AskaEth
2e9702790d
r1 复审 P1-1：上一版自检对外宣称「不可能被改成总是绿」，实测 19 条变异漏 11 条
（最要命的是把 `available` 写成恒 false —— 门禁永久关闭而每次运行都绿）。根因是三条
判据只是字符串存在性检查、自测步与自检步被整条豁免、`is True` 挡不住 `'true'`。
`packaging/tests/gpu_shot_ci_selftest.py` 重写为 6 条判据：
① 五步按 `id:` 必须在场且无条件（`vectors`/`exitcodes`/`wiring`/`detect`/`judge`，
`exitcodes` 还得真断言 `expect 0/1/2`）；
② 解析 `if:` 表达式后**求值**判「判定挂门」（`available` 取 `'true'`/`'false'`/`''`
必须真/假/假，认不出一律红）—— `|| always()` 这类文本层面的门现在真红了；
③ 两种调用形态都必须在场且产物来自 `needs.probe.outputs.*`；
④ 跳过路径**实跑 7 种输入**（新增 `packaging/tests/gpu_shot_ci_probe.py`），
断言 `$GITHUB_OUTPUT` 的 `available`、summary 文案、`::warning::` 与三条红线的 exit≠0；
⑤ `continue-on-error` 按 YAML 布尔语义判（`'true'` / `${{ true }}` 也算开启）；
⑥ 扫所有 `python`+`packaging/` 的 `run`（自测步、自检步都在内），`|| true` / `; true` /
`|| exit 0` / 裸 `exit 0` 全算吞。
变异向量搬到 `packaging/tests/gpu_shot_ci_vectors.json`：21 条，按解析后的结构注入。
P2-1：`gpu-evidence.yml` 加 `workflow_call:`，`dev-guards.yml` 里 `uses:` 同一份 ——
直推 dev 时 selftest 那一半现在会跑（evidence 在 dev 上照样显式跳过）。
P2-2：跳过摘要与 `gpu-evidence.md §8` 改成唯一可行的补救路径（把产物提交进被 dispatch
的 ref，`git add -f`），并写明「就地拷文件在 CI 上做不到」。
措辞：三处「不可能被改成总是绿」改成能复跑的覆盖面 + 边界。
146 lines
7.8 KiB
Python
146 lines
7.8 KiB
Python
#!/usr/bin/env python3
|
||
"""`gpu-evidence.yml` 的**功能断言**部分（M253 / r1 复审返工）：把 workflow 里的 `run:` 原文抽出来真跑。
|
||
|
||
为什么单独一份：判据脚本自己守 ≤400 行硬线，而这里全是「把 `run:` 落盘、用桩 env 起 bash、读
|
||
`$GITHUB_OUTPUT` / `$GITHUB_STEP_SUMMARY`」的**执行脚手架**，与「哪条判据该红」的判断无关
|
||
（同 `gpu_shot_selftest.py` 与 `check_gpu_shot.py` 的分工）。
|
||
|
||
被谁用：`packaging/tests/gpu_shot_ci_selftest.py` 的判据 4。它一次跑 7 种输入，因为字符串存在性
|
||
检查被 r1 复审证伪过：旧版只要求「同一步的 run 里同时出现 4 个字符串」，于是把 `available` 写成
|
||
恒 `false`（门禁永久关闭、每次运行都绿）它照样绿 —— **文案都在，执行路径却根本没走到**。
|
||
|
||
7 种输入（前 4 条断言「跳过要留痕」，后 3 条断言「红线不许退化成跳过」）：
|
||
|
||
| # | 输入 | 必须发生什么 |
|
||
|---|------|-------------|
|
||
| S1 | 无产物（CI 常态） | `available=false` + summary 写「跳过 / 未验证」+ stdout 有 `::warning::` |
|
||
| S2 | 约定产物成对 | `available=true` |
|
||
| S3 | 约定产物半套（只有日志） | `available=false`（半套不算可用） |
|
||
| S4 | dispatch 成对 | `available=true` |
|
||
| S5 | dispatch 点名不存在 | `exit≠0`，且**不写** `available`（写了就等于退化成跳过） |
|
||
| S6 | dispatch 正例缺 PNG | `exit≠0`，同上 |
|
||
| S7 | dispatch 负例只给日志 | `available=true`（`--expect-refuse` 允许省略 PNG） |
|
||
|
||
不降级：找不到能跑的 `bash` 就**判红**（「认不出/跑不了」≠「没问题」）。
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
import os
|
||
import shutil
|
||
import subprocess
|
||
import tempfile
|
||
from pathlib import Path
|
||
|
||
_CACHE: dict[str, list[str]] = {}
|
||
|
||
|
||
def _find_bash() -> str | None:
|
||
"""挑一个**真能跑**的 bash。Windows 上 `bash.exe` 可能是 WSL 的 relay（没有 `/bin/bash`），要验一次。"""
|
||
