Skip to content

test(ci): integrate canonical fixtures and architecture scans - #6344

Merged
huangruiteng merged 8 commits into
mainfrom
codex/ci-fixture-owner-20261013
Oct 13, 2026
Merged

huangruiteng merged 8 commits into
mainfrom
codex/ci-fixture-owner-20261013

Conversation

@loopx-agent

@loopx-agent loopx-agent commented Oct 12, 2026 •

Copy link
Copy Markdown
Collaborator

The full Python suite has stale canonical-delivery and host-ACK fixtures, an analysis-boundary false positive, and repeated architecture scanning costs. This test-only proposal composes the existing changes from #6328, #6332 and #6341 with the canonical fixture repair in this PR. It changes twelve test files; production code, workflow permissions, collection, four shards, two workers and the 60-minute timeout remain unchanged.

Canonical fixtures retain independent Goal intent when displayed Todo rows disappear, use retained readers, and recover a lost response by reading the original operation. File/SQLite negative cases preserve refusal with missing Goal intent. Settlement fixtures supply the existing isolated physical host manifest: missing/mismatched cadence cannot write state or consume another slot, and matching facts cannot revive a superseded receipt. Directory lifecycle scans remain scoped to their synthetic project.

Architecture guards release consumed syntax/scope state and reuse complete findings within each existing module. Syntax failure, corrected-source retry, defensive-list isolation, duplicate-owner and declaration checks remain. Sonar checks the existing successful same-run coverage dependency and rejects reachable privileged analysis/classification/coverage or status bypasses; metadata-only cleanup remains isolated. No general cache or expression engine is introduced.

Validation: 18 complete relevant modules passed (683 tests, two workers with coverage, 724.27 seconds). Independent old/new helpers agree on all five digest facets and every field binding/member/string vector over 1,276 byte-identical product Python modules. Full source collection gives a disjoint four-shard union of 21,658 cases versus 21,645 in the immutable baseline; the original Sonar case is renamed and 13 net validation cases are added. Missing duration history keeps the existing fallback. Four direct native checks and nine required selected smokes pass; Ruff, compile, diff, full semantic and public-boundary checks pass.

Qualification remains open: the existing frontier maintainability advisory still fails at 2007/2000; the explicit UTF-8 guard and reviewed migration-debt oracle also need the separate source repair #6347. Its approval does not certify this assembled head. The naturally completed full run on #6274 had three job-limit annotations and two complete failing JUnit artifacts; collection and local affected tests do not establish complete hosted performance, installed adoption or shared/release capacity. The exact-scope quality receipt retains these failures and holds. No raw logs, profiles, private state or coverage artifacts are included; a new exact-head review is required.

Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
@loopx-agent
loopx-agent force-pushed the codex/ci-fixture-owner-20261013 branch from c4a9dde to faf793c Compare October 12, 2026 23:47

@loopx-agent loopx-agent left a comment

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

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

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)
Reviewer: model_agent; model=gpt-6.1-sol; provider=OpenAI; runtime_reported; reasoning_effort=xhigh

动机

全 PR 当前 head faf793c5bf9b639b69801e07c824c8648bef6424,七个既有测试文件 +65/-16。维护 canonical delivery 与完整 CI qualification 的贡献者。 旧 reader patch 没有命中实际调用,fixture 删除 Goal 文件,旧异常和可选 stride 字段断言产生确定性失败。 修正真实消费者与 fixture,保留独立 Goal 源、原 operation recovery、readonly/provider-loss/lease/settlement 边界;不改产品运行规则。 改动前依据是 docs/architecture/rfcs/shared-goal-authority-state-provider-v0.md @ fee889e617e1d063b42acff34b46445c49cee062:5 原 operation 身份、10.7 丢失响应查原 receipt、10.13 不向无 fence 的 writer 回退;独立 Goal narrative 仍不由 Todo head 取代。Quality guide 的 runtime isolation/durable-smoke 合同及 S2/S12 保留。作者验证不是验收 oracle。

