auto-coder 1.0.0__py3-none-any.whl → 2.0.1__py3-none-any.whl

autocoder/common/pruner/test_conversation_normalizer.py

@@ -0,0 +1,502 @@
+
+
+"""
+Test cases for ConversationNormalizer
+
+This module contains comprehensive tests for the conversation normalization functionality,
+including merge and delete strategies, edge cases, and validation.
+"""
+
+import pytest
+from typing import List, Dict, Any
+from autocoder.common.pruner.conversation_normalizer import (
+    ConversationNormalizer,
+    NormalizationStrategy,
+    NormalizationResult
+)
+
+
+class TestConversationNormalizer:
+    """测试会话规整化器的核心功能"""
+    
+    def setup_method(self):
+        """测试前的设置"""
+        self.normalizer = ConversationNormalizer()
+    
+    def test_merge_strategy_basic(self):
+        """测试基本的合并策略"""
+        # 测试目标：验证连续的相同角色消息能够正确合并
+        conversations = [
+            {"role": "user", "content": "第一个用户消息"},
+            {"role": "user", "content": "第二个用户消息"},
+            {"role": "assistant", "content": "助手回复"}
+        ]
+        
+        result = self.normalizer.normalize_conversations(
+            conversations, 
+            strategy=NormalizationStrategy.MERGE
+        )
+        
+        # 验证结果
+        assert result.original_count == 3
+        assert result.normalized_count == 2
+        assert len(result.changes_made) == 1
+        assert result.strategy_used == "merge"
+        
+        # 验证合并后的内容
+        normalized = result.normalized_conversations
+        assert len(normalized) == 2
+        assert normalized[0]["role"] == "user"
+        assert "[消息 1]" in normalized[0]["content"]
+        assert "[消息 2]" in normalized[0]["content"]
+        assert "第一个用户消息" in normalized[0]["content"]
+        assert "第二个用户消息" in normalized[0]["content"]
+        assert normalized[1]["role"] == "assistant"
+        assert normalized[1]["content"] == "助手回复"
+    
+    def test_delete_strategy_basic(self):
+        """测试基本的删除策略"""
+        # 测试目标：验证连续的相同角色消息只保留第一个
+        conversations = [
+            {"role": "user", "content": "第一个用户消息"},
+            {"role": "user", "content": "第二个用户消息"},
+            {"role": "assistant", "content": "助手回复"}
+        ]
+        
+        result = self.normalizer.normalize_conversations(
+            conversations, 
+            strategy=NormalizationStrategy.DELETE
+        )
+        
+        # 验证结果
+        assert result.original_count == 3
+        assert result.normalized_count == 2
+        assert len(result.changes_made) == 1
+        assert result.strategy_used == "delete"
+        
+        # 验证删除后的内容
+        normalized = result.normalized_conversations
+        assert len(normalized) == 2
+        assert normalized[0]["role"] == "user"
+        assert normalized[0]["content"] == "第一个用户消息"  # 只保留第一个
+        assert normalized[1]["role"] == "assistant"
+        assert normalized[1]["content"] == "助手回复"
+        
+        # 验证变更记录
+        change = result.changes_made[0]
+        assert change["action"] == "delete"
+        assert change["deleted_count"] == 1
+        assert change["kept_message"]["content"] == "第一个用户消息"
+    
+    def test_complex_conversation_merge(self):
+        """测试复杂会话的合并策略"""
+        # 测试目标：验证多个连续组的正确处理
+        conversations = [
+            {"role": "user", "content": "用户消息1"},
+            {"role": "user", "content": "用户消息2"},
+            {"role": "user", "content": "用户消息3"},
+            {"role": "assistant", "content": "助手消息1"},
+            {"role": "assistant", "content": "助手消息2"},
+            {"role": "user", "content": "用户消息4"},
+            {"role": "assistant", "content": "助手消息3"}
+        ]
+        
+        result = self.normalizer.normalize_conversations(
+            conversations, 
+            strategy=NormalizationStrategy.MERGE
+        )
+        
+        # 验证结果
+        assert result.original_count == 7
+        assert result.normalized_count == 4
+        assert len(result.changes_made) == 2  # 两个合并操作
+        
+        normalized = result.normalized_conversations
+        assert len(normalized) == 4
+        
+        # 验证第一个合并（3个user消息）
+        assert normalized[0]["role"] == "user"
+        assert "[消息 1]" in normalized[0]["content"]
+        assert "[消息 2]" in normalized[0]["content"]
+        assert "[消息 3]" in normalized[0]["content"]
+        
+        # 验证第二个合并（2个assistant消息）
+        assert normalized[1]["role"] == "assistant"
+        assert "[消息 1]" in normalized[1]["content"]
+        assert "[消息 2]" in normalized[1]["content"]
+        
+        # 验证未合并的消息