candidates = [os.environ.get("BASH"), shutil.which("bash")]
|
||
if os.name == "nt":
|
||
candidates += [r"C:\Program Files\Git\bin\bash.exe", r"C:\Program Files\Git\usr\bin\bash.exe"]
|
||
candidates.append("bash")
|
||
for candidate in candidates:
|
||
if not candidate:
|
||
continue
|
||
try:
|
||
done = subprocess.run([candidate, "-c", "echo ok"], capture_output=True, text=True, timeout=60)
|
||
except (OSError, subprocess.SubprocessError):
|
||
continue
|
||
if done.returncode == 0 and "ok" in (done.stdout or ""):
|
||
return candidate
|
||
return None
|
||
|
||
|
||
def _run(bash: str, script: Path, workdir: Path, env_extra: dict):
|
||
"""跑一份 `run:` 原文，返回 `(CompletedProcess, GITHUB_OUTPUT 字典, GITHUB_STEP_SUMMARY 文本)`。"""
|
||
out_path, sum_path = workdir / "gh_output", workdir / "gh_summary"
|
||
env = os.environ.copy()
|
||
env.update({"GITHUB_OUTPUT": out_path.as_posix(), "GITHUB_STEP_SUMMARY": sum_path.as_posix()})
|
||
env.update({key: ("" if value is None else str(value)) for key, value in env_extra.items()})
|
||
done = subprocess.run([bash, "-e", str(script)], cwd=str(workdir), env=env, capture_output=True, text=True, encoding="utf-8", errors="replace")
|
||
outputs: dict[str, str] = {}
|
||
if out_path.exists():
|
||
for line in out_path.read_text(encoding="utf-8", errors="replace").splitlines():
|
||
key, sep, value = line.partition("=")
|
||
if sep:
|
||
outputs[key.strip()] = value
|
||
summary = sum_path.read_text(encoding="utf-8", errors="replace") if sum_path.exists() else ""
|
||
return done, outputs, summary
|
||
|
||
|
||
def _case(label: str, done, outputs: dict, summary: str, *, available: str, contains: tuple[str, ...] = (), warning: bool = False) -> list[str]:
|
||
problems: list[str] = []
|
||
if done.returncode != 0:
|
||
problems.append(f"{label}：退出码 {done.returncode}，应为 0（stderr: {(done.stderr or '').strip()[:200]}）")
|
||
if outputs.get("available") != available:
|
||
problems.append(f"{label}：`available={outputs.get('available')!r}`，应为 {available!r} —— 探测结论被改坏了，门就永久开或永久关")
|
||
for needle in contains:
|
||
if needle not in summary:
|
||
problems.append(f"{label}：job summary 里缺 `{needle}` —— 跳过不再留痕")
|
||
if warning and "::warning::" not in (done.stdout or ""):
|
||
problems.append(f"{label}：stdout 里没有 `::warning::` —— 跳过变成了静默 pass")
|
||
return problems
|
||
|
||
|
||
def _probe_scenarios(bash: str, script_text: str) -> list[str]:
|
||
problems: list[str] = []
|
||
with tempfile.TemporaryDirectory(prefix="ull-gpu-ci-", ignore_cleanup_errors=True) as tmp:
|
||
root = Path(tmp)
|
||
fixtures = root / "fixtures"
|
||
fixtures.mkdir()
|
||
log = fixtures / "a1-gpu.stderr.log"
|
||
