feat(migrate): Halo 全量内容迁移（压缩版）——6文/21图/1瞬间 URL 1:1

- 正文源=线上渲染页（Halo 快照层 patch 残迹：泥潭/赎罪仅 2 字符）， pandoc html→gfm + tags API 显示名换算后分类：5 书评（五档情绪色）+1 技术 - 图库 21 张：长边 2400/q86 压缩入仓（原图 196M→9.5M——公开仓历史膨胀与 nginx 100M 推送上限双杀；原图在 Phase 0 备份 tar 中永久留档） - 瞬间 1 条；URL 1:1（6 slug 全等，零 301） - 本提交由 7 个本地提交重写而成（清除已进历史的 196M blob，未推送过）；因 reset --soft 暂存竞态，一并包含 spec 归档文件（.spec/specs/archive/2026-10-24_migrate-blog-content/）验证: check_urls.py 6==6；astro build 11 页 exit 0（8.67s）； EXIF 管线跑通；staging 更新；halo healthy+200 全程未动 Co-authored-by: GLM-5.3
2026-10-25 00:39:08 +08:00
1 parent 7b71acf223
commit aea21c8694
@@ -0,0 +1,128 @@
+#!/usr/bin/env python3
+"""文章迁移（Phase 2 T2）：Halo 渲染页 → content/posts/*.md
+
+正文源用线上渲染页（reader 看到的即所得）——实测 DB snapshots 对
+shu-zui/ni-tan 只有 2 字符（patch 残迹），而渲染页完整。
+
+处理链：抓 /archives/{slug} → 剥主题噪音(button/nav/header) → 图片
+/upload/* 下载到 content/posts/assets/{slug}/ 并改写引用 → pandoc
+html→gfm → frontmatter（分类/标签沿用 Halo，书评分配五档情绪色）。
+"""
+import html as htmllib
+import json
+import re
+import subprocess
+import urllib.request
+from pathlib import Path
+
+BASE = "http://127.0.0.1:18090"
+ROOT = Path(".")
+ASSETS = ROOT / "content" / "posts" / "assets"
+
+# 书评五档情绪色（tokens --mood-1..5）；技术文不分配
+MOODS = ["#b45309", "#0f766e", "#7c3aed", "#9f1239", "#166534"]
+
+# Halo 分类名 → 集合枚举
+CAT_MAP = {"Reading": "reading", "Linux": "tech", "默认分类": "tech"}
+
+
+def fetch(url: str) -> bytes:
+    with urllib.request.urlopen(url, timeout=30) as r:
+        return r.read()
+
+
+def article_html(slug: str) -> str:
+    page = fetch(f"{BASE}/archives/{slug}").decode("utf-8")
+    m = re.search(r"<article[^>]*>(.*?)</article>", page, re.S)
+    if not m:
+        raise RuntimeError(f"no article for {slug}")
+    body = m.group(1)
+    # 主题噪音：higan 主题把主题切换钮等塞进 article；标题在 frontmatter 重建，正文 h1 去重
+    body = re.sub(r"<button.*?</button>", "", body, flags=re.S)
+    body = re.sub(r"<header.*?<h1[^>]*>.*?</h1>.*?</header>", "", body, count=1, flags=re.S)
+    return body
+
+
+def localize_images(slug: str, body: str) -> str:
+    def repl(m: re.Match) -> str:
+        src = m.group(1)
+        if not src.startswith("/upload/"):
+            return m.group(0)
+        dest_dir = ASSETS / slug
+        dest_dir.mkdir(parents=True, exist_ok=True)
+        fname = src.rsplit("/", 1)[1]
+        data = fetch(BASE + src)
+        (dest_dir / fname).write_bytes(data)
+        return m.group(0).replace(src, f"./assets/{slug}/{fname}")
+
+    return re.sub(r'src="([^"]+)"', repl, body)
+
+
+def to_markdown(body: str) -> str:
+    p = subprocess.run(
+        ["/home/wait4x/.local/bin/pandoc", "-f", "html", "-t", "gfm-raw_html", "--wrap=none"],
+        input=body.encode(), capture_output=True, check=True,
+    )
+    return p.stdout.decode("utf-8")
+
+
+def load_tag_display() -> dict[str, str]:
+    """内部标签名(tag-xxx) → 显示名(read/linux/...)。"""
+    with urllib.request.urlopen(BASE + "/apis/api.content.halo.run/v1alpha1/tags") as r:
+        data = json.loads(r.read())
+    return {t["metadata"]["name"]: t["spec"]["displayName"] for t in data.get("items", [])}
+
+
+def main() -> int:
+    posts = json.load(open(".migration/posts.json", encoding="utf-8"))
+    tag_disp = load_tag_display()
+    mood_i = 0
+    report = []
+    for p in posts:
+        spec = p["spec"]
+        slug = spec["slug"]
+        tags = [tag_disp.get(t, t) if isinstance(t, str) else t.get("name", "") for t in (spec.get("tags") or [])]
+        cat_raw = spec.get("category") or ""
+        # 分类判定：Halo 分类名优先，回退看标签（read→reading / linux→tech）
+        if CAT_MAP.get(cat_raw):
+            category = CAT_MAP[cat_raw]
+        elif any("read" in t.lower() for t in tags):
+            category = "reading"
+        else:
+            category = "tech"
+
+        mood = None
+        if category == "reading":
+            mood = MOODS[mood_i % len(MOODS)]
+            mood_i += 1
+
+        body = localize_images(slug, article_html(slug))
+        md = to_markdown(body).strip()
+        md = re.sub(r"^(-{4,}|_{4,})\s*", "", md)  # 原 hr 开头去掉（frontmatter 后首元素应是正文）
+        fm = [
+            "---",
+            f'title: "{htmllib.escape(spec["title"], quote=False)}"',
+            f"pubDate: {spec['publishTime'][:10]}",
+            f"category: {category}",
+            f"tags: [{', '.join(tags)}]",
+        ]
+        if mood:
+            fm.append(f"mood: '{mood}'")
+        fm.append("---")
+        out = ROOT / "content" / "posts" / f"{slug}.md"
+        out.write_text("\n".join(fm) + "\n\n" + md + "\n", encoding="utf-8")
+        n_img = len(list((ASSETS / slug).glob("*"))) if (ASSETS / slug).exists() else 0
+        report.append((slug, category, len(md), n_img))
+        print(f"  {slug} [{category}] {len(md)} md-chars, {n_img} imgs")
+
+    # 清理被真实内容取代的样例文件（真实技术文 slug=csge-lei-pei-zhi，样例为 cs-ge-lei-pei-zhi）
+    sample = ROOT / "content" / "posts" / "cs-ge-lei-pei-zhi.md"
+    if sample.exists():
+        sample.unlink()
+        print("  removed sample: cs-ge-lei-pei-zhi.md")
+    print(f"migrated {len(report)} posts")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())