#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
把意見書 PDF 的文字抽出並填入筆記中的 OPINION_TEXT_PLACEHOLDER。
品質檢查：中文字比例過低者視為抽取失敗，改走 OCR（tesseract chi_tra），
仍失敗則在筆記中明白標示「PDF 文字層無法擷取」，絕不留下空白或假內容。
"""
import json, re, subprocess, sys, os, shutil, tempfile
from pathlib import Path

CJK = re.compile(r'[一-鿿]')

def pdftotext(p):
    try:
        r = subprocess.run(['pdftotext', '-enc', 'UTF-8', '-nopgbrk', str(p), '-'],
                           capture_output=True, timeout=180)
        return r.stdout.decode('utf-8', 'replace')
    except Exception:
        return ''

def ocr(p):
    if not shutil.which('tesseract') or not shutil.which('pdftoppm'):
        return ''
    with tempfile.TemporaryDirectory() as td:
        subprocess.run(['pdftoppm', '-r', '200', '-png', str(p), td + '/pg'],
                       capture_output=True, timeout=900)
        chunks = []
        for img in sorted(Path(td).glob('pg*.png')):
            r = subprocess.run(['tesseract', str(img), 'stdout', '-l', 'chi_tra'],
                               capture_output=True, timeout=300)
            chunks.append(r.stdout.decode('utf-8', 'replace'))
        return '\n'.join(chunks)

def quality(t):
    t2 = re.sub(r'\s', '', t)
    if len(t2) < 80:
        return 0.0
    return len(CJK.findall(t2)) / len(t2)

def clean(t):
    t = t.replace('\r\n', '\n').replace('\x0c', '\n')
    t = re.sub(r'\n{3,}', '\n\n', t)
    return t.strip()

def main(pdfdir, notesroot, manifest_path):
    pdfdir, notesroot = Path(pdfdir), Path(notesroot)
    manifest = json.load(open(manifest_path, encoding='utf-8'))
    by_file = {}
    for key, items in manifest.items():
        for it in items:
            by_file[it['檔名']] = (key, it['下載id'], it['標題'])

    notes = {p.stem: p for p in list((notesroot / '10_釋字').glob('*.md')) + list((notesroot / '20_憲判').glob('*.md'))}
    # key（釋字0742）→ 筆記檔
    def note_for(key):
        if key.startswith('釋字'):
            return notes.get('釋字第%04d號' % int(key[2:]))
        m = re.match(r'憲判(\d+)-(\d+)', key)
        if m:
            return notes.get('%d年憲判字第%02d號' % (int(m.group(1)), int(m.group(2))))
        return None

    cache, stats = {}, {'ok': 0, 'ocr': 0, 'fail': 0, 'missing_note': 0}
    pdfs = sorted(pdfdir.rglob('*.pdf'))
    for i, p in enumerate(pdfs, 1):
        meta = by_file.get(p.name)
        if not meta:
            continue
        key, dlid, title = meta
        txt = clean(pdftotext(p))
        q = quality(txt)
        if q < 0.25:
            o = clean(ocr(p))
            if quality(o) >= 0.25:
                txt, q = o, quality(o)
                txt += '\n\n<!-- 本篇以 OCR 擷取，可能有辨識誤差 -->'
                stats['ocr'] += 1
            else:
                stats['fail'] += 1
                txt = ''
        else:
            stats['ok'] += 1
        cache.setdefault(key, {})[dlid] = (txt, q)
        if i % 200 == 0:
            print(f'  {i}/{len(pdfs)}', flush=True)

    for key, m in cache.items():
        np_ = note_for(key)
        if not np_:
            stats['missing_note'] += 1
            continue
        s = np_.read_text(encoding='utf-8')
        for dlid, (txt, q) in m.items():
            ph = f'<!--OPINION_TEXT_PLACEHOLDER:{dlid}-->'
            if ph not in s:
                continue
            body = txt if txt else '> [!failure] 本篇意見書之 PDF 無可擷取文字層，且 OCR 亦未通過品質門檻。請點上方原始檔連結查閱官網 PDF。'
            s = s.replace(ph, body)
        np_.write_text(s, encoding='utf-8')

    print(json.dumps(stats, ensure_ascii=False))

if __name__ == '__main__':
    main(sys.argv[1], sys.argv[2], sys.argv[3])