test(windows): launch the source CLI through the shipped host entry

[JasonBuildAI]

Goal And Delivered Outcome

• Outcome basis / optional anchor: Closes [Bug]: On native Windows the source-checkout test suite cannot launch the CLI (WinError 193 / WSL relay / locale decode) #6405 (reproduced: on native Windows the source-checkout suite cannot launch the CLI).

• Goal/source and gap: the 14 affected files exec the extensionless POSIX bash shim scripts/loopx via subprocess.run, so on native Windows 45 tests fail (OSError: [WinError 193], the WSL relay's missing /bin/bash, or a cp936 read of the CLI's UTF-8-pinned output). CLI-behavior coverage in these files is lost for Windows contributors and CI.

• Observable before → after, with the validation row that proves it: same 14-file command on Windows at 3300dfdd2 → 45 failed, 372 passed in 229.91s; on this head → 399 passed, 18 skipped in 366.76s (the 18 skips are the declared POSIX-only cases; see Validation row regression_parity / integration).

• Issue/task and intended base: Closes [Bug]: On native Windows the source-checkout test suite cannot launch the CLI (WinError 193 / WSL relay / locale decode) #6405; base main.

Author Declaration

• Written by: an OpenAI Codex agent executing a human operator's task; the operator set the constraints and reviewed the change before submission. AI disclosure: this change was drafted with AI assistance at my request, under my direction and review.

Implemented against

• Specification and revision: linked issue [Bug]: On native Windows the source-checkout test suite cannot launch the CLI (WinError 193 / WSL relay / locale decode) #6405; shipped Windows entry contract (loopx/windows_install.py builds [python, "-I", scripts/loopx_entry.py, ...] and scripts/loopx.ps1 delegates to it); POSIX-only guard precedent test(contract): keep the mode-000 read checks POSIX-only #6007; CLI UTF-8 stdio pinning fix(cli): pin the shipped CLI's own stdio to UTF-8 #6119. Revision: 3300dfdd2.
• Criteria:

Criterion (spec clause)	Disposition	Symbol / path	Test or command
Windows hosts exercise the CLI through the shipped Windows entry, not the POSIX shim	implemented	tests/source_cli.py (source_cli_command, source_cli_env, run_source_cli)	14-file set on Windows: 399 passed, 18 skipped
POSIX hosts keep the previous bash-shim invocation (argv, cwd, env)	implemented	tests/source_cli.py	POSIX branch keeps the prior form; to be confirmed by the ubuntu CI lane (no POSIX host in this environment)
POSIX-only launcher assertions declared POSIX-only instead of failing	implemented	tests/test_loopx_launcher_python_selection.py, tests/test_scheduler_native_launcher.py	20 prior Windows failures now skip (18) or pass (2)
CLI text reads pin UTF-8, matching the CLI's pinned stdio	implemented	tests/test_scheduler_native_launcher.py	previous UnicodeDecodeError: 'gbk' codec ... byte 0xb7 cases now pass

• Self-check before submission: read the shipped launcher chain (loopx/windows_install.py, scripts/loopx_entry.py, scripts/loopx.ps1) so the helper matches installed behavior; verified every migrated call keeps its original check semantics (30 sites compared old vs new); left out product-code changes, new dependencies, and unrelated Windows failures. Review-driven follow-up commit fee76fc77 pins LOOPX_RELEASE_ROOT like the shipped launchers instead of inheriting an ambient value, pins UTF-8 on the two remaining launcher-output reads, and adds the end-to-end --version smoke. Not verified locally: POSIX execution (no POSIX host here).

Scope And Continuation

• Completed scope and remaining work: the 14 files now run on Windows through the host entry and fail closed only where declared POSIX-only. Other Windows-host lanes (TS control-plane [Bug]: native Windows: remaining 32 host-dependent control-plane TypeScript test failures after #6207 #6351, edgebench fix(benchmark): run the EdgeBench report adapter and suite on native Windows #6368) are separate and untouched.
• Slice boundary / successor: complete within this scope; no successor required for these files.

