feat: 新增多格式文件上传入库与认证体系

- 新增 JWT 认证模块，支持登录/注册/用户管理 - 新增文件上传接口，支持 .txt/.md/.html/.pdf/.docx 等格式解析入库 - 新增检索结果 AI 总结功能 - 新增文本去重缓存机制 - 新增全局认证夹具简化测试 - 新增配置项与环境变量支持 - 完善文档与测试覆盖
2026-07-30 10:30:15 +08:00
parent 51dc8dc4f6
commit dce9e31bde
31 changed files with 3021 additions and 45 deletions
@@ -0,0 +1,138 @@
+"""ResultSummarizer 单元测试（FakeOllama，不依赖真实 Ollama）
+
+覆盖：
+- summarize 正常：调用 ollama，prompt 含 query 与 hits 文本，返回 strip 后的 summary
+- summarize 无 hits：返回空串且不调 ollama
+- summarize ollama 异常：返回空串
+- _build_context 拼接格式：编号 / title / section_path / 文本
+- 取前 settings.result_summary_max_hits 条：超量 hits 只取前 N 条进入 prompt
+"""
+
+import pytest
+
+from app.config import settings
+from app.core.result_summarizer import ResultSummarizer
+from app.models.search import SearchHit
+
+
+class FakeOllama:
+    """记录调用并返回固定响应的假 OllamaClient"""
+
+    def __init__(self, response: str = "这是总结") -> None:
+        self.response = response
+        self.calls: list[dict] = []
+
+    async def generate(self, prompt: str, json_mode: bool = False) -> str:
+        self.calls.append({"prompt": prompt, "json_mode": json_mode})
+        return self.response
+
+
+class FailingOllama:
+    """generate 抛异常的假 OllamaClient，用于测试容错降级"""
+
+    async def generate(self, prompt: str, json_mode: bool = False) -> str:
+        raise RuntimeError("ollama 不可用")
+
+
+def _hit(idx: int, title: str = "", section_path: str = "") -> SearchHit:
+    """构造测试用 SearchHit"""
+    return SearchHit(
+        text=f"文本内容-{idx}",
+        doc_id=f"doc-{idx}",
+        title=title or f"标题-{idx}",
+        section_path=section_path,
+        score=0.1 * idx,
+    )
+
+
+class TestSummarize:
+    """ResultSummarizer.summarize 主流程"""
+
+    async def test_summarize_normal(self):
+        """正常总结：调用 ollama，prompt 含 query 与 hits 文本，返回 strip 后的 summary"""
+        ollama = FakeOllama(response="  这是 AI 总结  ")
+        summarizer = ResultSummarizer(ollama=ollama)  # type: ignore[arg-type]
+        hits = [_hit(1, "文档A", "章节A"), _hit(2, "文档B", "章节B")]
+
+        result = await summarizer.summarize("安装步骤是什么", hits)
+
+        assert result == "这是 AI 总结"
+        assert len(ollama.calls) == 1
+        prompt = ollama.calls[0]["prompt"]
+        assert "安装步骤是什么" in prompt
+        assert "文本内容-1" in prompt
+        assert "文本内容-2" in prompt
+        assert "文档A" in prompt
+        assert "文档B" in prompt
+        assert "章节A" in prompt
+        assert "章节B" in prompt
+
+    async def test_summarize_empty_hits_returns_empty(self):
+        """无 hits：返回空串且不调用 ollama"""
+        ollama = FakeOllama()
+        summarizer = ResultSummarizer(ollama=ollama)  # type: ignore[arg-type]
+
+        result = await summarizer.summarize("任何问题", [])
+
+        assert result == ""
+        assert ollama.calls == []
+
+    async def test_summarize_ollama_error_returns_empty(self):
+        """ollama.generate 抛异常：返回空串（容错降级）"""
+        summarizer = ResultSummarizer(ollama=FailingOllama())  # type: ignore[arg-type]
+        hits = [_hit(1)]
+
+        result = await summarizer.summarize("问题", hits)
+
+        assert result == ""
+
+
+class TestBuildContext:
+    """_build_context 拼接格式：编号 / title / section_path / 文本"""
+
+    def test_build_context_format(self):
+        hits = [
+            _hit(1, "文档A", "第一章"),
+            _hit(2, "文档B", "第二章"),
+        ]
+        context = ResultSummarizer._build_context(hits)
+
+        # 每条带编号、title、section_path
+        assert "[1] / 文档A / 第一章" in context
+        assert "[2] / 文档B / 第二章" in context
+        # 文本内容拼接
+        assert "文本内容-1" in context
+        assert "文本内容-2" in context
+        # 分隔符
+        assert "---" in context
+
+    def test_build_context_empty_title_and_section(self):
+        """title 与 section_path 为空时头部仅保留编号"""
+        hit = SearchHit(text="纯文本", doc_id="d1", title="", section_path="", score=1.0)
+        context = ResultSummarizer._build_context([hit])
+        assert context.strip().startswith("[1]")
+        assert "纯文本" in context
+
+
+class TestMaxHitsLimit:
+    """取前 settings.result_summary_max_hits 条命中"""
+
+    async def test_only_first_n_hits_in_prompt(self, monkeypatch: pytest.MonkeyPatch):
+        """构造超过 max_hits 的 hits，验证 prompt 只含前 N 条文本"""
+        max_hits = 3
+        monkeypatch.setattr(settings, "result_summary_max_hits", max_hits)
+        ollama = FakeOllama(response="总结")
+        summarizer = ResultSummarizer(ollama=ollama)  # type: ignore[arg-type]
+
+        # 构造 max_hits + 5 条 hits
+        hits = [_hit(i) for i in range(max_hits + 5)]
+        await summarizer.summarize("问题", hits)
+
+        assert len(ollama.calls) == 1
+        prompt = ollama.calls[0]["prompt"]
+        # 前 max_hits 条文本出现在 prompt 中
+        for i in range(max_hits):
+            assert f"文本内容-{i}" in prompt
+        # 超出的文本不出现在 prompt 中
+        for i in range(max_hits, max_hits + 5):
+            assert f"文本内容-{i}" not in prompt