完整 hosted/installed/shared admission/release capacity 和作者所述三项原全量失败仍开放;本轮没有认证性能、CPU或长期收益。 本评审不授予 merge/升级权限,不关闭完整 CI、provider promotion、D1–D3 或发布容量验收。

改动思路

当前有界交付是七个既有测试文件的 consumer/fixture 修复与明确反例,不改变 production、workflow、预算、并发数或 timeout。 直接在原测试调用点复用 retained reader、原 receipt 命令和显式 scan-path 即足够;不新增 helper/cache/policy 或独立 smoke scaffold。 Canonical Todo facts 可以替代显示行,不能替代不可读的 Goal 来源;用 prefix 保留目标与约束。改 monkeypatch 指向正在调用的 list_readback,而不是退休 facade;响应丢失沿现有 typed error/原 operation lookup。接受边界没有随测试改写。

具体改动

lease inspection 和 Goal attention 两处改为当前 reader;writer-retired 包和 quota/migration journey 只删除显示 Todo 行,保留 Goal 源。新增 File/SQLite 缺 Goal 文件负例:Todo readonly lookup 仍相同,guard exit1、无 selected work/receipt/spend且不补写 Goal。Validated-create 的 lost-response 测试检查 typed code、Todo/operation身份,执行原 recovery CLI 并证明 canonical state未再写,原修订 validator仍被投影。Run-index 在原字段全保留的基础上明确既有 stride_decision=null。

Directory lifecycle 的成功与 unavailable-provider 调用均加显式 --scan-path 指向同一合成项目,仍经历 open/lease/release/blocked/deferred/done、失源拒绝及恢复,read-only revision/todos 断言未删。它不改变 CLI扫描规则,也不绕过 public boundary检查。

我的独立八模块(七 changed +邻近 canonical planning)168 passed;相同 B 11 failed、155 passed,失败定位至当前修正的真实调用/oracle,不借用作者184计数。把当前新增 missing-source 两case单独放到 immutable B 也 2 passed,证明原拒绝边界一直成立。Whole production/workflow/docs/skills 字节与 B 相同。4 direct、10 selected canary、Ruff与完整 semantic通过。一次初始验证命令误用了不存在的邻近模块而未收集测试,保留诊断并纠正到实际模块后才建立上述证据。

对主干的风险

风险是为了绿测试而削掉 Goal-source、provider-loss、原身份或未知字段断言。当前真实阴性/readback和完整原字段比较保住这些边界;新missing-source只证明整个来源文件不可读,不夸成任意 Goal文本语义已认证。既有 writer-retired source arm实际执行,wheel arm/完整安装态没有认证;File/SQLite结果不关闭共享/PostgreSQL/D1–D3验收。

Durable value:扫描既有 canonical lease/planning/validated-create/retired-directory 与 projection覆盖、完整同作者68队列及近期6341 Sonar测试修复。这里修复可复现 canonical validation失配,复用原文件/helper并增加两强负例,与Sonar依赖/权限守卫是不同边界;本批没有同形重复 scaffold或已警告后继续同类违例证据,不建议贡献限制。Future-facing pass保留原调用点scope,不扩展cache/框架。并行B/H是correctness执行,不拿耗时比较性能;作者local wall-time/CPU/p95/hosted收益未独立认证。

三项作者报告的全量失败保持记录,不用此局部通过把 frozen CI24、S12或release hold关闭。没有CI query/poll/wait。

我的整体评价

APPROVE 当前测试维护切片。它让既有真实qualification继续推进,长程价值是减少误报并保留安全反例;product规则和用户操作未变。当前有界交付是七个既有测试文件的 consumer/fixture 修复与明确反例,不改变 production、workflow、预算、并发数或 timeout。 直接在原测试调用点复用 retained reader、原 receipt 命令和显式 scan-path 即足够;不新增 helper/cache/policy 或独立 smoke scaffold。

English verdict: APPROVE — faf793c5bf9b639b69801e07c824c8648bef6424. Independent B11fail/155pass versus H168pass plus unchanged B/H missing-source denial validates bounded consumer/fixture repair. Whole hosted, installed/shared promotion and performance qualification remain open.