+        assert normalized[2]["role"] == "user"
+        assert normalized[2]["content"] == "用户消息4"
+        assert normalized[3]["role"] == "assistant"
+        assert normalized[3]["content"] == "助手消息3"
+    
+    def test_no_normalization_needed(self):
+        """测试不需要规整化的正常会话"""
+        # 测试目标：验证正常的交替会话不会被修改
+        conversations = [
+            {"role": "user", "content": "用户消息"},
+            {"role": "assistant", "content": "助手回复"},
+            {"role": "user", "content": "用户追问"},
+            {"role": "assistant", "content": "助手再回复"}
+        ]
+        
+        result = self.normalizer.normalize_conversations(conversations)
+        
+        # 验证结果
+        assert result.original_count == 4
+        assert result.normalized_count == 4
+        assert len(result.changes_made) == 0
+        assert result.normalized_conversations == conversations
+    
+    def test_empty_conversations(self):
+        """测试空会话列表"""
+        # 测试目标：验证空输入的正确处理
+        result = self.normalizer.normalize_conversations([])
+        
+        assert result.original_count == 0
+        assert result.normalized_count == 0
+        assert len(result.changes_made) == 0
+        assert result.normalized_conversations == []
+    
+    def test_single_message(self):
+        """测试单条消息"""
+        # 测试目标：验证单条消息不会被修改
+        conversations = [{"role": "user", "content": "单条消息"}]
+        
+        result = self.normalizer.normalize_conversations(conversations)
+        
+        assert result.original_count == 1
+        assert result.normalized_count == 1
+        assert len(result.changes_made) == 0
+        assert result.normalized_conversations == conversations
+    
+    def test_preserve_additional_fields(self):
+        """测试保留消息的额外字段"""
+        # 测试目标：验证合并时保留除role和content外的其他字段
+        conversations = [
+            {"role": "user", "content": "消息1", "timestamp": "2024-01-01", "id": "msg1"},
+            {"role": "user", "content": "消息2", "timestamp": "2024-01-02", "id": "msg2"},
+            {"role": "assistant", "content": "回复"}
+        ]
+        
+        result = self.normalizer.normalize_conversations(
+            conversations, 
+            strategy=NormalizationStrategy.MERGE
+        )
+        
+        normalized = result.normalized_conversations
+        merged_msg = normalized[0]
+        
+        # 验证额外字段被保留（来自第一条消息）
+        assert "timestamp" in merged_msg
+        assert "id" in merged_msg
+        assert merged_msg["timestamp"] == "2024-01-01"
+        assert merged_msg["id"] == "msg1"
+
+
+class TestConversationStructureAnalysis:
+    """测试会话结构分析功能"""
+    
+    def setup_method(self):
+        """测试前的设置"""
+        self.normalizer = ConversationNormalizer()
+    
+    def test_analyze_normal_structure(self):
+        """测试分析正常的会话结构"""
+        # 测试目标：验证正常会话结构的分析结果
+        conversations = [
+            {"role": "user", "content": "用户消息"},
+            {"role": "assistant", "content": "助手回复"},
+            {"role": "user", "content": "用户追问"},
+            {"role": "assistant", "content": "助手再回复"}
+        ]
+        
+        analysis = self.normalizer.analyze_conversation_structure(conversations)
+        
+        assert analysis["total_messages"] == 4
+        assert analysis["role_distribution"] == {"user": 2, "assistant": 2}
+        assert analysis["consecutive_groups"] == []
+        assert analysis["needs_normalization"] == False
+        assert len(analysis["issues"]) == 0
+    
+    def test_analyze_problematic_structure(self):
+        """测试分析有问题的会话结构"""
+        # 测试目标：验证问题会话结构的正确识别
+        conversations = [
+            {"role": "user", "content": "消息1"},
+            {"role": "user", "content": "消息2"},
+            {"role": "user", "content": "消息3"},
+            {"role": "assistant", "content": "回复1"},
+            {"role": "assistant", "content": "回复2"}
+        ]
+        
+        analysis = self.normalizer.analyze_conversation_structure(conversations)
+        
+        assert analysis["total_messages"] == 5
+        assert analysis["role_distribution"] == {"user": 3, "assistant": 2}
+        assert len(analysis["consecutive_groups"]) == 2
+        assert analysis["needs_normalization"] == True
+        assert len(analysis["issues"]) >= 2
+        
+        # 验证连续组信息
+        user_group = analysis["consecutive_groups"][0]
+        assert user_group["role"] == "user"
+        assert user_group["count"] == 3
+        assert user_group["start_index"] == 0
+        
+        assistant_group = analysis["consecutive_groups"][1]
+        assert assistant_group["role"] == "assistant"
