Skip to content

fix: sync resume-args branch and stabilize reader undo test - #14

Open
Cec1c wants to merge 54 commits into
WilliamWang1721:codex/issue-323-agent-resume-argsfrom
Cec1c:fix/pr457-resume-followup
Open

Cec1c wants to merge 54 commits into
WilliamWang1721:codex/issue-323-agent-resume-argsfrom
Cec1c:fix/pr457-resume-followup

Conversation

@Cec1c

@Cec1c Cec1c commented Oct 9, 2026 •

Copy link
Copy Markdown

Advertising, commercial promotion, and traffic solicitation are prohibited in PRs
and Issues. Business requests must go through the email in the
contribution policy.
PRs or Issues containing unauthorized promotion will be closed directly.
严禁利用 PR 或 Issue 推广、引流,未经授权的推广将直接关闭;商业需求请通过贡献指南中的邮箱沟通,已获批准的赞助按批准范围处理。

Result / 用户结果

Follow-up for Kuddev/pebrel#457, targeting the original author's feature branch. This brings the resume-arguments work onto upstream 244d6fe1 and fixes the unrelated reader undo test that failed on Intel Mac. The existing feature can restore Codex with explicitly configured ["--yolo"]; no new resume behavior is introduced here.

协助收尾 Kuddev/pebrel#457,目标为原作者的功能分支。同步上游 244d6fe1,并修复 Intel Mac 上失败的、与恢复参数无关的阅读器撤销测试。原有功能已能按显式配置的 ["--yolo"] 恢复 Codex;本 PR 未新增恢复行为。

Design / 设计边界

  • c4ad676f merges upstream main without rewriting the author's history. The only manual conflict resolution keeps both upstream's renamed completion test and this feature's settings test in .config/nextest.toml.

  • 61b2d1f5 changes one editor test fixture (+5/-1): prepare Updated using one clipboard paste. Simulated per-character typing can cross the production 600 ms undo-group interval. All original reader/source/undo assertions remain.

  • No production editor, dependency, persistence, threading, or compatibility changes are added by the follow-up fix.

  • The comparison against the author's older branch includes upstream changes (117 files, +5287/-992). Against the synchronized upstream base, the entire feature plus this fix is 23 files, +703/-7. The imported upstream changes are not newly authored work in this PR.

  • c4ad676f 合并上游 main,保留原作者提交历史。唯一手工冲突处理位于 .config/nextest.toml,同时保留上游重命名的补全测试和本功能的设置隔离测试。

  • 61b2d1f5 仅修改一个编辑器测试的准备步骤(+5/-1):通过一次剪贴板粘贴输入 Updated。原逐字模拟输入可能跨过生产代码的 600 毫秒撤销分组间隔;原阅读器、源码及撤销断言全部保留。

  • 此收尾修复没有新增编辑器生产逻辑、依赖、持久化、线程或兼容性变更。

  • 相对作者旧分支的比较包含上游更新(117 个文件,+5287/-992);相对已同步的上游基线,完整功能与此修复共 23 个文件,+703/-7。导入的上游更新不是本 PR 新编写的改动。

Evidence / 验证依据

  • Reproduced the original Intel Mac failure: inserting 650 ms between simulated Upda and ted makes undo yield Upda instead of Original. The affected baseline test is identical on upstream main. The delay is not retained in the fix.

  • Windows, Rust 1.97.1: 94 editor tests passed after the fix; 31 resume-related tests passed, including real settings controls and PowerShell 5.1/pwsh argument round trips. cargo test -p nebula-settings passed 99 tests; file-line-budget passed 2; isolated i18n contract passed 24 (1 ignored). Cargo workspace checks used .github/ci-profile.toml, profile ci, and --locked.

  • py -3.12 -m unittest discover -s scripts/tests: 289 tests, OK, 61 platform-specific skips. Architecture checks against both upstream and author bases passed. cargo fmt --all -- --check and git diff --check passed. The actual GPUI product was built with --features gpui-shell.

  • Real Windows acceptance used two synthetic saved conversations and isolated Pebrel/Codex directories with the installed Codex CCU 0.157.1 build. Both panes showed their own transcript marker, correct cwd, and Full access; the actual native commands contained the exact session IDs and --yolo. Abrupt process restart restored both again without reseeding the saved session. Clearing arguments to [] removed --yolo from both native commands.

  • Limits: no live model requests or vanilla Codex acceptance; no claim that clearing flags resets Codex's own saved permissions. The public Runtime API refuses to close a busy Codex window, so the GUI-confirmed clean-close path was not exercised. Some hook-disabled runtime snapshots omitted IDs while the native command, transcript, and persisted recovery identity were correct. No visual layout or hot-path behavior changes are introduced by this follow-up.

  • Final-head fork CI: full native tests passed on all five platforms, including both macOS release-workspace checks. The Intel Mac job ran 2,758 tests: all passed, with 27 platform/feature skips; its log explicitly records PASS for both the fixed reader test and the resume-settings control test. Architecture, PR size, and lint passed. Validation-only PR compares against the synchronized upstream base and was closed without merging after validation completed. This helper PR is ready for the original author to review.

