Update app.py

app.py

@@ -1,10 +1,13 @@
 import re
+import json
 import requests
 import gradio as gr
 
-BASE_URL = "https://laws.e-gov.go.jp/api/2/laws"
+LIST_URL = "https://laws.e-gov.go.jp/api/2/laws"
+GET_URL_A = "https://laws.e-gov.go.jp/api/2/laws/{lawId}"
+GET_URL_B = "https://laws.e-gov.go.jp/api/2/laws"  # ?lawId= フォールバック
 
-# よく使う法令はIDを直持ち（民法ほか数件）
+# よく使う法令のID（即ヒット用）
 KNOWN = {
     "民法": "429AC0000000089",
     "会社法": "429AC0000000086",
@@ -13,8 +16,8 @@ KNOWN = {
     "不動産登記法": "429AC0000000092",
 }
 
-# ざっくり漢数字変換（条番号用：1〜399くらいまでで十分）
 KANJI_DIG = {0:"零",1:"一",2:"二",3:"三",4:"四",5:"五",6:"六",7:"七",8:"八",9:"九"}
+
 def int_to_kanji(n:int)->str:
     if n <= 10:
         return ["零","一","二","三","四","五","六","七","八","九","十"][n]
@@ -32,80 +35,141 @@ def int_to_kanji(n:int)->str:
 
 def parse_query(q:str):
     q = q.strip().replace("　","")
-    # 「民法3上」みたいな誤入力も「条」に補正
-    q = re.sub(r"(?<=\d)上$", "条", q)
-    m = re.match(r"(?P<law>[^0-9条第第\s]+)\s*(第?\s*(?P<num>\d+)\s*条?)", q)
+    q = re.sub(r"(?<=\d)上$", "条", q)  # 「3上」を補正
+    m = re.match(r"(?P<law>[^0-9条第\s]+)\s*(第?\s*(?P<num>\d+)\s*条?)", q)
     if not m:
-        # 例: 「民法3条」「民法 第3条」「民法3」
         m = re.match(r"(?P<law>[^0-9\s]+)\s*(?P<num>\d+)", q)
     if not m:
         raise ValueError("「民法3条」「会社法10条」みたいに入力して。")
-    law = m.group("law")
-    num = int(m.group("num"))
-    return law, num
+    return m.group("law"), int(m.group("num"))
 
 def find_law_id(law_name:str)->str:
-    # まず内蔵マップ
     if law_name in KNOWN:
         return KNOWN[law_name]
-    # キーワード検索で候補取得（v2の一覧API）
-    params = {"keyword": law_name, "num": 20}
-    r = requests.get(BASE_URL, params=params, timeout=15)
+    # キーワード検索（一覧API）
+    r = requests.get(LIST_URL, params={"keyword": law_name, "num": 25}, timeout=20)
     r.raise_for_status()
-    data = r.json()
-    infos = data.get("lawListInfo", {}).get("lawInfo", [])
+    infos = r.json().get("lawListInfo", {}).get("lawInfo", [])
     if not infos:
         raise ValueError(f"法令が見つからない：「{law_name}」")
-    # 完全一致＞部分一致の順で拾う
+    # 完全一致優先
     for it in infos:
         if it.get("lawName") == law_name:
             return it.get("lawId")
     return infos[0].get("lawId")
 