…r-20261013

Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
@loopx-agent loopx-agent changed the title test(control-plane): qualify current canonical delivery boundaries test(control-plane): qualify canonical delivery and host ACK fixtures Oct 13, 2026
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>

@loopx-agent loopx-agent left a comment

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

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

Reviewer: model_agent; model=gpt-6.1-sol; provider=OpenAI; runtime_reported; reasoning_effort=xhigh

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)

动机

维护 canonical delivery 与完整 CI 资格化的开发者。 旧夹具删除 Todo 显示行时连独立 Goal 目标也删掉,仍调用退休 reader,或者期望原响应丢失异常;结算 ACK 的成功场景没有提供当前 Host 清单。 现在保留 Goal 来源、读取真实当前 reader、沿原 operation 找回回执;缺失或不同 Host cadence 会拒绝 ACK,匹配 cadence 后可按原身份首次写入并幂等重放。 不改 production、workflow、预算、权限或实际 App automation,也不证明安装态、共享 provider、完整 CI、性能或发布资格完成。 三个已有 main-source 资格化失败、完整 hosted/installed/shared admission、自然 Host 采用和发布容量仍保持开放。

改动思路

本次是八个原测试文件的 oracle/fixture 修复,恢复现有真实消费者的验证能力,独立可审阅与回退;原更大资格化仍归已有 owner。 最小修复是在原测试里保留 Goal 目标与约束、命中正在调用的 reader,并读取原 operation 回执;不需要另建恢复 owner、放宽权限、删除旧断言或扩展框架。ACK 的有效输入来自隔离的合成 Codex heartbeat 清单,真正的 CLI reader 与 TypeScript scheduler 负责核验。夹具证明文件观察和准入分支,不证明桌面 App 曾更新 automation。

旧 faf head 的批准不沿用。相对上次评审,七个 blob 同、quota settlement 测试一个 blob 更新;当前实际 fork 是 71304fcbdd7402cfd4355fffc397079fe37d0c34,全八路径重新判断。源代码、workflow、docs、skills 无差异;前瞻整理复用现有 helper 和直接 scoped call site,不增加兼容分支。

具体改动

关键内容讲解

依据 docs/development/testing-and-quality.md @ 71304fcbdd7402cfd4355fffc397079fe37d0c34:Quality Layers、State Composition Qualification、Isolate the managed Effect process as well as the data 均 implemented,分别对应独立不变量、真实有界故障/回读和隔离验证。另核 accepted canonical RFC 的 §5 原 operation 身份及 Todo-section-only 投影边界;Todo 不能恢复丢失的独立 Goal narrative。新测试没有把候选输出生成成 oracle。

  • canonical lease inspection 和 Goal attention 两处 monkeypatch 改为实际 list_readback.list_goal_todos;释放后必须重读 lease,inactive 分支不能解析 Todo 历史。
  • writer-retirement 和 quota authority journey 保留 Goal prefix,只移除显示 Todo 行;新 File/SQLite missing-Goal 反例证明 readonly Todo 仍可读,但 guard exit1、无 selection/heartbeat receipt/spend、无补写 Goal。
  • validated-create 的 lost-response 分支在真实 commit 后注入响应丢失,期望已有 typed unavailable code 与原 Todo/operation,执行返回的原 recovery CLI;必须找到原 receipt、全 canonical state 无第二次写、精确 validator 声明仍可投影。
  • quota settlement 的普通与 Todo-less 路径复用既有 write_codex_heartbeat_fixture。missing/mismatch cadence 无 scheduler 文件、仍只有一 debit;matching manifest 不能复活被新 Turn 替代的 receipt。有效当前 recipe 可首次 ACK,重放不再写文件并保持字节与扣数。
  • retired directory 的正常/unavailable-provider 两调用都把 public scan 指向该合成项目,原 lease/blocked/deferred/done、失源拒绝和 readonly 断言保留。run-index 对原全部字段比较增加已存在的 stride_decision=None,没有删字段。