Validation

• Tested revision: 320762cad6908b7b5672a0640de1e815ed70c3e1 (rebased onto latest main at 14112f36c; the 14-file Windows run was repeated on this head)
• Run state: finished
• Input classes: synthetic

Check kind	Result	Public-safe evidence / limitation
static	passed	ruff check over all changed files: All checks passed!; git diff --check clean
unit	passed	uv run --extra test python -m pytest -q -p no:cacheprovider tests/test_source_cli.py → 4 passed in 3.90s, including an end-to-end smoke that starts the real host entry (--version)
integration	passed	Windows, 14 affected files: 399 passed, 18 skipped in 366.76s; tests exec the real CLI entry end-to-end
regression_parity	passed	Same 14-file command at 3300dfdd2: 45 failed, 372 passed in 229.91s; before-failure list recorded in #6405; total collected 417 both sides
manual	passed	Byte-level check of the write-guard stdout (31,661 bytes): first non-ASCII byte at offset 732, decodes as UTF-8, not cp936 — the reader-side failure class
manual	not_run	POSIX execution unavailable in this environment (Windows host; bash is a broken WSL relay). CI's ubuntu lane must confirm POSIX parity

• Coverage and gaps: changed paths are test launchers only; every affected file was executed on Windows before and after. Gap: no local POSIX re-run, argued rather than measured; the POSIX branch is byte-for-byte the previous invocation and the skips/encoding are nt-scoped. Scope note: the required windows-powershell CI lane runs a curated file list that does not include these files (.github/workflows/python-tests.yml), so the demonstrated benefit is the Windows developer loop and any job that runs the suite there; durable CI regression protection for the Windows branch would need a lane change and is left out of this PR.

Frontend / Visual Evidence

• UI impact: none
• Before: N/A
• After: N/A
• States and viewports shown: N/A
• Source data: none
• Attention review: N/A (test-only change)
• Craft review: N/A (test-only change)

Type of Change

• Test update

LoopX Area

• Build, packaging, installer, or CI

Technical Direction

• Direction / acceptance reference, when applicable: N/A

Shared-authority RFC fixture impact

• Production-scale fixture schema: N/A
• Semantic dimensions changed, or reviewed no-impact rationale: N/A
• Provider conformance arms run: N/A
• Read-only legacy/file/PostgreSQL three-arm rehearsal (required for promotion, runtime-routing, or compatibility-projection changes): N/A

Boundary Checklist

• Neither the diff nor this PR body/comments/attachments disclose private state, credentials, raw traces or verifier output, internal links, or local machine paths (including .loopx/, .codex/goals/, and live ACTIVE_GOAL_STATE.md).
• I did not duplicate maintainer-owned benchmark work unless a maintainer split out a public issue for it.
• I kept the change scoped to the linked issue/task.
• I completed the visual evidence section for UI changes, or marked UI impact none.
• Every commit includes a DCO Signed-off-by trailer (git commit -s).

[loopx-agent]

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

Exact head: 320762c. Whole-PR review against 14112f3.

动机

在原生 Windows 上跑源码测试的贡献者，需要让测试真正启动该宿主可执行的 CLI。旧测试直接启动 POSIX bash shim；Windows 无法执行它，WSL relay 也不能代替已安装的 bash，按宿主默认编码读取 UTF-8 输出还会报解码错误。结果是 CLI 行为未得到检验，贡献者却要处理无关的启动失败。

此次将 30 处既有 CLI 调用集中到宿主入口 helper，保留原命令参数、cwd、env 和成功/失败断言，并声明真正的 POSIX 专用场景。独立 POSIX base/head 测试与实际 Python 隔离入口均通过，源码测试的可执行入口和既有检查语义得到保留。本次未在原生 Windows 或 PowerShell 宿主重跑，不把作者的 Windows 结果、分支模拟或本机 Python 入口验证当成该平台的独立验收；不启动 benchmark、外部模型或活动 Goal。

改动思路

复用产品已有的 Windows 启动链：sys.executable → -I → scripts/loopx_entry.py，明确设置当前 checkout 的 LOOPX_RELEASE_ROOT。POSIX 继续走原 bash shim；helper 提供原 cwd/text/capture_output 默认值及 UTF-8 解码，调用者仍保留 check 的决定。测试 fixture 只适配宿主启动与读取，不成为产品 CLI、评分或权限的新 owner。

这是一份有实际 30 处消费者的共享测试 helper，替代重复 subprocess 包装；现有 tests/control_plane/test_source_cli_entrypoint.py 检验的是 module/console 分支和 typed ownership，不是宿主 launcher 的选择，两个边界不同。相邻 windows_install/_entry_command 和 scripts/loopx.ps1 已核对，helper 的 Windows argv 与原产品 owner 一致。该 PR 边界是已有源码 CLI 测试的宿主适配，不修改已发布产品、控制面或 benchmark 的运行规则。

