amd-gaia 0.15.0__py3-none-any.whl → 0.15.2__py3-none-any.whl

gaia/agents/code/orchestration/steps/error_handler.py

@@ -1,314 +1,314 @@
-# Copyright(C) 2025-2026 Advanced Micro Devices, Inc. All rights reserved.
-# SPDX-License-Identifier: MIT
-"""
-Three-tier error recovery for orchestration workflows.
-
-Provides intelligent error handling with:
-- RETRY: For transient errors (network, resources)
-- FIX_AND_RETRY: For known fixable errors (missing deps, compilation)
-- ESCALATE: For complex errors requiring LLM intervention
-"""
-
-import logging
-import re
-import time
-from dataclasses import dataclass
-from enum import Enum, auto
-from typing import Any, Callable, Dict, List, Optional, Tuple
-
-from .base import ErrorCategory
-
-logger = logging.getLogger(__name__)
-
-
-class RecoveryAction(Enum):
-    """Actions the error handler can take."""
-
-    RETRY = auto()  # Wait and retry same operation
-    FIX_AND_RETRY = auto()  # Execute fix, then retry
-    ESCALATE = auto()  # Use LLM to diagnose and fix
-    ABORT = auto()  # Give up, unrecoverable
-
-
-@dataclass
-class ErrorPattern:
-    """Pattern for matching and handling specific errors."""
-
-    pattern: str  # Regex pattern to match
-    category: ErrorCategory
-    action: RecoveryAction
-    fix_command: Optional[str] = None  # Command to run for FIX_AND_RETRY
-    max_retries: int = 2
-    delay_seconds: float = 1.0
-
-
-# Error patterns based on existing cli_tools.py ERROR_PATTERNS
-ERROR_PATTERNS: List[ErrorPattern] = [
-    # Network errors - retry
-    ErrorPattern(
-        pattern=r"ECONNREFUSED|ETIMEDOUT|ENOTFOUND|network\s+error",
-        category=ErrorCategory.NETWORK,
-        action=RecoveryAction.RETRY,
-        max_retries=3,
-        delay_seconds=2.0,
-    ),
-    # Resource errors - retry with delay
-    ErrorPattern(
-        pattern=r"ENOSPC|ENOMEM|out\s+of\s+memory",
-        category=ErrorCategory.RESOURCE,
-        action=RecoveryAction.RETRY,
-        max_retries=2,
-        delay_seconds=5.0,
-    ),
-    # Missing dependencies - fix and retry
-    ErrorPattern(
-        pattern=r"Cannot find module '([^']+)'|Module not found.*'([^']+)'",
-        category=ErrorCategory.DEPENDENCY,
-        action=RecoveryAction.FIX_AND_RETRY,
-        fix_command="npm install {module}",
-        max_retries=2,
-    ),
-    ErrorPattern(
-        pattern=r"ModuleNotFoundError:\s+No module named '([^']+)'",
-        category=ErrorCategory.DEPENDENCY,
-        action=RecoveryAction.FIX_AND_RETRY,
-        fix_command="uv pip install {module}",
-        max_retries=2,
-    ),
-    # TypeScript compilation - escalate to LLM
-    ErrorPattern(
-        pattern=r"TS\d+:|error TS\d+|Type '.*' is not assignable",
-        category=ErrorCategory.COMPILATION,
-        action=RecoveryAction.ESCALATE,
-        max_retries=1,
-    ),
-    # Prisma errors - fix and retry
-    ErrorPattern(
-        pattern=r"prisma generate|@prisma/client.*not.*generated",
-        category=ErrorCategory.DEPENDENCY,
-        action=RecoveryAction.FIX_AND_RETRY,
-        fix_command="npx prisma generate",
-        max_retries=2,
-    ),
-    # Syntax errors - escalate to LLM
-    ErrorPattern(
-        pattern=r"SyntaxError:|Unexpected token|Parse error",
-        category=ErrorCategory.SYNTAX,
-        action=RecoveryAction.ESCALATE,
-        max_retries=1,
-    ),
-    # Lint errors - escalate to LLM for fix
-    ErrorPattern(
-        pattern=r"eslint.*error|warning.*react-hooks|unused variable",
-        category=ErrorCategory.VALIDATION,
-        action=RecoveryAction.ESCALATE,
-        max_retries=1,
-    ),
-    # CSS content type errors (Issue #1002) - escalate to LLM
-    # Catches TypeScript/JavaScript code in CSS files
-    ErrorPattern(
-        pattern=r"CSS file contains.*TypeScript|CRITICAL.*CSS file|globals\.css contains",
-        category=ErrorCategory.VALIDATION,
-        action=RecoveryAction.ESCALATE,
-        max_retries=2,  # Allow retries - LLM can regenerate correct CSS
-    ),
-]
-
-
-class ErrorHandler:
-    """Handles errors with three-tier recovery strategy."""
