Aislo/resources/knowledge/original/_pipeline/cak_hwp_md.py

# -*- coding: utf-8 -*-
"""건설업 임금실태조사 보고서 HWP(5.x) → 전체본 md + 장별 md 4개.  (PDF 가 없는 회차용)

사용: python cak_hwp_md.py <회차 폴더> ...   (폴더 안에 .hwp 한 개)
- HWP 본문 레코드에서 글·표(칸 좌표 · 병합)를 직접 읽음. 쪽 구분은 HWP 에 없으므로 쪽 표기 없음.
- 검증은 cak_verify.py 의 회차 사이 대조(같은 공표일 값이 다른 회차 PDF 에도 실려 있음)로 함.
"""

import re
import struct
import sys
import zlib
from pathlib import Path

import olefile

NUM = r"\(※\d{1,3}(?:,\d{3})+\)|\d{1,3}(?:,\d{3})+|(?<![\d,])-(?![\d,])"
INFO = {"2020-09-01": ("2020-08-31", 5311), "2021-01-01": ("2021-01-01", 5312)}
CHAPTERS = [
    ("1. 조사개요", "Ⅰ.조사개요"),
    ("2. 임금적용요령", "Ⅱ.임금적용요령"),
    ("3. 개별직종 노임단가", "Ⅲ.개별직종노임단가"),
    ("4. 직종해설", "Ⅳ.직종해설"),
]


def para_text(b):
    """PARA_TEXT → 글. 제어문자(8 단위 차지) 건너뜀."""
    u = struct.unpack("<%dH" % (len(b) // 2), b)
    out, i = [], 0
    while i < len(u):
        c = u[i]
        if c >= 32:
            out.append(chr(c))
            i += 1
        elif c == 9:
            out.append("\t")
            i += 8
        elif c in (1, 2, 3, 4, 5, 6, 7, 8, 11, 12, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23):
            i += 8
        elif c in (10, 13):
            i += 1
        else:
            i += 1
    return "".join(out)


def read_blocks(path):
    """→ [("p", 글) | ("t", 칸목록[(row, col, colspan, rowspan, 글)] , 행수, 열수)]"""
    o = olefile.OleFileIO(str(path))
    comp = o.openstream("FileHeader").read()[36] & 1
    recs = []
    i = 0
    while o.exists(f"BodyText/Section{i}"):
        d = o.openstream(f"BodyText/Section{i}").read()
        if comp:
            d = zlib.decompress(d, -15)
        p = 0
        while p < len(d):
            h = struct.unpack_from("<I", d, p)[0]
            tag, lvl, sz = h & 0x3FF, (h >> 10) & 0x3FF, h >> 20
            p += 4
            if sz == 0xFFF:
                sz = struct.unpack_from("<I", d, p)[0]
                p += 4
            recs.append((tag, lvl, d[p : p + sz]))
            p += sz
        i += 1
    blocks = []
    k = 0
    while k < len(recs):
        tag, lvl, b = recs[k]
        if tag == 71 and b[:4] == b" lbt":  # 표
            tl = lvl
            rows = cols = 0
            cells = []
            cur = None
            k += 1
            while k < len(recs) and (
                recs[k][1] > tl or recs[k][0] in (72, 77) and recs[k][1] >= tl
            ):
                t2, l2, b2 = recs[k]
                if t2 == 77:
                    rows, cols = struct.unpack_from("<HH", b2, 4)
                elif t2 == 72 and l2 == tl + 1:
                    col, row, cs, rs = struct.unpack_from("<HHHH", b2, 8)
                    cur = [row, col, cs, rs, []]
                    cells.append(cur)
                elif t2 == 67 and cur is not None:
                    cur[4].append(para_text(b2).rstrip("\r"))
                k += 1
            blocks.append(
                (
                    "t",
                    [
                        (r, c, cs, rs, "<br>".join(x.strip() for x in t if x.strip() != ""))
                        for r, c, cs, rs, t in cells
                    ],
                    rows,
                    cols,
                )
            )
            continue
        if tag == 67:
            blocks.append(("p", para_text(b).rstrip("\r")))
        k += 1
    return blocks


def esc(s):
    return s.replace("|", "\\|")


