From 38790e21478c99a7b85ec8def6b7bd82a7b879da Mon Sep 17 00:00:00 2001
From: Color2333 <1552429809@qq.com>
Date: Mon, 20 Jul 2026 03:37:09 +0800
Subject: [PATCH] =?UTF-8?q?fix(tools):=20search.py=20=E7=9F=AD=E5=89=8D?=
 =?UTF-8?q?=E7=BC=80=20paper=5Fid=20=E5=AF=BC=E8=87=B4=20get=5Fpaper=5Fdet?=
 =?UTF-8?q?ail/get=5Fsimilar/citation=5Ftree=20=E5=B4=A9?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

PR #60 修了 read/figures/reasoning，但 search.py 仍有同样 bug：
- _get_paper_detail: line 53 用 UUID(paper_id) 重新查（既多余又崩短前缀），
  且 line 49 已调 _require_paper 拿到完整论文对象，直接用即可
- _get_similar_papers: line 78 UUID(paper_id) 崩；line 107 返回原始短前缀给 LLM
- _get_citation_tree: line 181 传原始 paper_id 给 GraphService.citation_tree，
  内部用 paper.id 作 dict key，短前缀不匹配 → root_paper 为 None

修复：
- _get_paper_detail: 用 _require_paper 返回的 p 直接构造 data，删重复 UUID 查询
- _get_similar_papers: pid = UUID(paper.id)；返回 paper.id 给 LLM
- _get_citation_tree: 用 paper.id 传给 citation_tree

实测前：生产 b7e07388 → LLM 先调 get_paper_detail → "badly formed hexadecimal UUID"，
LLM 改调 search_papers(keyword=b7e07388) → 0 结果 → 放弃。
修复后：get_paper_detail 直接用 _require_paper 解析的完整论文，短前缀可用。

测试：69 passed / 2 skipped（无回归）。
---
 packages/ai/tools/handlers/search.py | 50 +++++++++++++++-------------
 1 file changed, 26 insertions(+), 24 deletions(-)

diff --git a/packages/ai/tools/handlers/search.py b/packages/ai/tools/handlers/search.py
index 4a64ad6..e097b1c 100644
--- a/packages/ai/tools/handlers/search.py
+++ b/packages/ai/tools/handlers/search.py
@@ -49,33 +49,34 @@ def _get_paper_detail(paper_id: str) -> ToolResult:
     p, err = _require_paper(paper_id)
     if err:
         return err
-    with session_scope() as session:
-        p = PaperRepository(session).get_by_id(UUID(paper_id))
-        title = p.title or ""
-        data = {
-            "id": str(p.id),
-            "title": title,
-            "arxiv_id": p.arxiv_id,
-            "abstract": (p.abstract or "")[:1000],
-            "publication_date": str(p.publication_date) if p.publication_date else None,
-            "read_status": p.read_status.value,
-            "pdf_path": p.pdf_path,
-            "has_embedding": p.embedding is not None,
-            "categories": (p.metadata_json or {}).get("categories", []),
-            "authors": (p.metadata_json or {}).get("authors", []),
-        }
-        return ToolResult(
-            success=True,
-            data=data,
-            summary=f"论文: {title[:60]}" + ("..." if len(title) > 60 else ""),
-        )
+    # p 已是 _require_paper 解析出的完整论文对象（detached 但属性已加载）。
+    # 不再用 UUID(paper_id) 重新查——短前缀会崩，且重复查询无意义。
+    title = p.title or ""
+    data = {
+        "id": str(p.id),
+        "title": title,
+        "arxiv_id": p.arxiv_id,
+        "abstract": (p.abstract or "")[:1000],
+        "publication_date": str(p.publication_date) if p.publication_date else None,
+        "read_status": p.read_status.value,
+        "pdf_path": p.pdf_path,
+        "has_embedding": p.embedding is not None,
+        "categories": (p.metadata_json or {}).get("categories", []),
+        "authors": (p.metadata_json or {}).get("authors", []),
+    }
+    return ToolResult(
+        success=True,
+        data=data,
+        summary=f"论文: {title[:60]}" + ("..." if len(title) > 60 else ""),
+    )
 
 
 def _get_similar_papers(paper_id: str, top_k: int = 5) -> ToolResult:
     paper, err = _require_paper(paper_id)
     if err:
         return err
-    pid = UUID(paper_id)
+    # 用 paper.id（完整 UUID），不用原始短前缀调 UUID()
+    pid = UUID(paper.id)
     if not paper.embedding:
         return ToolResult(
             success=False,
@@ -103,7 +104,7 @@ def _get_similar_papers(paper_id: str, top_k: int = 5) -> ToolResult:
         return ToolResult(
             success=True,
             data={
-                "paper_id": paper_id,
+                "paper_id": paper.id,
                 "similar_ids": [str(x) for x in ids],
                 "items": items,
             },
@@ -173,11 +174,12 @@ def on_progress(msg: str) -> None:
 
 
 def _get_citation_tree(paper_id: str, depth: int = 2) -> ToolResult:
-    _, err = _require_paper(paper_id)
+    paper, err = _require_paper(paper_id)
     if err:
         return err
     try:
-        result = GraphService().citation_tree(root_paper_id=paper_id, depth=depth)
+        # 用完整 UUID（paper.id），不用原始短前缀（citation_tree 内部用 paper.id 作 dict key）
+        result = GraphService().citation_tree(root_paper_id=paper.id, depth=depth)
         node_count = len(result.get("nodes", []))
         edge_count = len(result.get("edges", []))
         return ToolResult(