#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
全文抽取　第 1 步：建卡片
=========================
把 870 件釋憲案的官網原文全部區塊（含相關文件／聲請書）做成標註卡片，
並依字數做平衡分批。

    python _pipeline\\全文抽取_建卡片.py

輸出於 60_全文抽取/：
    卡片/<釋憲案>.txt   標註時實際閱讀的全文（不含任何既有碼）
    工作檔.json         每件的 meta 與字數
    批次.json           平衡分批結果
    進度.csv            逐件進度追蹤
"""

import csv, json, sys, collections
from pathlib import Path

ROOT = Path(__file__).resolve().parent.parent
PIPE = ROOT / "_pipeline"
OUT = ROOT / "60_全文抽取"
CARD = OUT / "卡片"

# 全部區塊都要，順序固定，讓卡片在各件間長得一樣
ORDER = ["解釋文", "主文", "理由書", "理由", "相關文件",
         "判決摘要", "主筆大法官記載", "大法官就主文所採立場表"]

# 憲判字的署名列不在官網頁面的區塊結構內（落在最末一段編號之後），
# 建庫時被解析器丟掉。以 60_全文抽取/補收/署名列_彙整.json 補回。
# 釋字的署名列在「理由書」末段，不受影響。
ROSTER = OUT / "補收" / "署名列_彙整.json"

# 意見書全文合計 1,941 萬字，是本文的 4.5 倍，不放進卡片；
# 但標題本身載明「某大法官提出、某大法官加入」，是意見結盟的直接證據，
# 且僅約 1,680 行，故以清單形式收進卡片。
TARGET = 55_000        # 每批目標字數


def main():
    try:
        sys.stdout.reconfigure(encoding="utf-8")
    except Exception:
        pass

    dbp = PIPE / "資料庫.json"
    if not dbp.exists():
        print("找不到", dbp)
        sys.exit(1)
    recs = json.load(open(dbp, encoding="utf-8"))
    roster = json.load(open(ROSTER, encoding="utf-8")) if ROSTER.exists() else {}

    OUT.mkdir(exist_ok=True)
    CARD.mkdir(exist_ok=True)

    work = []
    for r in recs:
        secs = {lab: ps for lab, ps in r["s"]}
        parts = []
        for lab in ORDER:
            if lab not in secs:
                continue
            body = "\n".join((f"（{n}）" if n else "") + t for n, t in secs[lab])
            parts.append(f"【{lab}】\n{body}")
        # 補收之署名列（僅憲判字需要）
        rr = roster.get(r["k"])
        if rr and rr.get("署名列原文"):
            parts.append("【署名列（自官網補收）】\n"
                         + rr["署名列原文"]
                         + "\n※本區塊係建庫後自官網原文補收，"
                           "經 markdown 轉換，空白排版與官網 HTML 未必逐字相同；"
                           "姓名與順序為逐字照錄。"
                         + (f"\n※迴避記載：{rr['迴避原文']}" if rr.get("迴避原文") else ""))

        # 意見書清單（僅標題，全文不入卡片）
        ops = r.get("o") or []
        if ops:
            lines = "\n".join(f"（{i}）{x.get('t','')}" for i, x in enumerate(ops, 1))
            parts.append(f"【意見書清單】（{len(ops)} 篇，僅列標題，全文不入卡片）\n" + lines)

        extra = [lab for lab in secs if lab not in ORDER]
        for lab in extra:
            body = "\n".join((f"（{n}）" if n else "") + t for n, t in secs[lab])
            parts.append(f"【{lab}】\n{body}")
        text = "\n\n".join(parts)
        head = (f"釋憲案　{r['k']}　{r.get('z','')}"
                + (f"【{r.get('n')}】" if r.get("n") else "")
                + f"\n類型　{r.get('ty','')}　公布　{r.get('roc') or r.get('dt','')}"
                + (f"　令號　{r.get('decree')}" if r.get("decree") else "")
                + f"\n官網原文　{r.get('u','')}"
                + f"\n本卡片區塊　{ '、'.join(lab for lab in ORDER if lab in secs) }"
                + ("　＋署名列（補收）" if roster.get(r["k"]) else "")
                + (f"　＋意見書清單（{len(r.get('o') or [])}篇）" if r.get("o") else "")
                + (f"（另有：{'、'.join(extra)}）" if extra else "")
                + f"\n相關法條（官網欄位，供參）　{'、'.join(r.get('l') or [])[:400]}"
                + "\n" + "=" * 78 + "\n")
        (CARD / f"{r['k']}.txt").write_text(head + text + "\n", encoding="utf-8")
        work.append({"k": r["k"], "z": r.get("z", ""), "n": r.get("n", ""),
                     "ty": r.get("ty", ""), "u": r.get("u", ""),
                     "roc": r.get("roc") or r.get("dt", ""),
                     "區塊": [lab for lab in ORDER if lab in secs] + extra,
                     "有相關文件": "相關文件" in secs,
                     "有署名列補收": bool(roster.get(r["k"])),
                     "意見書篇數": len(r.get("o") or []),
                     "chars": len(text)})

    json.dump(work, open(OUT / "工作檔.json", "w", encoding="utf-8"), ensure_ascii=False)

    # 平衡分批：先降冪，再貪心裝箱
    order = sorted(work, key=lambda w: -w["chars"])
    bins = []
    for w in order:
        b = min(bins, key=lambda x: x["c"]) if bins else None
        if b is None or (b["c"] + w["chars"] > TARGET and len(bins) < 400):
            bins.append({"ks": [w["k"]], "c": w["chars"]})
        else:
            b["ks"].append(w["k"])
            b["c"] += w["chars"]
    bins.sort(key=lambda x: -x["c"])
    json.dump([b["ks"] for b in bins], open(OUT / "批次.json", "w", encoding="utf-8"),
              ensure_ascii=False)

    with open(OUT / "進度.csv", "w", encoding="utf-8-sig", newline="") as fh:
        w = csv.writer(fh)
        w.writerow(["批次", "釋憲案", "字數", "有相關文件", "狀態"])
        for i, b in enumerate(bins, 1):
            for k in b["ks"]:
                it = next(x for x in work if x["k"] == k)
                w.writerow([f"B{i:03d}", k, it["chars"], int(it["有相關文件"]), "未標註"])

    ch = sorted(x["chars"] for x in work)
    print(f"卡片 {len(work)} 件，合計 {sum(ch):,} 字")
    print(f"字數　中位 {ch[len(ch)//2]:,}／最長 {ch[-1]:,}／最短 {ch[0]:,}")
    print(f"有相關文件（聲請書）者 {sum(1 for x in work if x['有相關文件'])} 件")
    print(f"補收署名列 {sum(1 for x in work if x['有署名列補收'])} 件；"
          f"意見書清單 {sum(1 for x in work if x['意見書篇數'])} 件／"
          f"{sum(x['意見書篇數'] for x in work)} 篇")
    print(f"分批 {len(bins)} 批，每批目標 {TARGET:,} 字，"
          f"實際最大 {bins[0]['c']:,}／最小 {bins[-1]['c']:,}")
    print("卡片在", CARD)


if __name__ == "__main__":
    main()