#!/usr/bin/env python3
"""
render_mermaid.py
=================
扫描合并后的 markdown 中的 ```mermaid 代码块,用 mmdc(@mermaid-js/mermaid-cli)
渲染为 PNG,并把这些块替换为可被 pandoc 直接吃下的 LaTeX includegraphics 命令。

用法(被 build_book.sh 调用):
    python3 render_mermaid.py <input_md> <out_md> <fig_dir>

特性:
- 内容 hash 缓存,二次构建不重复渲染
- 单个块渲染失败不会中断整体构建,会降级为 verbatim 源码 + 警告
- 详细错误信息打到 stderr 方便调试

依赖:
    mmdc 必须在 plans/PRISM/tools/node_modules/.bin/mmdc
"""
from __future__ import annotations
import hashlib
import pathlib
import re
import subprocess
import sys


TOOLS_DIR = pathlib.Path(__file__).resolve().parent
MMDC = TOOLS_DIR / "node_modules" / ".bin" / "mmdc"


def render_one(code: str, fig_dir: pathlib.Path) -> pathlib.Path:
    """渲染一个 mermaid 代码块,基于内容 hash 缓存。"""
    digest = hashlib.sha1(code.encode("utf-8")).hexdigest()[:12]
    out_png = fig_dir / f"mmd_{digest}.png"
    if out_png.exists():
        return out_png

    src_mmd = fig_dir / f"mmd_{digest}.mmd"
    src_mmd.write_text(code, encoding="utf-8")

    cmd = [
        str(MMDC),
        "-i", str(src_mmd),
        "-o", str(out_png),
        "-b", "white",
        "-s", "2",
        "--width", "1400",
    ]
    proc = subprocess.run(cmd, capture_output=True, text=True)
    if proc.returncode != 0:
        # 抽取 mermaid 的 Parse / Lexical error 行
        err_lines = [ln for ln in (proc.stdout + proc.stderr).splitlines()
                     if "Error" in ln or "error" in ln]
        err_head = "\n  ".join(err_lines[:5]) or "(no error line found)"
        sys.stderr.write(
            f"\n[mmdc ERROR] block {digest}:\n"
            f"  {err_head}\n"
            f"  --- mermaid source ---\n{code}\n"
            f"  ----------------------\n"
        )
        raise RuntimeError(f"mmdc failed on block {digest}")
    return out_png


MERMAID_RE = re.compile(
    r"```mermaid\s*\n(.*?)\n```",
    re.DOTALL,
)


def process(input_md: pathlib.Path, out_md: pathlib.Path, fig_dir: pathlib.Path) -> int:
    if not MMDC.exists():
        sys.stderr.write(
            f"[ERROR] mmdc not found at {MMDC}\n"
            f"        run: cd plans/PRISM/tools && npm install @mermaid-js/mermaid-cli\n"
        )
        sys.exit(2)
    fig_dir.mkdir(parents=True, exist_ok=True)
    text = input_md.read_text(encoding="utf-8")

    ok = 0
    fail = 0
    failures: list[str] = []

    def _sub(m: re.Match) -> str:
        nonlocal ok, fail
        code = m.group(1)
        try:
            png = render_one(code, fig_dir)
        except RuntimeError:
            fail += 1
            digest = hashlib.sha1(code.encode("utf-8")).hexdigest()[:12]
            failures.append(digest)
            # 降级:渲染失败 → 在 PDF 中插入"渲染失败"标记 + 源码 verbatim,
            # 这样书构建不会因为一个块挂掉而中断。
            return (
                f"\n\\begin{{quote}}\\textbf{{[mermaid 渲染失败 {digest}]}}"
                f"\\end{{quote}}\n\n"
                f"\\begin{{verbatim}}\n{code}\n\\end{{verbatim}}\n"
            )
        ok += 1
        rel = png.relative_to(input_md.parent)
        # 同时限制宽 (0.92 linewidth) 与高 (0.82 textheight),
        # keepaspectratio 自动取较紧者,保证任何极端长宽比的图都不溢出 A4 页面。
        return (
            f"\n\\begin{{center}}\n"
            f"\\includegraphics[width=0.92\\linewidth,"
            f"height=0.82\\textheight,keepaspectratio]{{{rel.as_posix()}}}\n"
            f"\\end{{center}}\n"
        )

    new_text = MERMAID_RE.sub(_sub, text)
    out_md.write_text(new_text, encoding="utf-8")
    if fail:
        print(f"[mermaid] rendered {ok} ok, {fail} FAILED → 降级为 verbatim")
        print(f"[mermaid] failed digests: {', '.join(failures)}")
    else:
        print(f"[mermaid] rendered {ok} block(s) → {fig_dir}")
    return ok


if __name__ == "__main__":
    if len(sys.argv) != 4:
        sys.stderr.write(
            "usage: render_mermaid.py <input_md> <out_md> <fig_dir>\n"
        )
        sys.exit(1)
    process(
        pathlib.Path(sys.argv[1]),
        pathlib.Path(sys.argv[2]),
        pathlib.Path(sys.argv[3]),
    )