-
-    def __init__(
-        self,
-        command_executor: Optional[Callable[[str], Tuple[int, str]]] = None,
-        llm_fixer: Optional[Callable[[str, str], Optional[str]]] = None,
-    ):
-        """Initialize error handler.
-
-        Args:
-            command_executor: Function to run shell commands (returns exit_code, output)
-            llm_fixer: Function to fix code using LLM (takes error, code, returns fixed code)
-        """
-        self.command_executor = command_executor
-        self.llm_fixer = llm_fixer
-        self.retry_counts: Dict[str, int] = {}
-
-    def categorize_error(self, error_text: str) -> Tuple[ErrorCategory, ErrorPattern]:
-        """Categorize an error and find matching pattern.
-
-        Args:
-            error_text: The error message to categorize
-
-        Returns:
-            Tuple of (ErrorCategory, matching ErrorPattern or default)
-        """
-        for pattern in ERROR_PATTERNS:
-            if re.search(pattern.pattern, error_text, re.IGNORECASE):
-                return pattern.category, pattern
-
-        # Default pattern for unknown errors
-        return ErrorCategory.UNKNOWN, ErrorPattern(
-            pattern=".*",
-            category=ErrorCategory.UNKNOWN,
-            action=RecoveryAction.ESCALATE,
-            max_retries=1,
-        )
-
-    def handle_error(
-        self,
-        step_name: str,
-        error_text: str,
-        context: Optional[Dict[str, Any]] = None,
-    ) -> Tuple[RecoveryAction, Optional[str]]:
-        """Handle an error with appropriate recovery action.
-
-        Args:
-            step_name: Name of the step that failed
-            error_text: The error message
-            context: Optional context (e.g., code being executed)
-
-        Returns:
-            Tuple of (action to take, optional fix result/message)
-        """
-        category, pattern = self.categorize_error(error_text)
-        retry_key = f"{step_name}:{category.name}"
-
-        # Check retry count
-        current_retries = self.retry_counts.get(retry_key, 0)
-        if current_retries >= pattern.max_retries:
-            logger.warning(
-                f"Max retries ({pattern.max_retries}) exceeded for {step_name}"
-            )
-            return RecoveryAction.ABORT, f"Max retries exceeded: {error_text}"
-
-        self.retry_counts[retry_key] = current_retries + 1
-
-        logger.info(
-            f"Error recovery: {category.name} -> {pattern.action.name} "
-            f"(attempt {current_retries + 1}/{pattern.max_retries})"
-        )
-
-        if pattern.action == RecoveryAction.RETRY:
-            return self._handle_retry(pattern)
-
-        if pattern.action == RecoveryAction.FIX_AND_RETRY:
-            return self._handle_fix_and_retry(pattern, error_text)
-
-        if pattern.action == RecoveryAction.ESCALATE:
-            return self._handle_escalate(error_text, context)
-
-        return RecoveryAction.ABORT, error_text
-
-    def _handle_retry(
-        self, pattern: ErrorPattern
-    ) -> Tuple[RecoveryAction, Optional[str]]:
-        """Handle retry action with delay."""
-        if pattern.delay_seconds > 0:
-            logger.info(f"Waiting {pattern.delay_seconds}s before retry...")
-            time.sleep(pattern.delay_seconds)
-        return RecoveryAction.RETRY, None
-
-    def _handle_fix_and_retry(
-        self, pattern: ErrorPattern, error_text: str
-    ) -> Tuple[RecoveryAction, Optional[str]]:
-        """Handle fix-and-retry action."""