+        assert assistant_group["count"] == 2
+        assert assistant_group["start_index"] == 3
+    
+    def test_analyze_empty_conversations(self):
+        """测试分析空会话"""
+        # 测试目标：验证空会话的分析结果
+        analysis = self.normalizer.analyze_conversation_structure([])
+        
+        assert analysis["total_messages"] == 0
+        assert analysis["role_distribution"] == {}
+        assert analysis["consecutive_groups"] == []
+        assert analysis["needs_normalization"] == False
+        assert analysis["issues"] == []
+
+
+class TestConversationValidation:
+    """测试会话验证功能"""
+    
+    def setup_method(self):
+        """测试前的设置"""
+        self.normalizer = ConversationNormalizer()
+    
+    def test_validate_perfect_conversation(self):
+        """测试验证完美的会话"""
+        # 测试目标：验证理想的user-assistant交替模式
+        conversations = [
+            {"role": "user", "content": "用户消息"},
+            {"role": "assistant", "content": "助手回复"},
+            {"role": "user", "content": "用户追问"},
+            {"role": "assistant", "content": "助手再回复"}
+        ]
+        
+        validation = self.normalizer.validate_normalized_conversation(conversations)
+        
+        assert validation["is_valid"] == True
+        assert len(validation["errors"]) == 0
+        assert len(validation["warnings"]) == 0
+        assert validation["structure_info"]["follows_user_assistant_pattern"] == True
+        assert validation["structure_info"]["starts_with_user"] == True
+        assert validation["structure_info"]["ends_with_assistant"] == True
+    
+    def test_validate_consecutive_roles(self):
+        """测试验证连续相同角色的问题"""
+        # 测试目标：验证连续相同角色被正确识别为错误
+        conversations = [
+            {"role": "user", "content": "用户消息1"},
+            {"role": "user", "content": "用户消息2"},
+            {"role": "assistant", "content": "助手回复"}
+        ]
+        
+        validation = self.normalizer.validate_normalized_conversation(conversations)
+        
+        assert validation["is_valid"] == False
+        assert len(validation["errors"]) == 1
+        assert "连续的 'user' 角色" in validation["errors"][0]
+    
+    def test_validate_wrong_pattern(self):
+        """测试验证错误的角色模式"""
+        # 测试目标：验证不符合user-assistant交替模式的会话
+        conversations = [
+            {"role": "assistant", "content": "助手先开始"},
+            {"role": "user", "content": "用户回应"},
+            {"role": "user", "content": "用户继续"}
+        ]
+        
+        validation = self.normalizer.validate_normalized_conversation(conversations)
+        
+        assert validation["structure_info"]["follows_user_assistant_pattern"] == False
+        assert validation["structure_info"]["starts_with_user"] == False
+        assert validation["structure_info"]["ends_with_assistant"] == False
+        assert len(validation["warnings"]) >= 2  # 至少有模式不匹配的警告
+
+
+class TestEdgeCases:
+    """测试边缘情况"""
+    
+    def setup_method(self):
+        """测试前的设置"""
+        self.normalizer = ConversationNormalizer()
+    
+    def test_empty_content_messages(self):
+        """测试空内容消息的处理"""
+        # 测试目标：验证空内容消息的正确处理
+        conversations = [
+            {"role": "user", "content": ""},
+            {"role": "user", "content": "有内容的消息"},
+            {"role": "assistant", "content": "回复"}
+        ]
+        
+        result = self.normalizer.normalize_conversations(
+            conversations, 
+            strategy=NormalizationStrategy.MERGE
+        )
+        
+        # 验证空内容也被正确合并
+        assert result.normalized_count == 2
+        normalized = result.normalized_conversations
+        assert "[消息 2]" in normalized[0]["content"]
+        assert "有内容的消息" in normalized[0]["content"]
+    
+    def test_missing_role_field(self):
+        """测试缺少role字段的消息"""
+        # 测试目标：验证缺少role字段时的处理
+        conversations = [
+            {"content": "没有role字段"},
+            {"content": "也没有role字段"},
+            {"role": "assistant", "content": "有role字段"}
+        ]
+        
+        result = self.normalizer.normalize_conversations(conversations)
+        
+        # 验证缺少role字段的消息被当作相同角色处理
+        assert result.normalized_count == 2
+        assert len(result.changes_made) == 1
+    
+    def test_unknown_role(self):
+        """测试未知角色的处理"""
+        # 测试目标：验证未知角色的正确处理
+        conversations = [
+            {"role": "system", "content": "系统消息1"},
+            {"role": "system", "content": "系统消息2"},
+            {"role": "user", "content": "用户消息"},
+            {"role": "assistant", "content": "助手回复"}
+        ]
+        
