Robot_Arm/tests/verify.py

# -*- coding: utf-8 -*-
"""重构期的不变量关卡。

    python -m tests.verify            # 数据哈希 + IK 数值（秒级，每步都跑）
    python -m tests.verify --shots    # 再加全套截图比对（约 3 分钟，前端改动后跑）

基线在 tests/baselines/ 下，随仓库走。产物按**文件名**匹配，不写死目录：
mesh 从 web/ 搬到 webapp/data/ 这类移动不该让关卡失效，也不该让我以为它通过了。
任何一项对不上就退出码 1。
"""
import hashlib
import io
import json
import os
import subprocess
import sys

sys.stdout.reconfigure(encoding="utf-8")
HERE = os.path.dirname(os.path.abspath(__file__))
REPO = os.path.dirname(HERE)
BASE = os.path.join(HERE, "baselines")

# 每个产物允许出现的目录，按优先级；找到第一个即取。
CANDIDATES = {
    "mesh_arm.json": ["webapp/data", "web"],
    "sheets.jsdata": ["webapp/data", "web"],
    "travel.json": ["docs", "webapp/data"],
    "travel_detail.json": ["docs", "webapp/data"],
    "collide.json": ["docs", "webapp/data"],
    "screws.json": ["docs"],
}


def md5(path):
    h = hashlib.md5()
    with open(path, "rb") as f:
        for chunk in iter(lambda: f.read(1 << 20), b""):
            h.update(chunk)
    return h.hexdigest()


def read_list(name):
    """基线文件是 'hash  filename'；md5sum 的二进制模式会在文件名前加 '*'，剥掉。"""
    out = {}
    for ln in io.open(os.path.join(BASE, name), encoding="utf-8"):
        ln = ln.strip()
        if ln:
            h, f = ln.split()
            out[f.lstrip("*")] = h
    return out


def locate(fname, dirs):
    for d in dirs:
        p = os.path.join(REPO, d, fname)
        if os.path.exists(p):
            return p
    return None


def check_data():
    base = read_list("data_by_name.md5")
    bad, rows = 0, []
    for fname, dirs in CANDIDATES.items():
        want = base.get(fname)
        if want is None:
            rows.append(("SKIP", fname, "基线里没有这项"))
            continue
        p = locate(fname, dirs)
        if p is None:
            bad += 1
            rows.append(("MISS", fname, "找不到文件（试过的目录: %s）" % ", ".join(dirs)))
            continue
        got = md5(p)
        ok = got == want
        if not ok:
            bad += 1
        rows.append(("OK " if ok else "DIFF", fname,
                     os.path.relpath(p, REPO).replace("\\", "/") + "  " + got[:12] +
                     ("" if ok else "  基线 " + want[:12])))
    for st, f, msg in rows:
        print("  [%s] %-20s %s" % (st, f, msg))
    return bad


def check_ik():
    r = subprocess.run(["node", os.path.join(HERE, "ik_baseline.mjs"),
                        "--compare", os.path.join(BASE, "ik_exact.json")],
                       capture_output=True, text=True, cwd=REPO,
                       encoding="utf-8", errors="replace")
    out = r.stdout or ""
    for ln in out.splitlines():
        print("   " + ln)
    if r.returncode != 0:
        print("   node 退出码 %d: %s" % (r.returncode, (r.stderr or "").strip()[:300]))
    # 没有 "== exact" 这一行就说明探针根本没跑出比对结果（编码炸了、node 不在、路径错…），
    # 绝不能因为"没报差异"就算通过。
    if "== exact" not in out:
        print("   [MISS] ik_baseline 没有输出比对结果，判 FAIL")
        return 1
    return 0 if r.returncode == 0 else 1


def check_shots():
    """两层判定，依据是"同一构建连跑两轮"的实测（scratch/_gate/flaky_scan.py）：
      104 张逐字节可复现  -> md5 必须相同；
      3 张线束 x-ray 不可复现（mean<=0.0017、无像素差>8，纯抗锯齿噪声）
                          -> 与 tests/baselines/flaky/ 里的参考像素比，超阈值才算回归。
    把不可复现的也要求 md5 相同，等于给每一步埋假 FAIL，最后只会被人肉忽略。"""
    import numpy as np
    from PIL import Image

    base = read_list("shots_stable.md5")
    th = json.load(io.open(os.path.join(BASE, "flaky", "thresholds.json"), encoding="utf-8"))
    r = subprocess.run([sys.executable, "-m", "arm.viewer.shot"],
                       capture_output=True, text=True, cwd=REPO,
                       encoding="utf-8", errors="replace")
    out = r.stdout or ""
    if "all shots ok" not in out and "OK " not in out:
        print("   [MISS] shot.py 没有正常输出（退出码 %s），判 FAIL" % r.returncode)
        return 999
    bad = [l for l in out.splitlines() if "FAIL" in l]
    if bad:
        print("   [MISS] 截图失败 %d 个: %s" % (len(bad), bad[:3]))

    fails = len(bad)
    for fname, want in sorted(base.items()):
        p = os.path.join(REPO, fname.replace("/", os.sep))
        if not os.path.exists(p):
            print("   [MISS] %s" % fname)
            fails += 1
            continue
        if md5(p) != want:
            print("   [DIFF] %-40s" % fname)
            fails += 1
    print("   逐字节比对: %d 张稳定视图，%d 张不一致" % (len(base), fails - len(bad)))

    tol_fails = 0
    for name, ref in sorted(th["views"].items()):
        p = os.path.join(REPO, "docs", "shots", name + ".png")
        q = os.path.join(BASE, "flaky", name + ".png")
        if not (os.path.exists(p) and os.path.exists(q)):
            print("   [MISS] 容差比对缺文件: %s" % name)
            tol_fails += 1
            continue
        d = np.abs(np.asarray(Image.open(p).convert("RGB"), dtype=np.int16)
                   - np.asarray(Image.open(q).convert("RGB"), dtype=np.int16))
        mean, mx = float(d.mean()), int(d.max())
        over = mean > th["mean_max"] or mx > th["pixel_max"]
        tol_fails += 1 if over else 0
        print("   [%s] %-34s mean=%.5f (<= %.2f)  最大单像素差=%d (<= %d)"
              % ("DIFF" if over else "OK  ", name, mean, th["mean_max"], mx, th["pixel_max"]))
    print("   容差比对: %d 张噪声视图，%d 张超阈值" % (len(th["views"]), tol_fails))
    return fails + tol_fails


def main():
    argv = sys.argv[1:]
    fails = 0
    print("== 数据产物哈希 ==")
    fails += check_data()
    print("== IK 数值基线 ==")
    fails += check_ik()
    if "--shots" in argv:
        print("== 截图逐字节比对 ==")
        fails += check_shots()
    print("RESULT:", "PASS" if fails == 0 else "FAIL（%d 项不一致）" % fails)
    return 1 if fails else 0


if __name__ == "__main__":
    sys.exit(main())