-        if not pattern.fix_command or not self.command_executor:
-            return RecoveryAction.ESCALATE, "No fix command available"
-
-        # Extract module name from error if applicable
-        fix_cmd = pattern.fix_command
-        match = re.search(pattern.pattern, error_text, re.IGNORECASE)
-        if match and match.groups():
-            # Use first captured group as module name
-            module = (
-                match.group(1) or match.group(2)
-                if len(match.groups()) > 1
-                else match.group(1)
-            )
-            if module:
-                fix_cmd = fix_cmd.format(module=module)
-
-        logger.info(f"Executing fix command: {fix_cmd}")
-        exit_code, output = self.command_executor(fix_cmd)
-
-        if exit_code == 0:
-            return RecoveryAction.RETRY, f"Fix applied: {fix_cmd}"
-        return RecoveryAction.ESCALATE, f"Fix failed: {output}"
-
-    def _handle_escalate(
-        self, error_text: str, context: Optional[Dict[str, Any]] = None
-    ) -> Tuple[RecoveryAction, Optional[str]]:
-        """Handle escalation to LLM.
-
-        For TypeScript errors, parses file path from error and reads content.
-        """
-        from pathlib import Path
-
-        if not self.llm_fixer:
-            return RecoveryAction.ABORT, "No LLM fixer available"
-
-        code = context.get("code", "") if context else ""
-        project_dir = context.get("project_dir", "") if context else ""
-        error_file_path = None
-
-        # If no code provided, try to extract file path from TypeScript error
-        # Format: filename.ts(line,col): error TSxxxx: message
-        if not code and project_dir:
-            ts_error_match = re.search(
-                r"^([^\s(]+\.tsx?)\(\d+,\d+\):", error_text, re.MULTILINE
-            )
-            if ts_error_match:
-                error_file_path = ts_error_match.group(1)
-                full_path = Path(project_dir) / error_file_path
-                if full_path.exists():
-                    try:
-                        code = full_path.read_text(encoding="utf-8")
-                        logger.info(f"Read file for LLM fix: {error_file_path}")
-                    except Exception as e:
-                        logger.warning(f"Could not read {error_file_path}: {e}")
-
-        fixed_code = self.llm_fixer(error_text, code)
-
-        if fixed_code:
-            # If we read a file, write the fix back
-            if error_file_path and project_dir:
-                full_path = Path(project_dir) / error_file_path
-                try:
-                    full_path.write_text(fixed_code, encoding="utf-8")
-                    logger.info(f"Wrote LLM fix to: {error_file_path}")
-                except Exception as e:
-                    logger.warning(f"Could not write fix to {error_file_path}: {e}")
-
-            return RecoveryAction.RETRY, fixed_code
-        return RecoveryAction.ABORT, "LLM could not fix the error"
-
-    def reset_retry_count(self, step_name: str) -> None:
-        """Reset retry count for a step after success."""
-        keys_to_remove = [k for k in self.retry_counts if k.startswith(f"{step_name}:")]
-        for key in keys_to_remove:
-            del self.retry_counts[key]
-
-    def get_recovery_suggestion(self, error_text: str) -> str:
-        """Get a human-readable recovery suggestion for an error.
-
-        Args:
-            error_text: The error message
-
-        Returns:
-            Suggestion string for how to fix the error
-        """
-        category, pattern = self.categorize_error(error_text)
-
-        suggestions = {
-            ErrorCategory.NETWORK: "Check network connection and retry",
-            ErrorCategory.RESOURCE: "Free up system resources (disk/memory) and retry",
-            ErrorCategory.DEPENDENCY: f"Install missing dependency: {pattern.fix_command or 'check error message'}",
-            ErrorCategory.COMPILATION: "Fix TypeScript/compilation errors in the code",
-            ErrorCategory.SYNTAX: "Fix syntax errors in the code",
-            ErrorCategory.RUNTIME: "Debug runtime error and fix logic",
-            ErrorCategory.VALIDATION: "Fix linting/validation warnings",
-            ErrorCategory.CONFIGURATION: "Check configuration files for errors",
-            ErrorCategory.UNKNOWN: "Review error message and investigate",
-        }
-
-        return suggestions.get(category, "Unknown error - investigate")
+# Copyright(C) 2025-2026 Advanced Micro Devices, Inc. All rights reserved.