log.write_text("[shot] 200x120\n", encoding="utf-8")
|
||
png = fixtures / "a1-gpu.png"
|
||
png.write_bytes(b"\x89PNG\r\n\x1a\n")
|
||
|
||
def case(slug: str, tree: tuple[tuple[str, bytes], ...] = ()):
|
||
workdir = root / slug
|
||
workdir.mkdir()
|
||
for rel, payload in tree:
|
||
path = workdir / rel
|
||
path.parent.mkdir(parents=True, exist_ok=True)
|
||
path.write_bytes(payload)
|
||
script = workdir / "probe.sh"
|
||
script.write_text(script_text, encoding="utf-8")
|
||
return script, workdir
|
||
|
||
base = {"EVENT": "push", "SHOT_LOG": "", "SHOT_PNG": "", "EXPECT_REFUSE": None}
|
||
convention = "DevDocs/artifacts/gpu-probe/gpu-shot"
|
||
script, workdir = case("s1-none")
|
||
problems += _case("S1 无产物（CI 常态）", *_run(bash, script, workdir, base), available="false", contains=("跳过", "未验证"), warning=True)
|
||
script, workdir = case("s2-convention-pair", ((f"{convention}.stderr.log", b"x"), (f"{convention}.png", b"x")))
|
||
problems += _case("S2 约定产物成对", *_run(bash, script, workdir, base), available="true")
|
||
script, workdir = case("s3-convention-half", ((f"{convention}.stderr.log", b"x"),))
|
||
problems += _case("S3 约定产物半套（只有日志）", *_run(bash, script, workdir, base), available="false", contains=("跳过",))
|
||
script, workdir = case("s4-dispatch-pair")
|
||
problems += _case("S4 dispatch 成对", *_run(bash, script, workdir, dict(base, SHOT_LOG=str(log), SHOT_PNG=str(png))), available="true")
|
||
for slug, label, extra in (
|
||
("s5-dispatch-missing", "S5 dispatch 点名不存在", {"SHOT_LOG": str(root / "nope.log"), "SHOT_PNG": str(png)}),
|
||
("s6-dispatch-no-png", "S6 dispatch 正例缺 PNG", {"SHOT_LOG": str(log), "SHOT_PNG": ""}),
|
||
):
|
||
script, workdir = case(slug)
|
||
done, outputs, _summary = _run(bash, script, workdir, dict(base, **extra))
|
||
if done.returncode == 0:
|
||
problems.append(f"{label}：退出码 0 —— 「明确说要判、东西不在」退化成了跳过（假绿）")
|
||
if outputs:
|
||
problems.append(f"{label}：该红却写了 `$GITHUB_OUTPUT`（{outputs}）")
|
||
script, workdir = case("s7-dispatch-refuse")
|
||
problems += _case("S7 dispatch 负例只给日志", *_run(bash, script, workdir, dict(base, SHOT_LOG=str(log), SHOT_PNG="", EXPECT_REFUSE="true")), available="true")
|
||
return problems
|
||
|
||
|
||
def probe_problems(script_text: str) -> list[str]:
|
||
"""判据 4：`id: detect` 那一步的 `run` 原文实跑 7 种输入（带缓存，同文本只跑一次）。"""
|
||
if not script_text.strip():
|
||
return ["`id: detect` 那一步没有 `run:` —— 跳过路径无从执行"]
|
||
if script_text not in _CACHE:
|
||
bash = _find_bash()
|
||
_CACHE[script_text] = (
|
||
["找不到能跑的 `bash`：跳过路径的功能断言无法执行，本判据不降级（认不出就红）"]
|
||
if not bash
|
||
else _probe_scenarios(bash, script_text)
|
||
)
|
||
return _CACHE[script_text]
|