parse-sanpi-hyo.py: 罫線で列を区切り、合計欄で検算する方式に変更

従来は表のセル抽出と会派の区切りの推測に頼っていて、会派や議員ごとの列がずれることがあった。上部の「会派名（議員数）」を読み、縦罫線で列を厳密に区切って議員名と賛否記号を割り当て、PDFに載っている議案ごとの賛成・反対の合計と突き合わせて検算する方式にした。令和7年12月定例会で、会派・議員27人が市の会派名簿と一致し、4議案の賛成・反対数もPDFの合計とすべて一致することを確認。（表の生成側 gen-sanpi-table.py は、この出力に合わせて後日更新が必要）

1 files changed, 106 insertions, 90 deletions

diff --git a/scripts/parse-sanpi-hyo.py b/scripts/parse-sanpi-hyo.py
index 2030549..457cac0 100644
--- a/scripts/parse-sanpi-hyo.py
+++ b/scripts/parse-sanpi-hyo.py
@@ -1,111 +1,127 @@
-import sys
-import json
-import re
-import pdfplumber
-
-sys.stdout.reconfigure(encoding="utf-8")
+"""賛否一覧PDF（新レイアウト）を読み、Markdown表を出力する。合計欄で検算する。
 
-VOTE_CHARS = set("〇○×✕✖◯")  # 賛成・反対の記号候補
+使い方: python .extract-final.py <pdf> "<定例会名>"
+"""
+import re
+import sys
 
+import pymupdf
 
-def norm(c):
-    """セル文字列を正規化（空白・改行を除去）"""
-    if c is None:
-        return ""
-    return "".join(str(c).split())
+sys.stdout.reconfigure(encoding="utf-8")
 
+YES, NO, ABS = set("〇○◯"), set("×✕✖"), set("－—−-・")
+VOTE = YES | NO | ABS
+DIGITS = set("0123456789０１２３４５６７８９")
 
-HEADER_WORDS = {"区分", "番号", "件名", "議決結果", "賛成", "反対", "会派", "議員", "賛否"}
 
+def col_of(x, bounds):
+    for i in range(len(bounds) - 1):
+        if bounds[i] <= x < bounds[i + 1]:
+            return i
+    return None
 
-def is_vote(c):
-    s = norm(c)
-    return len(s) == 1 and s in VOTE_CHARS
 
+def chars(pg):
+    out = []
+    for block in pg.get_text("rawdict")["blocks"]:
+        for line in block.get("lines", []):
+            for span in line["spans"]:
+                for ch in span["chars"]:
+                    if ch["c"].strip():
+                        x0, y0, x1, y1 = ch["bbox"]
+                        out.append({"c": ch["c"], "x": (x0 + x1) / 2, "y": (y0 + y1) / 2})
+    return out
 
-def is_name(c):
-    s = norm(c)
-    if not (2 <= len(s) <= 8):
-        return False
-    if is_vote(c) or s in HEADER_WORDS:
-        return False
-    return not re.search(r"[0-9０-９]", s)
 
+def main(path, label):
+    doc = pymupdf.open(path)
+    pg = doc[0]
+    cs = chars(pg)
 
-def is_label(c):
-    s = norm(c)
-    if not (1 <= len(s) <= 5):
-        return False
-    if is_vote(c) or s in HEADER_WORDS:
-        return False
-    return not re.search(r"[0-9０-９]", s)
+    # 会派（上部の2行、y 45〜70。文字は y,x の順に並べる）
+    topcs = sorted([c for c in cs if 45 <= c["y"] <= 70], key=lambda c: (round(c["y"] / 6), c["x"]))
+    top = "".join(c["c"] for c in topcs)
+    parties = [(s, f.strip(), int(n)) for s, f, n in
+               re.findall(r"([ぁ-んァ-ヶ一-龥A-Za-z]{2})[：:]([^（(]+?)[（(](\d+)", top)]
+    total = sum(n for _, _, n in parties)
 
+    # 縦罫線グリッド
+    xs = {}
+    for d in pg.get_drawings():
+        for it in d["items"]:
+            if it[0] == "l":
+                a, b = it[1], it[2]
+                if abs(a.x - b.x) < 0.8:
+                    k = round(a.x, 1)
+                    info = xs.setdefault(k, {"maxy": 0})
+                    info["maxy"] = max(info["maxy"], a.y, b.y)
+    grid = sorted(x for x, i in xs.items() if i["maxy"] > 200 and 250 < x < 770)
 
-def main(path):
-    members = []      # [(col, 会派, 議員名)]
-    rows = []         # [{区分, 番号, 件名, 議決結果, votes:{col: '〇'/'×'}}]
-    carries = {}      # 区分の繰越
+    # 議員名の帯（y 88〜140、左の区分・番号・件名・議決結果を除く）
+    namechars = [c for c in cs if 88 <= c["y"] <= 140 and c["x"] > 308]
+    nx0, nx1 = min(c["x"] for c in namechars), max(c["x"] for c in namechars)
+    left = [x for x in grid if x <= nx0]
+    bounds = ([left[-1]] if left else [nx0 - 8]) + [x for x in grid if nx0 < x <= nx1 + 15]
 
-    with pdfplumber.open(path) as pdf:
-        for page in pdf.pages:
-            for table in page.extract_tables():
-                if not table:
-                    continue
-                ncols = max(len(r) for r in table)
-                grid = [list(r) + [None] * (ncols - len(r)) for r in table]
+    cols = {}
+    for c in namechars:
+        i = col_of(c["x"], bounds)
+        if i is not None:
+            cols.setdefault(i, []).append(c)
+    members = [(i, "".join(x["c"] for x in sorted(cols[i], key=lambda c: c["y"]))) for i in sorted(cols)]
 