+# SPDX-License-Identifier: MIT
+"""
+Three-tier error recovery for orchestration workflows.
+
+Provides intelligent error handling with:
+- RETRY: For transient errors (network, resources)
+- FIX_AND_RETRY: For known fixable errors (missing deps, compilation)
+- ESCALATE: For complex errors requiring LLM intervention
+"""
+
+import logging
+import re
+import time
+from dataclasses import dataclass
+from enum import Enum, auto
+from typing import Any, Callable, Dict, List, Optional, Tuple
+
+from .base import ErrorCategory
+
+logger = logging.getLogger(__name__)
+
+
+class RecoveryAction(Enum):
+    """Actions the error handler can take."""
+
+    RETRY = auto()  # Wait and retry same operation
+    FIX_AND_RETRY = auto()  # Execute fix, then retry
+    ESCALATE = auto()  # Use LLM to diagnose and fix
+    ABORT = auto()  # Give up, unrecoverable
+
+
+@dataclass
+class ErrorPattern:
+    """Pattern for matching and handling specific errors."""
+
+    pattern: str  # Regex pattern to match
+    category: ErrorCategory
+    action: RecoveryAction
+    fix_command: Optional[str] = None  # Command to run for FIX_AND_RETRY
+    max_retries: int = 2
+    delay_seconds: float = 1.0
+
+
+# Error patterns based on existing cli_tools.py ERROR_PATTERNS
+ERROR_PATTERNS: List[ErrorPattern] = [
+    # Network errors - retry
+    ErrorPattern(
+        pattern=r"ECONNREFUSED|ETIMEDOUT|ENOTFOUND|network\s+error",
+        category=ErrorCategory.NETWORK,
+        action=RecoveryAction.RETRY,
+        max_retries=3,
+        delay_seconds=2.0,
+    ),
+    # Resource errors - retry with delay
+    ErrorPattern(
+        pattern=r"ENOSPC|ENOMEM|out\s+of\s+memory",
+        category=ErrorCategory.RESOURCE,
+        action=RecoveryAction.RETRY,
+        max_retries=2,
+        delay_seconds=5.0,
+    ),
+    # Missing dependencies - fix and retry
+    ErrorPattern(
+        pattern=r"Cannot find module '([^']+)'|Module not found.*'([^']+)'",
+        category=ErrorCategory.DEPENDENCY,
+        action=RecoveryAction.FIX_AND_RETRY,
+        fix_command="npm install {module}",
+        max_retries=2,
+    ),
+    ErrorPattern(
+        pattern=r"ModuleNotFoundError:\s+No module named '([^']+)'",
+        category=ErrorCategory.DEPENDENCY,
+        action=RecoveryAction.FIX_AND_RETRY,
+        fix_command="uv pip install {module}",
+        max_retries=2,
+    ),
+    # TypeScript compilation - escalate to LLM
+    ErrorPattern(
+        pattern=r"TS\d+:|error TS\d+|Type '.*' is not assignable",
+        category=ErrorCategory.COMPILATION,
+        action=RecoveryAction.ESCALATE,
+        max_retries=1,
+    ),
+    # Prisma errors - fix and retry
+    ErrorPattern(
+        pattern=r"prisma generate|@prisma/client.*not.*generated",
+        category=ErrorCategory.DEPENDENCY,
+        action=RecoveryAction.FIX_AND_RETRY,
+        fix_command="npx prisma generate",
+        max_retries=2,
+    ),
+    # Syntax errors - escalate to LLM
+    ErrorPattern(
+        pattern=r"SyntaxError:|Unexpected token|Parse error",
+        category=ErrorCategory.SYNTAX,
+        action=RecoveryAction.ESCALATE,
+        max_retries=1,
+    ),
+    # Lint errors - escalate to LLM for fix
+    ErrorPattern(
+        pattern=r"eslint.*error|warning.*react-hooks|unused variable",
+        category=ErrorCategory.VALIDATION,
+        action=RecoveryAction.ESCALATE,
+        max_retries=1,
+    ),
+    # CSS content type errors (Issue #1002) - escalate to LLM
+    # Catches TypeScript/JavaScript code in CSS files
+    ErrorPattern(
+        pattern=r"CSS file contains.*TypeScript|CRITICAL.*CSS file|globals\.css contains",
+        category=ErrorCategory.VALIDATION,
+        action=RecoveryAction.ESCALATE,
+        max_retries=2,  # Allow retries - LLM can regenerate correct CSS
+    ),
+]
+
+
+class ErrorHandler:
+    """Handles errors with three-tier recovery strategy."""