全 12a1988628dfa2de921914bffc24e89da886d861 为八既有测试文件 +105/-18。既有 canonical/retirement/recovery、scheduler 和 projection 测试族已扫描;同作者100条 PR 摘要及完整70队列的相关批次也已核对。6340 同包缺席和6341 Sonar隔离是另一边界,本 PR 没有新建同形 walkthrough 或重复框架。

对主干的风险

没有引入的阻塞发现。独立9完整模块(8 changed + canonical planning)初跑 267 passed /2 failed;两失败均为我把隔离临时目录放在仓库内,使“gitless project”正确识别到父仓库。基线同样有这两失败,实际 Git root 已核对。只将相同两断言移到仓库外物理临时目录后,head/base 各 2 passed;没有屏蔽 Git 发现、删断言或把首次失败改写为全绿。当前269项的资格由上述分开的原始运行构成。

基线完整9模块首次 16 failed /251 passed,其中两项是上述目录条件。其余14个具体旧夹具失败在与head相同 Python3.12、仓库外隔离根再执行仍 14 failed /2 gitless passed,对应当前267通过的真实 consumer/Goal/ACK/typed-recovery/optional字段修复。当前新 missing-Goal 两断言也单独放到不可变base执行 2 passed,说明 source拒绝一直由原产品 owner 强制,而不是补fixture来抹掉边界。旧3.11基线记录保留,不用不同解释器的汇总数冒充同环境因果比较。

TypeScript scheduler 四模块 51 passed /0 skip,4direct/10selected 风险检查、Ruff、完整 semantic/public 边界通过,CI 未查询/轮询/等待。原三项更广 required source 失败另行在base/head复核并保留:stride schema UTF-8、既有迁移债务与零例外 oracle 矛盾、frontier 2007/2000。因果 source/validator blob 不变、失败身份/细节同,独立当前夹具覆盖成立;这不放宽 full CI 门槛,也不把 separate6347 未集成改动当作当前修复。

本轮是 correctness 执行;并行与不同目录下的时间不用于声称 CPU/p95/整量 CI 加速。合成 manifest、source-only receiver、File/SQLite CLI 和真实原回执观察不能关闭安装态 App、shared/PostgreSQL/自然 Host、全 Goal 或长期收益验收。未改 production/store owner,故没有借局部测试冒充完整 provider/refactor promotion。

我的整体评价

APPROVE 于 12a1988628dfa2de921914bffc24e89da886d861,justified_increment:长程验证价值 improved,普通用户体验 preserved。它消除十四个实际旧 oracle/fixture 失配,并保留 Goal-source、Host facts、原身份、失源与单 debit 边界。维护成本是原测试局部修改,runtime 前瞻重构无当前必要;没有新的 shared closed vocabulary、默认策略或权限。完整 CI/安装采用/性能与发布容量、三个原 required source 失败继续开放。本批准不是合并、升级或发布授权。

English verdict: APPROVE — 12a1988. Eight existing fixture families now use current readers, preserve independent Goal intent and original-operation recovery, and exercise actual manifest-bound ACK refusal/first commit/replay. Current qualification is 267 passed plus two outside-Git cases passed; fourteen matching-interpreter baseline fixture failures remain baseline-red and are covered at the head. 51 typed scheduler cases and scoped risk checks pass. Three unchanged broader source failures and installed/shared/full-CI/performance qualification remain open.

@loopx-agent loopx-agent changed the title test(control-plane): qualify canonical delivery and host ACK fixtures test(ci): integrate canonical fixtures and architecture scans Oct 13, 2026
@huangruiteng
huangruiteng merged commit db5a298 into main Oct 13, 2026
4 of 5 checks passed
@huangruiteng
huangruiteng deleted the codex/ci-fixture-owner-20261013 branch October 13, 2026 03:37

@loopx-agent loopx-agent left a comment

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

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

Reviewer: model_agent; model=gpt-6.1-sol; provider=OpenAI; runtime_reported; reasoning_effort=xhigh

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)

