fix: 知识库key格式统一、过滤逻辑修复、布局宽度稳定

backend/app/services/knowledge_service.py

@@ -27,6 +27,30 @@ class KnowledgeService:
         "报题单": "baoti",
     }
 
+    # 来源大类固定显示顺序（制片人 Obsidian 习惯）
+    SOURCE_TYPE_ORDER = [
+        "manuscript",      # 节目文稿
+        "military_report", # 杂志文章
+        "baoti",           # 报题单
+        "manual",          # 其他
+    ]
+
+    # 二级分组维度映射（与前端 useKnowledgeGrouping.js 的 SECONDARY_GROUP_FIELD 保持一致）
+    # key = source_type, value = 用来做二级分组的字段名，None = 不建二级节点
+    SECONDARY_GROUP_FIELD = {
+        "manuscript": "author",            # 节目文稿 → 按作者(编导)
+        "military_report": "source_detail", # 杂志文章 → 按出处
+        "baoti": None,                     # 报题单 → 不分组
+        "manual": None,                    # 其他 → 不分组
+    }
+
+    SOURCE_TYPE_LABEL = {
+        "military_report": "杂志文章",
+        "manuscript": "节目文稿",
+        "baoti": "报题单",
+        "manual": "其他",
+    }
+
     def __init__(self):
         self.embedder = EmbeddingService()
 
@@ -243,46 +267,22 @@ class KnowledgeService:
 
     def get_grouped_items(self) -> list[dict]:
         """
-        按 source_type → source_detail 两层聚合，返回树形结构数据。
+        按 source_type → 二级字段（author / source_detail）两层聚合，返回树形结构数据。
+        按 SOURCE_TYPE_ORDER 固定顺序排列。
 
-        返回格式：
-        [
-          {
-            "key": "全部",
-            "label": "全部（N条）",
-            "count": 总条目数,
-            "children": [
-              {
-                "key": "manuscript",
-                "label": "节目文稿（n条）",
-                "count": n,
-                "children": [  # source_detail 为 null 时直接为空数组，不造空节点
-                  {"key": "...", "label": "具体出处", "count": n}
-                ]
-              },
-              ...
-            ]
-          }
-        ]
+        二级节点 key 格式：`{source_type}|{二级字段名}|{字段值}`
+        例：manuscript|author|左鑫
+            military_report|source_detail|航空知识 2026年第1期
 
-        注意：source_detail 为 null 的条目（如节目文稿、报题单），
-        归入对应 source_type 大类下，但不单独成子节点（大类的 children 直接为其条目列表，
-        但前端 Tree 组件渲染时，若 children 为空则不展示二级节点，大类直接可点击）。
+        二级字段值为 null / 空字串 → 归入对应大类，不造空节点。
         """
-        SOURCE_TYPE_LABEL = {
-            "military_report": "杂志文章",
-            "manuscript": "节目文稿",
-            "baoti": "报题单",
-            "manual": "其他",
-        }
-
         with Session(engine) as session:
             items = session.exec(select(KnowledgeItem)).all()
 
         total_count = len(items)
 
-        # 按 source_type 分组
-        type_groups: dict = {}
+        # 按 source_type 分组，初始化所有已知类别（保证空类别也按顺序出现）
+        type_groups: dict = {st: [] for st in self.SOURCE_TYPE_ORDER}
         for item in items:
             st = item.source_type or "manual"
             if st not in type_groups:
@@ -290,30 +290,48 @@ class KnowledgeService:
             type_groups[st].append(item)
 
         children = []
-        for st, st_items in type_groups.items():
-            # 按 source_detail 细分
-            detail_groups: dict = {}
-            for item in st_items:
-                tags = item.tags or {}
-                sd = tags.get("source_detail") if isinstance(tags, dict) else None
-                if sd not in detail_groups:
-                    detail_groups[sd] = []
-                detail_groups[sd].append(item)
+        for st in self.SOURCE_TYPE_ORDER:
+            st_items = type_groups.get(st, [])
+            if not st_items:
+                # 空类别也按顺序出现（0条）
+                children.append({
+                    "key": st,
+                    "label": f"{self.SOURCE_TYPE_LABEL.get(st, st)}（0条）",
+                    "count": 0,
+                    "children": [],
+                })
+                continue
 
-            # 构建二级节点：仅 source_detail 非 null 的才建子节点
+            secondary_field = self.SECONDARY_GROUP_FIELD.get(st)
             grandchildren = []
-            for sd, sd_items in detail_groups.items():
-                if sd is not None:
-                    grandchildren.append({
-                        "key": f"{st}|{sd}",
-                        "label": f"{sd}（{len(sd_items)}条）",
-                        "count": len(sd_items),
-                    })
 
-            label = SOURCE_TYPE_LABEL.get(st, st)
+            if secondary_field is not None:
+                # 按二级字段分组
+                detail_groups: dict = {}
+                for item in st_items:
+                    # 取二级字段值：author 直接取，source_detail 从 tags(JSONB) 取
+                    if secondary_field == "source_detail":
+                        tags = item.tags or {}
+                        sd = tags.get("source_detail") if isinstance(tags, dict) else None
+                        field_val = sd
+                    else:
+                        field_val = (getattr(item, secondary_field, None) or "").strip() or None
+
+                    if field_val not in detail_groups:
+                        detail_groups[field_val] = []
+                    detail_groups[field_val].append(item)
+
+                for sd, sd_items in detail_groups.items():
+                    if sd is not None:
+                        grandchildren.append({
+                            "key": f"{st}|{secondary_field}|{sd}",
+                            "label": f"{sd}（{len(sd_items)}条）",
+                            "count": len(sd_items),
+                        })
+
             children.append({
                 "key": st,
-                "label": f"{label}（{len(st_items)}条）",
+                "label": f"{self.SOURCE_TYPE_LABEL.get(st, st)}（{len(st_items)}条）",
                 "count": len(st_items),
                 "children": grandchildren,
             })
@@ -323,4 +341,4 @@ class KnowledgeService:
             "label": f"全部（{total_count}条）",
             "count": total_count,
             "children": children,
-        }]
+        }]