宿主或输入	入口与结果
POSIX	原 scripts/loopx，原参数与结果保持
Windows 分支	当前解释器 -I 运行既有 loopx_entry.py，当前 root 覆盖过期环境 root
CLI 非零退出	check=False 返回退出码；check=True 仍抛 CalledProcessError
POSIX 专用 launcher 场景	仅 nt 宿主跳过；其余平台继续检验

表格说明当前 head 的代码分支；Windows 分支的实际 Python 入口本次在 POSIX 执行，不能据此声称 Windows 内核或 PowerShell 已验证。

具体改动

全量 16 文件 +197/-219。tests/source_cli.py 的 source_cli_command（L28）选择 launcher；source_cli_env（L41）复制输入环境，并在 Windows 固定当前 root；run_source_cli（L51）保持 cwd、显式 check 和 subprocess 返回语义，默认读取 UTF-8。11 个 benchmark capability 测试文件与 Todo cadence 文件迁移 30 处调用，完整函数 AST 对照证明只删除被 helper 提供的 launcher/cwd/text/capture 包装，原 check=True/False、参数、断言和其他函数逻辑未改。

tests/test_loopx_launcher_python_selection.py 增加 nt-only 的 POSIX 声明；test_scheduler_native_launcher.py 只在 Windows 跳过 unix 分支，并固定三个原有输出读取位置的 UTF-8。tests/test_source_cli.py 用四个聚焦用例覆盖 Windows/POSIX 选择、过期 root、非 ASCII 输出和真实 host entry 启动。生产目录、依赖、CI 选择和原 benchmark 用例语义均未修改。

独立依据是 #6405 的 Expected behavior，按本次读取的当前 issue body 判断：宿主使用可执行的原有入口、POSIX 保留旧行为、POSIX 专用断言有明确范围、CLI 输出按 UTF-8 读取。Windows 入口同时对照上述 immutable base 的 shipped launcher，而非只对照新 helper 自己的返回值。

独立实际探针在非仓库 cwd 放置会报错的同名模块，并传入过期 root/PYTHONPATH：Windows 分支生成的 argv 与 windows_install owner 相同，实际 Python -I 子进程读回当前源码版本 1.4.0，未被环境或 cwd 模块截获。POSIX 原调用和新 helper 的完整 stdout/stderr/退出码相同；实际非法 CLI 参数仍返回 2 或按 check=True 抛异常。

对主干的风险

相同的原 14 文件在本机 immutable base 是 404 passed、13 skipped；当前加上四个 helper 用例是 408 passed、13 skipped。13 项均依赖本机没有的 pwsh，未作为通过项；与作者原生 Windows 的 399/18 是不同平台和选择，不能混算。风险 canary 的 4 项直接检查及 10 项选中检查通过，完整语义与维护性检查通过；没有获取或等待远端 CI。

风险集中于测试准入和 source root，而非新产品行为。用整函数 AST 与实际子进程检验防止把原失败断言改成成功；没有把 benchmark 本身跳过。Windows 环境 pin 是刻意与产品启动链对齐，POSIX 环境不增加该 pin。helper 只在 tests 内被消费，不安装新能力、写入真实配置或改变执行授权。

非阻塞建议： 后续可将 tests/test_source_cli.py 纳入现有 windows-powershell 的 curated 文件列表，使四项测试及真实 host-entry smoke 持续在 Windows CI 执行。目前该列表没有这些文件，作者也已明确披露；本 PR 已交付有用的本地贡献者测试修复，但不能声称新增 Windows CI 防护已生效。该建议不要求在此 PR 扩大到其他 Windows、TS 或 EdgeBench 任务。

我的整体评价

APPROVE。共享 helper 有 30 处真实消费者，减少重复维护并修复已演示的宿主入口问题；既有正负检查、POSIX 行为和输入环境隔离均有独立证据。原生 Windows/PowerShell 的本次独立运行缺口保留，当前无要求作者更改产品 runtime 的 blocker。

未来维护检查已应用：调用者集中到同一个测试入口，复用既有产品 owner，不再增加另一套 installer 或通用测试框架。同作者最近 30 个 PR 已检查；相邻 #6356、#6368、#6399 和 #6422 处理其他明确宿主或产品边界，没有发现重复同形 scaffold，未触发贡献限制。历史 RFC 经验对本测试修复不提供结论，只提示核验实际使用路径；不宣称模型采纳或长期收益。

English verdict: APPROVE — 320762c. All 30 migrated calls preserve their checks and assertions. POSIX base: 404 passed/13 pwsh skips; head: 408 passed/13 skips. The real isolated Python entry and ambient-root rejection are verified on POSIX; native Windows and PowerShell were not independently rerun. Consider adding the helper tests to the existing Windows CI selection.