-def get_article_text(law_id:str, article_no:int)->str:
-    # v2の法令取得（lawId指定）。articles配列から該当条を拾う前提
-    r = requests.get(BASE_URL, params={"lawId": law_id}, timeout=30)
-    r.raise_for_status()
-    j = r.json()
-
-    law_name = j.get("lawName", "不明な法令")
-    articles = j.get("articles", [])
-
-    target_label = f"第{int_to_kanji(article_no)}条"
-    # articles: [{articleTitle, articleNumber, paragraphs:[{paragraphSentence:..}, ...]}, ...]
-    for a in articles:
-        # API仕様により articleNumber / articleTitle のどちらかに「第◯条」が入るケースを吸収
-        number = a.get("articleNumber") or a.get("articleTitle") or ""
-        if target_label in str(number):
-            paras = a.get("paragraphs", [])
-            # 段落を結合
-            body = []
-            for p in paras:
-                # 号・項などの入れ子にも軽く対応
-                s = p.get("paragraphSentence") or ""
-                items = p.get("items") or []
-                body.append(s)
-                for it in items:
-                    isent = it.get("itemSentence") or ""
-                    body.append(isent)
-            content = "\n".join([x for x in body if x])
-            return f"{target_label}\n\n{content.strip() or '本文が見つかりませんでした。'}"
-    raise ValueError(f"{law_name} に {target_label} が見つからない。")
-
-def query_runner(q:str)->str:
+def fetch_law_json(law_id:str)->dict:
+    # 形式A: /laws/{lawId}
+    r = requests.get(GET_URL_A.format(lawId=law_id), timeout=30)
+    if r.status_code == 200:
+        return r.json()
+    # 形式B: /laws?lawId=...
+    r2 = requests.get(GET_URL_B, params={"lawId": law_id}, timeout=30)
+    r2.raise_for_status()
+    return r2.json()
+
+def flatten_text(x):
+    """dict/list/str をテキストにフラット化"""
+    if x is None:
+        return ""
+    if isinstance(x, str):
+        return x
+    if isinstance(x, list):
+        return "\n".join(flatten_text(i) for i in x)
+    if isinstance(x, dict):
+        parts = []
+        # よく出るキーを優先的に拾う
+        for k in ["paragraphSentence","itemSentence","supplProvisionLabel","chapterTitle","articleTitle","articleCaption","note"]:
+            if k in x:
+                parts.append(flatten_text(x[k]))
+        # その他のキーも一応辿る
+        for v in x.values():
+            parts.append(flatten_text(v))
+        return "\n".join(p for p in parts if p)
+    return ""
+
+def find_article_blocks(obj, target_label):
+    """
+    JSONツリーを再帰的に探索し、「第◯条」にマッチする塊（dict）を返す。
+    e-Gov v2 は Article が色々な入れ子に出現するので総当たりで探す。
+    """
+    hits = []
+
+    if isinstance(obj, dict):
+        # Article相当の塊を検知
+        cand_title = ""
+        for key in ["articleNumber","articleTitle","ArticleNumber","ArticleTitle"]:
+            if key in obj and isinstance(obj[key], str):
+                cand_title = obj[key]
+                break
+        if target_label in str(cand_title):
+            hits.append(obj)
+
+        # 再帰探索
+        for v in obj.values():
+            hits.extend(find_article_blocks(v, target_label))
+
+    elif isinstance(obj, list):
+        for v in obj:
+            hits.extend(find_article_blocks(v, target_label))
+
+    return hits
+
+def extract_article_text(article_block:dict)->str:
+    """
+    条文塊から本文を抽出。段落・号などをざっくり連結。
+    v2のJSONは schema が揺れるので、汎用的に全要素からテキストを吸い出す。
+    """
+    # まず見出し
+    title = ""
+    for k in ["articleTitle","articleNumber","ArticleTitle","ArticleNumber"]:
+        if k in article_block and isinstance(article_block[k], str):
+            title = article_block[k]
+            break
+
+    # 本文候補（paragraphs/items などを総なめ）
+    body = []
+    # 代表的な場所
+    for k in ["paragraphs","Paragraph","paragraph","items","Item"]:
+        if k in article_block:
+            body.append(flatten_text(article_block[k]))
+    # 念のため全体からも
+    body.append(flatten_text(article_block))
+
+    text = "\n".join([b for b in body if b]).strip()
+    # 見出し重複を軽く除去
+    text = re.sub(r"\n{2,}", "\n", text)
+    return (title + "\n\n" if title else "") + text
+
+def run(q:str)->str:
     try:
         law, num = parse_query(q)
         law_id = find_law_id(law)
-        return get_article_text(law_id, num)
+        data = fetch_law_json(law_id)
+
+        # 法令名
+        law_name = data.get("lawName") or data.get("LawName") or "不明な法令"
+        target = f"第{int_to_kanji(num)}条"
+
+        blocks = find_article_blocks(data, target)
+        if not blocks:
+            raise ValueError(f"{law_name} に {target} が見つからない。")
+
+        # 最初の一致を表示
+        text = extract_article_text(blocks[0]).strip()
+        if not text:
+            raise ValueError("本文の抽出に失敗。")
+        return f"{target}\n\n{text}"
+
     except Exception as e:
         return f"❌ エラー: {e}"
 
 demo = gr.Interface(
-    fn=query_runner,
-    inputs=gr.Textbox(label="例）民法3条 / 会社法10条 / 不動産登記法47条"),
-    outputs=gr.Textbox(label="結果", lines=16),
-    title="e-Gov 法令API v2｜条文ひき",
-    description="法令名＋条番号で、該当条文を取得して表示します（最小構成）。"
+    fn=run,
+    inputs=gr.Textbox(label="例）民法3条 / 会社法10条 / 不動産登記法47条（「3上」でもOK）"),
+    outputs=gr.Textbox(label="結果", lines=18),
+    title="e-Gov 法令API v2｜条文ピンポイント検索（最小構成）",
+    description="法令名＋条番号を入れると該当条文を表示。JSON構造の違いにも（わりと）頑健。"
 )
 
 demo.launch()