+
+    def __init__(
+        self,
+        command_executor: Optional[Callable[[str], Tuple[int, str]]] = None,
+        llm_fixer: Optional[Callable[[str, str], Optional[str]]] = None,
+    ):
+        """Initialize error handler.
+
+        Args:
+            command_executor: Function to run shell commands (returns exit_code, output)
+            llm_fixer: Function to fix code using LLM (takes error, code, returns fixed code)
+        """
+        self.command_executor = command_executor
+        self.llm_fixer = llm_fixer
+        self.retry_counts: Dict[str, int] = {}
+
+    def categorize_error(self, error_text: str) -> Tuple[ErrorCategory, ErrorPattern]:
+        """Categorize an error and find matching pattern.
+
+        Args:
+            error_text: The error message to categorize
+
+        Returns:
+            Tuple of (ErrorCategory, matching ErrorPattern or default)
+        """
+        for pattern in ERROR_PATTERNS:
+            if re.search(pattern.pattern, error_text, re.IGNORECASE):
+                return pattern.category, pattern
+
+        # Default pattern for unknown errors
+        return ErrorCategory.UNKNOWN, ErrorPattern(
+            pattern=".*",
+            category=ErrorCategory.UNKNOWN,
+            action=RecoveryAction.ESCALATE,
+            max_retries=1,
+        )
+
+    def handle_error(
+        self,
+        step_name: str,
+        error_text: str,
+        context: Optional[Dict[str, Any]] = None,
+    ) -> Tuple[RecoveryAction, Optional[str]]:
+        """Handle an error with appropriate recovery action.
+
+        Args:
+            step_name: Name of the step that failed
+            error_text: The error message
+            context: Optional context (e.g., code being executed)
+
+        Returns:
+            Tuple of (action to take, optional fix result/message)
+        """
+        category, pattern = self.categorize_error(error_text)
+        retry_key = f"{step_name}:{category.name}"
+
+        # Check retry count
+        current_retries = self.retry_counts.get(retry_key, 0)
+        if current_retries >= pattern.max_retries:
+            logger.warning(
+                f"Max retries ({pattern.max_retries}) exceeded for {step_name}"
+            )
+            return RecoveryAction.ABORT, f"Max retries exceeded: {error_text}"
+
+        self.retry_counts[retry_key] = current_retries + 1
+
+        logger.info(
+            f"Error recovery: {category.name} -> {pattern.action.name} "
+            f"(attempt {current_retries + 1}/{pattern.max_retries})"
+        )
+
+        if pattern.action == RecoveryAction.RETRY:
+            return self._handle_retry(pattern)
+
+        if pattern.action == RecoveryAction.FIX_AND_RETRY:
+            return self._handle_fix_and_retry(pattern, error_text)
+
+        if pattern.action == RecoveryAction.ESCALATE:
+            return self._handle_escalate(error_text, context)
+
+        return RecoveryAction.ABORT, error_text
+
+    def _handle_retry(
+        self, pattern: ErrorPattern
+    ) -> Tuple[RecoveryAction, Optional[str]]:
+        """Handle retry action with delay."""
+        if pattern.delay_seconds > 0:
+            logger.info(f"Waiting {pattern.delay_seconds}s before retry...")
+            time.sleep(pattern.delay_seconds)
+        return RecoveryAction.RETRY, None
+
+    def _handle_fix_and_retry(
+        self, pattern: ErrorPattern, error_text: str
+    ) -> Tuple[RecoveryAction, Optional[str]]:
+        """Handle fix-and-retry action."""