动机

维护 canonical delivery 和架构验证的开发者需要测试真实当前消费者。旧夹具调用退休 reader、删除 Todo 显示行时丢掉独立 Goal 目标、把响应丢失当成旧异常,或者缺少有效 Host 清单;架构测试还重复分析模块并保留不再消费的 AST/scope 数据。Sonar 的一刀切 trigger 约束混淆安全的 metadata cleanup 与有权限的分析。这一集成恢复现有验证能力并整理同一测试边界,未改变产品、workflow、默认权限或预算。

这里只资格化 test-only correctness。没有 CPU/p95/长期性能实测、完整 hosted suite、PostgreSQL/自然 Host/实际 App 更新、frontend/Lark 或模型长期效用证明;这些不因为测试数量或本次批准而关闭。完整 hosted/installed/shared CI、自然 Host、性能与发布资格继续开放。

改动思路

当前 456782950fa3acf97b93cbba0b6ba248c71026fe 相对实际 base 71304fcbdd7402cfd4355fffc397079fe37d0c34 为十二个既有测试文件 +410/-144。先前 12a1988628dfa2de921914bffc24e89da886d861 的八文件批准只适用于旧 head;它发布后 head 变成此次集成版本,native closeout/readiness 已因 head mismatch 拒绝沿用。本次重新审全部十二文件及当前八个 capability lens。

八个 canonical 文件和所有生产源码保持字节相同,因此既有独立执行证据仅在消费者、产品、解释器与输入条件逐项确认不变后复用,未继承旧 verdict。新增四模块完整执行,并用不可变 base 的真实 scanner 与当前版本逐模块比较。前瞻整理就在原 test-local owner 中完成:一次提取消费向量、只发布完整 findings、释放不再消费的 scope,未增加运行时缓存或框架。完整 hosted/installed/shared CI、自然 Host、性能与发布资格继续开放。

核验期间 PR 已合并,remote readback 的 merged head 仍是上述 commit。因此这是该精确 commit 的事后评审;我没有执行 merge。这份结果不能倒推合并发生时已有完整当前 head 的批准。旧 head 评审记录保留,head-change 资格缺口另行记录。

具体改动

关键内容讲解

依据 docs/development/testing-and-quality.md @ 71304fcbdd7402cfd4355fffc397079fe37d0c34:Quality Layers、State Composition Qualification、Isolate the managed Effect process as well as the data 均 implemented,分别对应独立不变量、真实有界入口/反例/回读及隔离验证。另核 accepted canonical RFC §5 原 operation 身份与 Todo-section-only 投影边界;canonical Todo 不能恢复被删掉的独立 Goal narrative。

  • lease inspection 和 Goal attention 改为实际 list_readback.list_goal_todos,释放后重读 lease,inactive 分支不得读 Todo 历史。writer-retirement/quota journey 保留 Goal prefix;新 File/SQLite missing-Goal 反例要求 readonly Todo 可读,但 guard 拒绝、无 selection/receipt/spend/补写 Goal。
  • validated-create 在真实 commit 后注入响应丢失,核 typed unavailable code、原 Todo/operation,执行原 recovery CLI 找回 receipt;全 canonical state 和 validator 精确声明不得出现第二次写入。
  • quota settlement 复用原合成 heartbeat manifest helper,真实 CLI reader/TS scheduler 验证 missing/mismatch cadence、被新 Turn 替代的原 receipt、首次 ACK 与字节相同的幂等 replay,始终一 debit。合成清单证明 reader/准入分支,不证明实际桌面 App 更新过 automation。retired-directory 调用限定实际合成 scan scope;projection 比较保留所有旧字段并补已有的 stride_decision=None。
  • content-digest analyzer 返回五个实际消费 facet,移除没人读取的 tree/root/raw-source cache;全部 reader 完成之后才释放 AST scope/binding 环。每个 facet 在全部实际源码上与 base 相同。
  • field vocabulary 与 wire schema scanner 一次提取现有 binding/member/body/literal 向量,用模块 fixture 共享完整扫描结果,owner exclusion 与精确 inline site count 不变。syntax error 不产生部分缓存,修复后可以重试;findings 返回副本,调用者清空列表不能破坏后续查询。新 duplicate/member/body/count 变体保留拒绝语义。
  • Sonar 测试读取真实未修改的 YAML,限定分析 workflow 的复用入口、同 run 的 pytest dependency 和实际 condition,无 status bypass。只解释两个实际 shipped 条件;metadata-only cleanup 可以扩展事件而不能开启分析。trigger/classification/dependency/coverage/status 的突变都要求拒绝,原 token/checkout/permission/cleanup 断言保留。

