点。
  3. 影响范围与严重程度:仅 dsh-go 宿主触发;纯日志、不改行为/流量/配置;属可诊断性增强,无外部行为影响(其他宿主 isDshGoHost() 恒 false)。
  4. 怎么修复的:纯函数 isDshGoHost(读 DSH_GO_RUNTIME_ROOT,空白视为未盖章、fail-open)+ apply() 里一次/进程的 logHostEvent 日志,接在 #4545 note 之后,沿用文件既有的 once-per-process 模式。
  5. 是否有回归问题:无。纯增量;无 wire/config/持久化/行为变更;try/catch fail-open,env 异常保持静默。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及——读的是宿主自带的 DSH_GO_RUNTIME_ROOT,不是 bili 新加的开关/env。

判定:可合(APPROVE)。合并动作归 owner,我不代合。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

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

Re-review complete on the rebased head 7a3b3d1 — posted as APPROVE (review #5938925873)。

八问与历轮一致(无变化,全文见 review 正文);判定:可合(APPROVE)。合并动作归 owner,我不代合。

OMSociety and others added 2 commits November 19, 2026 00:28
The DSH Go Android app ships its own dsh runtime inside a proot rootfs and its
client is native, so this bundle's web client face (dsh.client.platform: web —
the Context tab and the settings page) has no surface to render into there,
while the host-side lane is unaffected. Nothing in bili.log — the only readable
record on that host — identified it, so a support read could not tell an
Android host from a desktop one.

Adds isDshGoHost() next to the lane's existing soft-note helpers in
client-build-note.ts (unit-tested decision table) and one once-per-process
[dsh-client] line in the dsh lane, keyed on DSH_GO_RUNTIME_ROOT. No behavior
change, no new config surface, no new src module.

Refs ranxianglei#6588
@OMSociety
OMSociety force-pushed the 2026-11-14_dsh-go-host-note branch from 7a3b3d1 to 161dd4f Compare November 18, 2026 16:28
@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

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

The failing check test (windows-latest, 22, 1) is a Windows flake, not caused by this PR. I've triggered a rerun of the failed job (zero code changes, PR branch untouched).

Evidence it's a flake, not the PR's fault:

  1. PR was green before — CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 floors 1544/1545: PR [PR] feat(dsh): detect dsh-go host and log once | feat(dsh): 识别 dsh-go 宿主并打一次日志 #6591 had fully green CI (37 green / 13 skipped / 0 failed) on heads 2e05b7a7ce and 7a3b3d183c.
  2. PR diff is a no-op for the failing tests — the PR adds isDshGoHost() (reads DSH_GO_RUNTIME_ROOT) + one log line. That env var is set ONLY in tests/client-build-note.test.ts (cleaned in finally), NOT in the two failing test files. So isDshGoHost() returns false there; the PR code is inert.
  3. logHostEvent can't hang — synchronous best-effort appendFileSync in try/catch (src/agent/host-log.ts:26).
  4. Both failures are Windows-specific:
  5. Master is green — CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 floor 1574: a PR on base a446a3c1c (= this PR's base = current master head) has 30/30 green. Not an inherited failure.

Action: rerun-failed-jobs on run 44347677508 (HTTP 201); test (windows-latest, 22, 1) is now queued. If it goes red again on a clean runner, it escalates to ci-fixes per the standard flake pattern (#4599 floors 1511/1512).

八问:

  1. 是什么问题: PR [PR] feat(dsh): detect dsh-go host and log once | feat(dsh): 识别 dsh-go 宿主并打一次日志 #6591 的 CI 检查 test (windows-latest, 22, 1) 红了。
  2. 什么原因导致的: Windows 平台特有的测试 flake(非 PR 引入)——bug: dsh-native — attach 目标瞬时不可达 → 回退 spawn 第二实例:routed baseURL 下工具/模型通道必然脑裂(工具 404 → preflight 502 → 会话不可自愈) #1365 attach-mode respawn 测试的 event-loop 时序竞态 + SQLite unified storage degraded to legacy-only (bili v0.1.208, node v24.21.0, win32-x64) #4816 测试的 Windows teardown 文件锁 EBUSY(index.db 被锁删不掉)。PR 的 diff 对这两个失败测试是 no-op(DSH_GO_RUNTIME_ROOT 只在 client-build-note.test.ts 里设,不在失败测试里;logHostEvent 是同步 best-effort,不可能 hang)。
  3. 影响范围与严重程度: 仅影响 CI 的 Windows 分片,不影响用户;PR 代码本身没问题(此前 head 2e05b7a/7a3b3d183c 全绿)。
  4. 怎么修复的: 触发 rerun-failed-jobs 重跑失败 job(零代码改动,PR 分支未动)。
  5. 是否有回归问题: 不涉及——没有改任何代码。
  6. 是否破坏缓存: 不涉及。
  7. 是否改了设计方向: 不涉及。
  8. 是否引入了新的配置: 不涉及。

