29.5 除錯與最佳化模組

29.5.1 除錯模組概述

除錯模組是程式設計 Agent 的重要能力，它能夠自動識別、定位和修復程式碼中的錯誤。除錯涉及錯誤檢測、根因分析、修復建議等多個環節。

除錯流程

錯誤報告 ↓ 錯誤檢測 ↓ 錯誤定位 ↓ 根因分析 ↓ 修復建議 ↓ 修復驗證 ↓ 問題解決

## 29.5.2 错误检测

### 错误检测器

    python


    python

    class ErrorDetector:
        """错误检测器"""

        def __init__(self):
            self.detectors = {
                'syntax': SyntaxErrorDetector(),
                'runtime': RuntimeErrorDetector(),
                'logic': LogicErrorDetector(),
                'performance': PerformanceErrorDetector(),
                'security': SecurityErrorDetector()
            }

        async def detect(self, code: str,
                         execution_result: ExecutionResult = None) -> List[DetectedError]:
            """检测错误"""
            errors = []

            # 运行各种检测器
            for detector_name, detector in self.detectors.items():
                try:
                    detected = await detector.detect(code, execution_result)
                    errors.extend(detected)
                except Exception as e:
                    logger.error(f"Error in {detector_name} detector: {e}")

            # 去重
            errors = self._deduplicate_errors(errors)

            # 排序（按严重程度）
            errors.sort(key=lambda e: e.severity, reverse=True)

            return errors

        def _deduplicate_errors(self,
                                errors: List[DetectedError]) -> List[DetectedError]:
            """去重错误"""
            seen = set()
            unique_errors = []

            for error in errors:
                key = (error.type, error.location, error.message)
                if key not in seen:
                    seen.add(key)
                    unique_errors.append(error)

            return unique_errors

    class SyntaxErrorDetector:
        """语法错误检测器"""

        async def detect(self, code: str,
                         execution_result: ExecutionResult = None) -> List[DetectedError]:
            """检测语法错误"""
            errors = []

            try:
                # 尝试编译代码
                compile(code, '<string>', 'exec')
            except SyntaxError as e:
                error = DetectedError(
                    type='syntax',
                    severity='high',
                    location=f"Line {e.lineno}, Column {e.offset}",
                    message=e.msg,
                    code_snippet=self._get_code_snippet(code, e.lineno),
                    suggestion=self._get_syntax_suggestion(e)
                )
                errors.append(error)

            return errors

        def _get_code_snippet(self, code: str, line_no: int) -> str:
            """获取代码片段"""
            lines = code.split('\n')
            if 1 <= line_no <= len(lines):
                return lines[line_no - 1]
            return ""

        def _get_syntax_suggestion(self, error: SyntaxError) -> str:
            """获取语法错误建议"""
            suggestions = {
                'invalid syntax': "检查语法是否正确，确保括号、引号等匹配",
                'unexpected EOF': "检查代码是否完整，确保所有括号都已闭合",
                'unterminated string literal': "检查字符串是否正确闭合",
                'expected an indented block': "检查缩进是否正确",
                'unindent does not match any outer indentation level': "检查缩进层级是否一致"
            }

            return suggestions.get(error.msg, "请检查语法错误")

    class RuntimeErrorDetector:
        """运行时错误检测器"""

        async def detect(self, code: str,
                         execution_result: ExecutionResult = None) -> List[DetectedError]:
            """检测运行时错误"""
            errors = []

            if not execution_result or not execution_result.error:
                return errors

            # 分析错误信息
            error_info = self._parse_error(execution_result.error)

            error = DetectedError(
                type='runtime',
                severity='high',
                location=error_info.get('location', 'Unknown'),
                message=error_info.get('message', 'Unknown error'),
                code_snippet=error_info.get('code_snippet', ''),
                traceback=error_info.get('traceback', ''),
                suggestion=self._get_runtime_suggestion(error_info)
            )
            errors.append(error)

            return errors

        def _parse_error(self, error: str) -> Dict[str, str]:
            """解析错误信息"""
            error_info = {
                'message': error,
                'location': 'Unknown',
                'traceback': '',
                'code_snippet': ''
            }