  • 已复现原 Intel Mac 失败:在模拟输入 Upda 与 ted 之间延迟 650 毫秒,撤销结果为 Upda,而非 Original。该基线测试与上游 main 完全相同;最终修复不保留延迟。

  • Windows、Rust 1.97.1:修复后 94 项编辑器测试通过;31 项恢复相关测试通过,包含真实设置控件及 PowerShell 5.1/pwsh 参数往返。设置模块 99 项、文件行数预算 2 项通过;独立翻译合同 24 项通过、1 项忽略。工作区 Cargo 检查使用 .github/ci-profile.toml、ci profile 与 --locked。

  • Python 脚本套件共 289 项,结果 OK,61 项因平台跳过。以当前上游及作者分支分别为基线的架构检查通过;格式与差异空白检查通过;已构建 --features gpui-shell 的真实产品。

  • Windows 实测使用两份合成历史会话、隔离的 Pebrel/Codex 目录及本机 Codex CCU 0.157.1。两个面板分别显示自己的历史标记、正确目录和 Full access,原生进程命令包含精确会话 ID 与 --yolo。受控终止后不重写保存的会话,再启动仍恢复正确;清空为 [] 后,两条原生命令均不再携带 --yolo。

  • 边界:未发送真实模型请求,未验收原版 Codex;不声称清空参数会重置 Codex 自己保存的权限。Runtime API 拒绝关闭忙碌的 Codex 窗口,因此未覆盖 GUI 确认后的正常关窗路径。关闭 hooks 时,部分运行时快照未填会话 ID,但原生命令、会话历史及持久化恢复身份均正确。此收尾不引入布局或热路径行为变更。

  • 最终提交的 fork CI:五平台原生测试与两个 macOS 发布工作区检查均已通过。Intel Mac 任务运行 2,758 项测试并全部通过,另有 27 项平台或功能条件跳过;日志明确记录了修复后的阅读器测试及恢复设置控件测试通过。架构、PR 大小及 lint 已通过。仅供验证的 PR以上游同步基线进行比较,验证完成后已关闭且未合并。本协助 PR 已准备好供原作者审阅。

Required Review / 必须确认

  • I followed CONTRIBUTING.md, docs/architecture.md, and docs/project-constraints.md.

  • I split responsibilities, not arbitrary line ranges; no duplicate behavior authority was added.

  • python3 scripts/check_architecture.py --base <PR-base-commit> passes; budgets were not inflated to fit the change.

  • Tests cover success and failure; platform/feature coverage limitations are stated.

  • New messages use typed i18n IDs and matching placeholders; untranslated content has an explicit fallback. No new UI messages in this follow-up.

  • Governance changes include a counterexample, corrected contract, tests, and a maintainer-reviewed decision. Not applicable: no governance rule changes are proposed.

  • 已遵循贡献指南、架构说明及工程约束。

  • 按职责处理改动,未机械拆分或新增重复行为权威。

  • 以 PR 目标提交为基线的架构检查通过,未抬高预算。

  • 覆盖成功与失败,并已注明平台及功能验收边界。

  • 此收尾未新增 UI 文案;原功能的类型化翻译合同已验证。

  • 此项不适用:未提出治理规则修订。

Checkboxes explain the review; they do not replace CI or maintainer approval.

勾选项只说明审阅依据,不替代 CI 或维护者审批。

Kuddev and others added 30 commits October 6, 2026 16:43
…ev#504)

Keep valid candidates visible during same-input source refresh and preserve the POSIX editor query protocol.
Retain main media backend ownership and its CI profile; keep the authenticated mobile scrolling changes.
…dev#498)

Reuse authenticated bounded input and terminal scrolling while returning the controlled viewport. Preserve the canonical main media backend and compact wheel replies.
…Kuddev#500)

Keep Android installation identity, record merged completion and mobile scrolling fixes, and leave public release publication separate.
Kuddev#505)

Keep short choices readable and keyboard accessible without a fixed-width dropdown fallback. Show advanced effect controls on demand while preserving stored configuration.
…v#507)

Clear inherited application identity only for the isolated test child. Preserve normal Runtime routing, unrelated environment and owned process cleanup.
Restore the renamed capsule writer to its existing settings group and compare canonical launch fixture paths. Preserve production behavior, test assertions and native validation coverage.
Kuddev and others added 24 commits October 7, 2026 13:02
…#514)

Refresh saved managed PowerShell bootstrap references at process launch.
Version generated scripts by release and content while preserving shell arguments and existing scripts.
…uddev#519)

Keep terminal history scrolling local to the phone with bounded independent history pages.
Preserve reading position across paging and reflow, retain fling momentum, and cancel stale frames or input gestures.
Separate LAN route changes from relay connection state. Preserve active relay sessions while refreshing LAN addresses, and keep local pairing available when relay is offline.
Align adapter selection, port input and refresh action in one row. Keep dropdown hover local, animate the one-pixel port focus line, preserve automatic port assignment and match backup action padding.
)

Use data-oriented theme import and export diagnostics and descriptive internal names. Retain compatible format identifiers, protocol semantics, upstream attribution and recorded terminal fixtures.
…Kuddev#513)

Use per-window postprocess capability and a shared bounded terminal-effect controller. Pin the qualified native renderer and component revisions, preserve ordered WGSL inputs and resource lifetimes, and retain paint-driven single-wake cadence.
…on-20261007

feat(ui): integrate bounded search and day-night transitions
@Cec1c
Cec1c marked this pull request as ready for review October 9, 2026 03:11
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