+        result = self.normalizer.normalize_conversations(
+            conversations,
+            strategy=NormalizationStrategy.MERGE
+        )
+        
+        # 验证系统消息被正确合并
+        assert result.normalized_count == 3
+        normalized = result.normalized_conversations
+        assert normalized[0]["role"] == "system"
+        assert "[消息 1]" in normalized[0]["content"]
+        assert "[消息 2]" in normalized[0]["content"]
+    
+    def test_all_same_role(self):
+        """测试全部相同角色的会话"""
+        # 测试目标：验证全部相同角色时的处理
+        conversations = [
+            {"role": "user", "content": "消息1"},
+            {"role": "user", "content": "消息2"},
+            {"role": "user", "content": "消息3"},
+            {"role": "user", "content": "消息4"}
+        ]
+        
+        # 测试合并策略
+        merge_result = self.normalizer.normalize_conversations(
+            conversations,
+            strategy=NormalizationStrategy.MERGE
+        )
+        assert merge_result.normalized_count == 1
+        assert len(merge_result.changes_made) == 1
+        
+        # 测试删除策略
+        delete_result = self.normalizer.normalize_conversations(
+            conversations,
+            strategy=NormalizationStrategy.DELETE
+        )
+        assert delete_result.normalized_count == 1
+        assert len(delete_result.changes_made) == 1
+        assert delete_result.changes_made[0]["deleted_count"] == 3
+
+
+class TestIntegrationScenarios:
+    """测试集成场景"""
+    
+    def setup_method(self):
+        """测试前的设置"""
+        self.normalizer = ConversationNormalizer()
+    
+    def test_real_world_scenario_1(self):
+        """测试真实场景1：用户多次提问后得到回复"""
+        # 测试目标：模拟用户连续提问的真实场景
+        conversations = [
+            {"role": "user", "content": "我想了解Python"},
+            {"role": "user", "content": "特别是关于数据结构的部分"},
+            {"role": "user", "content": "你能详细介绍一下列表吗？"},
+            {"role": "assistant", "content": "好的，我来详细介绍Python中的列表..."},
+            {"role": "user", "content": "那字典呢？"},
+            {"role": "assistant", "content": "字典是另一种重要的数据结构..."}
+        ]
+        
+        # 分析原始结构
+        analysis = self.normalizer.analyze_conversation_structure(conversations)
+        assert analysis["needs_normalization"] == True
+        assert len(analysis["consecutive_groups"]) == 1
+        
+        # 应用合并策略
+        result = self.normalizer.normalize_conversations(
+            conversations,
+            strategy=NormalizationStrategy.MERGE
+        )
+        
+        assert result.normalized_count == 4
+        normalized = result.normalized_conversations
+        
+        # 验证用户的三个问题被合并
+        assert normalized[0]["role"] == "user"
+        assert "Python" in normalized[0]["content"]
+        assert "数据结构" in normalized[0]["content"]
+        assert "列表" in normalized[0]["content"]
+        
+        # 验证后续对话保持不变
+        assert normalized[1]["content"] == "好的，我来详细介绍Python中的列表..."
+        assert normalized[2]["content"] == "那字典呢？"
+        assert normalized[3]["content"] == "字典是另一种重要的数据结构..."
+        
+        # 验证规整化后的会话
+        validation = self.normalizer.validate_normalized_conversation(normalized)
+        assert validation["is_valid"] == True
+    
+    def test_real_world_scenario_2(self):
+        """测试真实场景2：助手多次回复后用户提问"""
+        # 测试目标：模拟助手连续回复的真实场景
+        conversations = [
+            {"role": "user", "content": "请解释机器学习"},
+            {"role": "assistant", "content": "机器学习是人工智能的一个分支..."},
+            {"role": "assistant", "content": "它主要包括监督学习、无监督学习..."},
+            {"role": "assistant", "content": "常见的算法有线性回归、决策树..."},
+            {"role": "user", "content": "能举个具体例子吗？"},
+            {"role": "assistant", "content": "当然，比如预测房价..."}
+        ]
+        
+        # 应用删除策略
+        result = self.normalizer.normalize_conversations(
+            conversations,
+            strategy=NormalizationStrategy.DELETE
+        )
+        
+        assert result.normalized_count == 4
+        normalized = result.normalized_conversations
+        
+        # 验证只保留第一个助手回复
+        assert normalized[1]["role"] == "assistant"
+        assert normalized[1]["content"] == "机器学习是人工智能的一个分支..."
+        
+        # 验证变更记录
+        change = result.changes_made[0]
+        assert change["action"] == "delete"
+        assert change["deleted_count"] == 2
+        
+        # 验证规整化后的会话
+        validation = self.normalizer.validate_normalized_conversation(normalized)
+        assert validation["is_valid"] == True
+
+
+if __name__ == "__main__":
+    # 运行所有测试
+    pytest.main([__file__, "-v", "-s"])
+
+