            # 解析 traceback
            if 'Traceback' in error:
                lines = error.split('\n')
                for line in lines:
                    if 'File' in line and 'line' in line:
                        # 提取文件和行号
                        parts = line.split(',')
                        if len(parts) >= 2:
                            error_info['location'] = parts[1].strip()
                    elif line.strip() and not line.startswith(' ') and 'Error' in line:
                        error_info['message'] = line.strip()

            return error_info

        def _get_runtime_suggestion(self, error_info: Dict) -> str:
            """获取运行时错误建议"""
            message = error_info.get('message', '').lower()

            suggestions = {
                'nameerror': "检查变量名是否正确，确保在使用前已定义",
                'typeerror': "检查数据类型是否匹配，确保操作适用于该类型",
                'valueerror': "检查值是否在有效范围内",
                'indexerror': "检查索引是否在有效范围内",
                'keyerror': "检查字典键是否存在",
                'attributeerror': "检查对象是否具有该属性或方法",
                'zerodivisionerror': "检查除数是否为零"
            }

            for error_type, suggestion in suggestions.items():
                if error_type in message:
                    return suggestion

            return "请检查错误信息和代码逻辑"

    class LogicErrorDetector:
        """逻辑错误检测器"""

        def __init__(self, llm_client: LLMClient):
            self.llm_client = llm_client

        async def detect(self, code: str,
                         execution_result: ExecutionResult = None) -> List[DetectedError]:
            """检测逻辑错误"""
            errors = []

            # 使用 LLM 分析代码逻辑
            prompt = f"""
            分析以下代码中的潜在逻辑错误：

            {code}

            请识别：
            1. 无限循环
            2. 空指针引用
            3. 边界条件错误
            4. 逻辑矛盾
            5. 死代码

            以 JSON 格式返回检测到的错误列表。
            """

            response = await self.llm_client.complete(prompt)
            detected_errors = self._parse_logic_errors(response)

            for error in detected_errors:
                detected_error = DetectedError(
                    type='logic',
                    severity='medium',
                    location=error.get('location', 'Unknown'),
                    message=error.get('message', 'Logic error'),
                    code_snippet=error.get('code_snippet', ''),
                    suggestion=error.get('suggestion', 'Review the logic')
                )
                errors.append(detected_error)

            return errors

        def _parse_logic_errors(self, response: str) -> List[Dict]:
            """解析逻辑错误"""
            try:
                return json.loads(response)
            except json.JSONDecodeError:
                return []