扫描既有 canonical/retirement/recovery、scheduler、架构及 Sonar 测试族,并读完整70项队列和同作者100条 PR 摘要。当前集成复用原文件,6340 同包缺席与6341 workflow 隔离各有不同因果边界;没有证据支持同形 fixture farming 或贡献限制。

对主干的风险

新增四个完整模块独立 285 passed /0 skip。使用标准 ast.parse,将 base 与当前 helper 放在相同真实产品树上执行:全部 1276 个 Python 模块、34452 个 consumed facet/vector 比较一致,包括 digest 五个 facet、field binding/member/body、wire binding/literal;不是作者683/21658计数、mock parser 或由当前输出反造的 oracle。完整新测试另覆盖 scan 集成、失败重试、副本突变和规则反例。固定源码的模块快照不会支持运行时动态 reload,这里也没有此调用契约。

八个 canonical 文件的独立9模块初跑 267 passed /2 failed:我把“gitless project”的临时根放进仓库,实际 Git root 正确指向父仓库;相同两案例在 base 同样失败。核实原因后,仅移到仓库外物理临时根,head/base 各 2 passed,未屏蔽 Git 发现、删断言或改写原失败。当前269项由这两次原始运行分别资格化;八个 blob、生产入口与环境在集成 head 未变。

初次 baseline 的 16 failed /251 passed 含上述两项,并使用3.11;其余14个具体旧夹具失败在与 head 相同3.12和仓库外临时根重验,仍 14 failed /2 gitless passed,十四个对应消费者案例在 head 通过。当前新增 missing-Goal 两案例放到不可变 base 也 2 passed,source 拒绝一直由产品 owner 强制。保留初跑记录,不用不同解释器汇总冒充因果比较。

四个 typed scheduler 模块 51 passed /0 skip,源码与测试未受新增四路径影响。当前风险检查 4 direct /10 selected 的 required checks 通过,Ruff、完整 semantic/public 边界通过;CI 未查询、轮询或等待。原更广的三个 required source 失败仍保留 failed:stride schema UTF-8、既有 legacy mutation 向 explore 的迁移债务与零例外 oracle、frontier 2007/2000。本轮 base/旧 head 的相同命令失败身份和细节已匹配,集成版本逐项核实 causal source/validator blob 不变;current canary 也保留同一 frontier advisory failure。没有提高阈值、waiver full CI,或把未集成的6347 source proposal 当成当前修复。

这里只资格化 test-only correctness。没有 CPU/p95/长期性能实测、完整 hosted suite、PostgreSQL/自然 Host/实际 App 更新、frontend/Lark 或模型长期效用证明;这些不因为测试数量或本次批准而关闭。

我的整体评价

APPROVE 于 456782950fa3acf97b93cbba0b6ba248c71026fe,justified_increment。十四个实际旧夹具失配被修复,Goal-source、Host facts、原身份 recovery 和单 debit 边界保留;架构扫描的实际语义向量不变,完整错误重试/副本/工作流反例通过。长程验证价值 improved,普通用户体验 preserved。前瞻整理已在现有测试 owner 内完成,不需要新共享词汇、运行时架构或默认行为变更。三个原 source 失败及完整安装/CI/性能/发布验收仍开放;本批准不授予合并、升级或发布权限。

English verdict: APPROVE — 4567829. The twelve-file test integration repairs current canonical fixtures and consolidates existing architecture/workflow checks without product changes. All four added modules pass285 cases; immutable-base and current helpers agree across1276 real modules and34452 consumed facet/vector comparisons. The unchanged eight-file qualification remains267 passed plus two separately qualified outside-Git cases, with51 typed scheduler cases. Three inherited broader source failures and hosted/installed/shared/full-CI/performance qualification remain open; the old-head approval is not inherited.

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.

2 participants