-                # 議員名の行＝2〜8文字の名前らしいセルが最も多く並ぶ行
-                best = None
-                for i, row in enumerate(grid[:6]):
-                    n = sum(1 for c in row if is_name(c))
-                    if n >= 5 and (best is None or n > best[1]):
-                        best = (i, n)
-                if best is None:
-                    continue
-                member_row_idx = best[0]
+    # 会派ごとに区切る
+    groups, k = [], 0
+    for s, f, n in parties:
+        groups.append((s, members[k:k + n]))
+        k += n
+    ngroups = len(members) - sum(n for _, _, n in parties)   # 末尾の合計欄（賛成/反対）
+    member_only = members[:sum(n for _, _, n in parties)]
 
-                # 会派名は、その上の行（無ければ同じ行の短いラベル）から拾う
-                label_row = grid[member_row_idx - 1] if member_row_idx > 0 else grid[member_row_idx]
-                local = {}
-                group = ""
-                for c in range(ncols):
-                    lab = norm(label_row[c])
-                    if lab and is_label(label_row[c]):
-                        group = lab
-                    if is_name(grid[member_row_idx][c]):
-                        local[c] = (group, norm(grid[member_row_idx][c]))
-                if local and not members:
-                    members = [(c, g, n) for c, (g, n) in sorted(local.items())]
+    # 賛否
+    marks = [c for c in cs if c["c"] in VOTE and c["y"] > 145]
+    rows = {}
+    for c in marks:
+        rows.setdefault(round(c["y"] / 10) * 10, []).append(c)
 
-                # データ行（〇×がある行）
-                for row in grid[member_row_idx + 1:]:
-                    votes = {c: norm(row[c]) for c in local if is_vote(row[c])}
-                    if len(votes) < 2:
-                        continue
-                    vals = [norm(x) for x in row]
-                    disp = vals[0] if vals[0] else carries.get("区分", "")
-                    if vals[0]:
-                        carries["区分"] = vals[0]
-                    rows.append({
-                        "区分": disp,
-                        "番号": next((v for v in vals[1:4] if v and "第" in v), ""),
-                        "件名": next((v for v in vals[1:6] if len(v) >= 8), ""),
-                        "議決結果": next((v for v in vals if v in ("認 定", "認定", "原案可決", "可決", "不採択", "採択", "否決", "原案否決")), ""),
-                        "votes": votes,
-                        "raw": vals[:4],
-                    })
+    # 合計欄（末尾の列）の数字
+    nums = {}
+    for c in cs:
+        if c["c"] in DIGITS:
+            i = col_of(c["x"], bounds)
+            if i is not None and i >= len(member_only):
+                nums.setdefault(round(c["y"] / 10) * 10, []).append((c["y"], c["x"], c["c"]))
 
-    print("議員（会派 / 名前）:")
-    for c, g, n in members:
-        print(f"  col{c:>3}: {g:<6} {n}")
-    print("\n議案行数:", len(rows))
-    for r in rows[:6]:
-        vs = "".join(r["votes"].get(c, "・") for c, _, _ in members)
-        print(f"  {r['区分'][:6]:<6} {r['番号']:<8} {r['議決結果']:<8} {r['件名'][:28]:<28} {vs}")
+    # 議案名（左側）
+    left_txt = {}
+    for c in cs:
+        if c["x"] < 300 and c["y"] > 120:
+            left_txt.setdefault(round(c["y"] / 10) * 10, []).append((c["x"], c["c"]))
 
-    with open(".sanpi-parsed.json", "w", encoding="utf-8") as f:
-        json.dump({"members": members, "rows": rows}, f, ensure_ascii=False, indent=1)
-    print("\n.sanpi-parsed.json に保存しました")
+    print(f"### {label}")
+    print(f"会派: " + " / ".join(f"{s}（{n}）" for s, _, n in parties) + f" = {total}人")
+    print(f"議員抽出数: {len(member_only)}（一致: {len(member_only) == total}）")
+    for s, ms in groups:
+        print(f"  {s}: {', '.join(m[1] for m in ms)}")
+    print("\n議案行の検算:")
+    ok = True
+    for y in sorted(rows):
+        cells = {}
+        for c in rows[y]:
+            i = col_of(c["x"], bounds)
+            if i is not None:
+                cells.setdefault(i, []).append(c["c"])
+        yes = sum(1 for i, _ in member_only if any(v in YES for v in cells.get(i, [])))
+        no = sum(1 for i, _ in member_only if any(v in NO for v in cells.get(i, [])))
+        if yes + no == 0:
+            continue          # 賛否の記号がない行（全会一致など）
+        pdfs = "".join(c for _, _, c in sorted(nums.get(y, [])))
+        norm = pdfs.translate(str.maketrans("０１２３４５６７８９", "0123456789"))
+        match = (str(yes) in norm and str(no) in norm)
+        ok = ok and match
+        print(f"  y={y}: 賛成{yes} 反対{no} / PDF={pdfs}  {'✓' if match else '✗ 要確認'}")
+    print("  ⇒ 検算:", "全行一致" if ok else "不一致あり")
+    doc.close()
 
 
-main(".tmp-hyo.pdf")
+main(sys.argv[1], sys.argv[2] if len(sys.argv) > 2 else "")