    ```## 29.5.3 根因分析

```python
    ### 根因分析器

    class RootCauseAnalyzer:
    """根因分析器"""
    def __init__(self, llm_client: LLMClient):
    self.llm_client = llm_client
    async def analyze(self, error: DetectedError,
    code: str,
    context: Dict = None) -> RootCauseAnalysis:
    """分析根因"""
    analysis = RootCauseAnalysis(error=error)
    # 收集相关信息
    relevant_code = self._collect_relevant_code(error, code)
    # 分析根本原因
    analysis.root_cause = await self._identify_root_cause(
    error,
    relevant_code,
    context
    )
    # 分析影响范围
    analysis.impact = await self._analyze_impact(
    error,
    relevant_code
    )
    # 分析修复难度
    analysis.difficulty = await self._estimate_difficulty(
    error,
    analysis.root_cause
    )
    return analysis
    def _collect_relevant_code(self, error: DetectedError,
    code: str) -> str:
    """收集相关代码"""
    lines = code.split('\n')
    # 提取错误位置周围的代码
    if error.location:
    try:
    line_no = int(error.location.split(',')[0].split()[-1])
    start = max(0, line_no - 5)
    end = min(len(lines), line_no + 5)
    return '\n'.join(lines[start:end])
    except (ValueError, IndexError):
    pass
    return code
    async def _identify_root_cause(self, error: DetectedError,
    relevant_code: str,
    context: Dict) -> str:
    """识别根本原因"""
    prompt = f"""
    分析错误的根本原因：
    错误类型：{error.type}
    错误信息：{error.message}
    错误位置：{error.location}
    相关代码：
    {relevant_code}
    上下文：{context}
    请分析并说明：
    1. 错误的根本原因
    2. 为什么会发生这个错误
    3. 代码中的具体问题
    """
    return await self.llm_client.complete(prompt)
    async def _analyze_impact(self, error: DetectedError,
    relevant_code: str) -> ImpactAnalysis:
    """分析影响范围"""
    prompt = f"""
    分析错误的影响范围：
    错误：{error.message}
    相关代码：
    {relevant_code}
    请分析：
    1. 错误影响哪些功能
    2. 影响的严重程度
    3. 是否会影响其他代码
    4. 是否会影响数据完整性
    """
    response = await self.llm_client.complete(prompt)
    return self._parse_impact(response)
    async def _estimate_difficulty(self, error: DetectedError,
    root_cause: str) -> str:
    """估计修复难度"""
    prompt = f"""
    估计修复错误的难度：
    错误：{error.message}
    根本原因：{root_cause}
    请评估修复难度（简单/中等/困难）并说明理由。
    """
    return await self.llm_client.complete(prompt)

## 29.5.4 修复建议

### 修复建议生成器

    python

    ```python

    class FixSuggestionGenerator:
        """修復建議生成器"""

        def __init__(self, llm_client: LLMClient):
            self.llm_client = llm_client

        async def generate_suggestions(self,
                                       analysis: RootCauseAnalysis,
                                       code: str) -> List[FixSuggestion]:
            """生成修復建議"""
            suggestions = []

            # 生成多個修復方案

            prompt = f"""
            為以下錯誤生成修復建議：

            錯誤：{analysis.error.message}
            根本原因：{analysis.root_cause}
            原始程式碼：
            {code}

            請生成 2-3 個不同的修復方案，每個方案包括：
            1. 修復方法
            2. 修改後的程式碼
            3. 優缺點分析
            4. 適用場景

            以 JSON 格式返回修復方案列表。
            """

            response = await self.llm_client.complete(prompt)
            fix_suggestions = self._parse_fix_suggestions(response)

            for suggestion in fix_suggestions:
                fix_suggestion = FixSuggestion(
                    method=suggestion.get('method', ''),
                    fixed_code=suggestion.get('fixed_code', ''),
                    advantages=suggestion.get('advantages', []),
                    disadvantages=suggestion.get('disadvantages', []),
                    applicable_scenarios=suggestion.get('applicable_scenarios', ''),
                    confidence=suggestion.get('confidence', 0.5)
                )
                suggestions.append(fix_suggestion)

            # 按置信度排序

            suggestions.sort(key=lambda s: s.confidence, reverse=True)

            return suggestions

    ```## 29.5.5 性能优化

    ### 性能分析器

    class PerformanceAnalyzer:
    """性能分析器"""
    def __init__(self, llm_client: LLMClient):
    self.llm_client = llm_client
    async def analyze(self, code: str,
    execution_result: ExecutionResult = None) -> PerformanceAnalysis:
    """分析性能"""
    analysis = PerformanceAnalysis()
    # 分析时间复杂度
    analysis.time_complexity = await self._analyze_time_complexity(code)
    # 分析空间复杂度
    analysis.space_complexity = await self._analyze_space_complexity(code)
    # 识别性能瓶颈
    analysis.bottlenecks = await self._identify_bottlenecks(code)
    # 生成优化建议
    analysis.optimization_suggestions = await self._generate_optimization_suggestions(
    code,
    analysis
    )
    return analysis
    async def _analyze_time_complexity(self, code: str) -> str:
    """分析时间复杂度"""
    prompt = f"""
    分析以下代码的时间复杂度：
    {code}
    请分析并说明：
    1. 整体时间复杂度
    2. 关键部分的复杂度
    3. 影响复杂度的主要因素
    """
    return await self.llm_client.complete(prompt)
    async def _analyze_space_complexity(self, code: str) -> str:
    """分析空间复杂度"""
    prompt = f"""
    分析以下代码的空间复杂度：
    {code}
    请分析并说明：
    1. 整体空间复杂度
    2. 内存使用情况
    3. 可能的内存泄漏