+        if not pattern.fix_command or not self.command_executor:
+            return RecoveryAction.ESCALATE, "No fix command available"
+
+        # Extract module name from error if applicable
+        fix_cmd = pattern.fix_command
+        match = re.search(pattern.pattern, error_text, re.IGNORECASE)
+        if match and match.groups():
+            # Use first captured group as module name
+            module = (
+                match.group(1) or match.group(2)
+                if len(match.groups()) > 1
+                else match.group(1)
+            )
+            if module:
+                fix_cmd = fix_cmd.format(module=module)
+
+        logger.info(f"Executing fix command: {fix_cmd}")
+        exit_code, output = self.command_executor(fix_cmd)
+
+        if exit_code == 0:
+            return RecoveryAction.RETRY, f"Fix applied: {fix_cmd}"
+        return RecoveryAction.ESCALATE, f"Fix failed: {output}"
+
+    def _handle_escalate(
+        self, error_text: str, context: Optional[Dict[str, Any]] = None
+    ) -> Tuple[RecoveryAction, Optional[str]]:
+        """Handle escalation to LLM.
+
+        For TypeScript errors, parses file path from error and reads content.
+        """
+        from pathlib import Path
+
+        if not self.llm_fixer:
+            return RecoveryAction.ABORT, "No LLM fixer available"
+
+        code = context.get("code", "") if context else ""
+        project_dir = context.get("project_dir", "") if context else ""
+        error_file_path = None
+
+        # If no code provided, try to extract file path from TypeScript error
+        # Format: filename.ts(line,col): error TSxxxx: message
+        if not code and project_dir:
+            ts_error_match = re.search(
+                r"^([^\s(]+\.tsx?)\(\d+,\d+\):", error_text, re.MULTILINE
+            )
+            if ts_error_match:
+                error_file_path = ts_error_match.group(1)
+                full_path = Path(project_dir) / error_file_path
+                if full_path.exists():
+                    try:
+                        code = full_path.read_text(encoding="utf-8")
+                        logger.info(f"Read file for LLM fix: {error_file_path}")
+                    except Exception as e:
+                        logger.warning(f"Could not read {error_file_path}: {e}")
+
+        fixed_code = self.llm_fixer(error_text, code)
+
+        if fixed_code:
+            # If we read a file, write the fix back
+            if error_file_path and project_dir:
+                full_path = Path(project_dir) / error_file_path
+                try:
+                    full_path.write_text(fixed_code, encoding="utf-8")
+                    logger.info(f"Wrote LLM fix to: {error_file_path}")
+                except Exception as e:
+                    logger.warning(f"Could not write fix to {error_file_path}: {e}")
+
+            return RecoveryAction.RETRY, fixed_code
+        return RecoveryAction.ABORT, "LLM could not fix the error"
+
+    def reset_retry_count(self, step_name: str) -> None:
+        """Reset retry count for a step after success."""
+        keys_to_remove = [k for k in self.retry_counts if k.startswith(f"{step_name}:")]
+        for key in keys_to_remove:
+            del self.retry_counts[key]
+
+    def get_recovery_suggestion(self, error_text: str) -> str:
+        """Get a human-readable recovery suggestion for an error.
+
+        Args:
+            error_text: The error message
+
+        Returns:
+            Suggestion string for how to fix the error
+        """
+        category, pattern = self.categorize_error(error_text)
+
+        suggestions = {
+            ErrorCategory.NETWORK: "Check network connection and retry",
+            ErrorCategory.RESOURCE: "Free up system resources (disk/memory) and retry",
+            ErrorCategory.DEPENDENCY: f"Install missing dependency: {pattern.fix_command or 'check error message'}",
+            ErrorCategory.COMPILATION: "Fix TypeScript/compilation errors in the code",
+            ErrorCategory.SYNTAX: "Fix syntax errors in the code",
+            ErrorCategory.RUNTIME: "Debug runtime error and fix logic",
+            ErrorCategory.VALIDATION: "Fix linting/validation warnings",
+            ErrorCategory.CONFIGURATION: "Check configuration files for errors",
+            ErrorCategory.UNKNOWN: "Review error message and investigate",
+        }
+
+        return suggestions.get(category, "Unknown error - investigate")