def md_grid(cells, rows, cols):
    g = [[""] * cols for _ in range(rows)]
    for r, c, cs, rs, t in cells:
        if r < rows and c < cols:
            g[r][c] = t
    o = []
    for i, row in enumerate(g):
        o.append("| " + " | ".join(esc(x) for x in row) + " |")
        if i == 0:
            o.append("|" + "---|" * cols)
    return o


def jobs_table(cells):
    """개별직종 표 → 행 [(코드, 이름, 신뢰도, [4값])] · 머리 글"""
    byrow = {}
    for r, c, cs, rs, t in cells:
        byrow.setdefault(r, {})[c] = t
    rows, head = [], []
    for r in sorted(byrow):
        row = [byrow[r][c] for c in sorted(byrow[r])]
        m = re.fullmatch(r"(\*{0,2})\s*(\d{4})", row[0].strip()) if row else None
        if m and len(row) >= 6:
            vals = row[-4:]
            rows.append((m.group(2), "".join(row[1:-4]).replace(" ", ""), m.group(1), vals))
        else:
            head.append(row)
    return rows, head


def convert(folder):
    folder = Path(folder)
    hwp = next(folder.glob("*.hwp"))
    stem = hwp.stem
    apply = folder.name
    reg, seq = INFO[apply]
    blocks = read_blocks(hwp)
    title = f"{stem[:4]}년 {stem[4:6]} 적용"
    hdr = (
        f"> 원문: {hwp.name} (대한건설협회, 게시 {reg}, 적용 {apply})\n"
        "> 변환: cak_hwp_md.py — HWP 본문에서 글 · 표 칸 좌표를 직접 읽음. HWP 에는 쪽 구분이 없어 쪽 표기 없음 · 원문 PDF 없음(쪽 그림 대조 불가) — "
        "표 값은 같은 공표일이 실린 다른 회차 PDF 와 기계 대조(cak_verify.py). 표 안 글자 사이 벌림은 붙여 씀.\n\n"
    )
    # 장 나누기: 표 밖 문단 중 목차(탭 없는) 장 제목
    idx = {}
    for i, b in enumerate(blocks):
        if b[0] == "p" and "\t" not in b[1]:
            n = re.sub(r"\s", "", b[1])
            for name, key in CHAPTERS:
                if n == key and name not in idx:
                    idx[name] = i
    order = [idx[n] for n, _ in CHAPTERS]
    assert order == sorted(order), idx
    bounds = order + [len(blocks)]

    def render(bs):
        out = []
        for b in bs:
            if b[0] == "p":
                out.append(b[1] + "  ")
                continue
            _, cells, rows, cols = b
            jr, head = jobs_table(cells)
            if len(jr) >= 10:  # 개별직종 표
                dates = []
                for h in head:
                    dates += [x for x in h if re.fullmatch(r"\d{4}\.\d{1,2}\.\d{1,2}", x.strip())]
                out.append("")
                out.append("| 직종코드 | 직종명 | 신뢰도 | " + " | ".join(dates[:4]) + " |")
                out.append("|---|---|---|---|---|---|---|")
                for c, n, f, v in jr:
                    out.append(f"| {c} | {n} | {f} | " + " | ".join(esc(x) for x in v) + " |")
                out.append("")
            else:
                out.append("")
                out += md_grid(cells, rows, cols)
                out.append("")
        return out

    pre = render(blocks[: order[0]])
    chapters = {}
    for j, (name, _) in enumerate(CHAPTERS):
        chapters[name] = render(blocks[bounds[j] : bounds[j + 1]])
    full_md = (
        f"# {title}\n\n> 원본: `{hwp.name}`\n\n"
        + "\n".join(pre + [x for n, _ in CHAPTERS for x in chapters[n]])
        + "\n"
    )
    (folder / f"{stem}.md").write_text(full_md, encoding="utf-8")
    for name, _ in CHAPTERS:
        (folder / f"{name}.md").write_text(
            f"# {name.split('. ', 1)[1]}\n\n{hdr}" + "\n".join(chapters[name]) + "\n",
            encoding="utf-8",
        )
    return len([1 for b in blocks if b[0] == "t"])


if __name__ == "__main__":
    for f in sys.argv[1:]:
        print(f, "표", convert(f))