    """
    return await self.llm_client.complete(prompt)
    async def _identify_bottlenecks(self, code: str) -> List[PerformanceBottleneck]:
    """识别性能瓶颈"""
    prompt = f"""
    识别以下代码中的性能瓶颈：
    {code}
    请识别：
    1. 循环嵌套
    2. 重复计算
    3. 不必要的内存分配
    4. 低效的数据结构使用
    5. I/O 操作
    以 JSON 格式返回瓶颈列表。
    """
    response = await self.llm_client.complete(prompt)
    return self._parse_bottlenecks(response)
    async def _generate_optimization_suggestions(self,
    code: str,
    analysis: PerformanceAnalysis) -> List[OptimizationSuggestion]:
    """生成优化建议"""
    suggestions = []
    for bottleneck in analysis.bottlenecks:
    prompt = f"""
    为以下性能瓶颈生成优化建议：
    瓶颈类型：{bottleneck.type}
    瓶颈位置：{bottleneck.location}
    瓶颈描述：{bottleneck.description}
    请生成具体的优化建议，包括：
    1. 优化方法
    2. 优化后的代码
    3. 预期性能提升
    """
    response = await self.llm_client.complete(prompt)
    suggestion = OptimizationSuggestion(
    bottleneck_type=bottleneck.type,
    method=self._extract_method(response),
    optimized_code=self._extract_code(response),
    expected_improvement=self._extract_improvement(response)
    )
    suggestions.append(suggestion)
    return suggestions

### 代码优化器

    python

    ```python

    class CodeOptimizer:
        """程式碼最佳化器"""

        def __init__(self, llm_client: LLMClient):
            self.llm_client = llm_client

        async def optimize(self, code: str,
                          suggestions: List[OptimizationSuggestion]) -> OptimizedCode:
            """最佳化程式碼"""
            optimized_code = code

            for suggestion in suggestions:

                # 應用最佳化

                optimized_code = await self._apply_optimization(
                    optimized_code,
                    suggestion
                )

            return OptimizedCode(
                original_code=code,
                optimized_code=optimized_code,
                applied_suggestions=suggestions
            )

        async def _apply_optimization(self, code: str,
                                     suggestion: OptimizationSuggestion) -> str:
            """應用單個最佳化"""
            prompt = f"""
            應用以下最佳化到程式碼：

            原始程式碼：
            {code}

            最佳化方法：{suggestion.method}
            最佳化後的程式碼示例：
            {suggestion.optimized_code}

            請返回應用最佳化後的完整程式碼。
            """

            return await self.llm_client.complete(prompt)

    ```## 29.5.6 调试工作流

    ### 调试工作流管理器

    class DebuggingWorkflow:
    """调试工作流"""
    def __init__(self, llm_client: LLMClient,
    tool_manager: ToolManager):
    self.llm_client = llm_client
    self.tool_manager = tool_manager
    # 初始化组件
    self.error_detector = ErrorDetector()
    self.root_cause_analyzer = RootCauseAnalyzer(llm_client)
    self.fix_suggestion_generator = FixSuggestionGenerator(llm_client)
    self.performance_analyzer = PerformanceAnalyzer(llm_client)
    self.code_optimizer = CodeOptimizer(llm_client)
    async def debug(self, code: str,
    execution_result: ExecutionResult = None) -> DebuggingResult:
    """执行调试工作流"""
    result = DebuggingResult(original_code=code)
    # 1. 检测错误
    result.errors = await self.error_detector.detect(code, execution_result)
    if not result.errors:
    result.status = "no_errors"
    return result
    # 2. 分析每个错误
    for error in result.errors:
    # 根因分析
    analysis = await self.root_cause_analyzer.analyze(error, code)
    result.root_cause_analyses.append(analysis)
    # 生成修复建议
    suggestions = await self.fix_suggestion_generator.generate_suggestions(
    analysis,
    code
    )
    result.fix_suggestions.extend(suggestions)
    # 3. 性能分析
    result.performance_analysis = await self.performance_analyzer.analyze(code)
    # 4. 生成优化建议
    if result.performance_analysis.optimization_suggestions:
    result.optimization_suggestions = result.performance_analysis.optimization_suggestions
    result.status = "completed"
    return result
    async def apply_fix(self, code: str,
    suggestion: FixSuggestion) -> FixedCode:
    """应用修复"""
    fixed_code = FixedCode(
    original_code=code,
    fixed_code=suggestion.fixed_code,
    applied_suggestion=suggestion
    )
    # 验证修复
    validation_result = await self._validate_fix(fixed_code)
    fixed_code.validation_result = validation_result
    return fixed_code
    async def _validate_fix(self, fixed_code: FixedCode) -> ValidationResult:
    """验证修复"""
    # 检查语法
    try:
    compile(fixed_code.fixed_code, '<string>', 'exec')
    syntax_valid = True
    except SyntaxError:
    syntax_valid = False
    # 运行测试（如果有）
    test_result = await self._run_tests(fixed_code.fixed_code)
    return ValidationResult(
    syntax_valid=syntax_valid,
    tests_passed=test_result.passed,
    test_output=test_result.output
    )
    async def _run_tests(self, code: str) -> TestResult:
    """运行测试"""
    # 简化实现：执行代码并检查错误
    try:
    exec(code)
    return TestResult(passed=True, output="Execution successful")
    except Exception as e:
    return TestResult(passed=False, output=str(e))