判定: PR 代码无问题,CI 红是 Windows flake。已触发重跑,等结果。若干净 runner 再红则升档 ci-fixes。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

Re-review on the rebased head 161dd4f — APPROVE。

  • 该 head 是同样的两个 commit 干净 rebase 到当前 master 尖端 a446a3c(#6827 batch11-wave11 合并)之上:feat e783755、fix 161dd4f。
  • 两个 commit 的新增行内容与我历轮批准的状态逐字节一致(feat ≡ 01c2a26 ≡ 3c60533 ≡ 1c9f521;fix ≡ 7a3b3d1 ≡ 2e05b7a ≡ 310c1b9);唯一差异是 feat 的 hunk 头行号(@@ -2877,6 +2878,23 @@,较上一 head 下移 10 行)——与 master 在本块之前净加 10 行一致(#6770 的 compactionCoverageNoted 标志 +6、attributionOf 内覆盖读数 +9、gateCoverageSuffix 收缩 -5)。
  • master 对 dsh-native.ts 的改动(#6770 压缩覆盖读数:import 换 compactionCoverageText、gateCoverageSuffix 委托、首个路由请求打一次覆盖读数、_resetCompactionCoverageNoteForTest 钩子,位于 70/294/1251/1341/3129 行)与本 PR 块(2887 行,只读 env 打一条日志)不同机制、不同位置,无共享状态、无交互。
  • 新 base 上锚点全部复核在位:plan.mode === "off" 早退(src/agent/dsh-native.ts:1971)先于新块;新块位于 #4545 dshNewerBuildNoted 块(2875 行结束)与 #1772/#3354 注释(2898 行起)之间;once-per-process 标志(279 行)沿用既有模式;logHostEvent("info", "[dsh-client]", …) 与 host-log.ts:26 签名一致。
  • CI on head 161dd4f:37 success / 13 skipped / 0 failure(50 个 check run 全部完成;e2e 车道按三层 CI #3295 设计跳过)。
  • 实质不变:纯增量的 dsh-go 宿主一次/进程 [dsh-client] 启动注记,经纯函数、有单测的 isDshGoHost 判定;env 未盖章时(所有既有宿主形态)行为与日志零漂移;不碰 wire/config/持久化/kernel/缓存面;无 kernel 改动 → 无需决策升级流程。
  • 本 head 无新发现——无需顺手修。

结构化八问:

  1. 是什么问题:dsh-go(第三方安卓 App,自带 dsh 运行时)宿主上,bili.log 里没有任何能区分它与桌面宿主的地方;启动时打一条可辨认的 [dsh-client] 日志。
  2. 什么原因导致的:该宿主唯一能传到这条车道的信号是它给每个子进程盖的 DSH_GO_RUNTIME_ROOT env;除此之外无其他可区分点。
  3. 影响范围与严重程度:仅 dsh-go 宿主触发;纯日志、不改行为/流量/配置;属可诊断性增强,无外部行为影响(其他宿主 isDshGoHost() 恒 false)。
  4. 怎么修复的:纯函数 isDshGoHost(读 DSH_GO_RUNTIME_ROOT,空白视为未盖章、fail-open)+ apply() 里一次/进程的 logHostEvent 日志,接在 #4545 note 之后,沿用文件既有的 once-per-process 模式。
  5. 是否有回归问题:无。纯增量;无 wire/config/持久化/行为变更;try/catch fail-open,env 异常保持静默;master 的 #6770 改动与本块无交互(已逐行核对)。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及——读的是宿主自带的 DSH_GO_RUNTIME_ROOT,不是 bili 新加的开关/env。

判定:可合(APPROVE)。合并动作归 owner,我不代合。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

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

Re-review complete on the rebased head 161dd4f — posted as APPROVE (review #5962821913)。

八问与历轮一致(无变化,全文见 review 正文);判定:可合(APPROVE)。合并动作归 owner,我不代合。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 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.

2 participants