tunacode-cli 0.0.48__py3-none-any.whl → 0.0.49__py3-none-any.whl

tunacode/core/recursive/executor.py

@@ -0,0 +1,470 @@
+"""Module: tunacode.core.recursive.executor
+
+Main RecursiveTaskExecutor class for orchestrating recursive task decomposition and execution.
+"""
+
+import logging
+from dataclasses import dataclass, field
+from typing import Any, Dict, List, Optional, Tuple
+from uuid import uuid4
+
+from pydantic import BaseModel
+
+from tunacode.core.state import StateManager
+
+from .aggregator import AggregationStrategy, ResultAggregator, TaskResult
+from .budget import BudgetManager
+from .decomposer import TaskDecomposer
+from .hierarchy import TaskHierarchy
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class TaskNode:
+    """Represents a single task in the recursive execution tree."""
+
+    id: str = field(default_factory=lambda: str(uuid4()))
+    parent_id: Optional[str] = None
+    title: str = ""
+    description: str = ""
+    complexity_score: float = 0.0
+    iteration_budget: int = 10
+    subtasks: List["TaskNode"] = field(default_factory=list)
+    status: str = "pending"  # pending, in_progress, completed, failed
+    result: Optional[object] = None
+    error: Optional[str] = None
+    context: Dict[str, Any] = field(default_factory=dict)
+    depth: int = 0
+
+
+class TaskComplexityResult(BaseModel):
+    """Result of task complexity analysis."""
+
+    is_complex: bool
+    complexity_score: float
+    reasoning: str
+    suggested_subtasks: List[str] = []
+
+
+class RecursiveTaskExecutor:
+    """Orchestrates recursive task decomposition and execution."""
+
+    def __init__(
+        self,
+        state_manager: StateManager,
+        max_depth: int = 5,
+        min_complexity_threshold: float = 0.7,
+        default_iteration_budget: int = 10,
+    ):
+        """Initialize the RecursiveTaskExecutor.
+
+        Args:
+            state_manager: The StateManager instance for accessing agents and state
+            max_depth: Maximum recursion depth allowed
+            min_complexity_threshold: Minimum complexity score to trigger decomposition
+            default_iteration_budget: Default iteration budget for tasks
+        """
+        self.state_manager = state_manager
+        self.max_depth = max_depth
+        self.min_complexity_threshold = min_complexity_threshold
+        self.default_iteration_budget = default_iteration_budget
+        self._task_hierarchy: Dict[str, TaskNode] = {}
+        self._execution_stack: List[str] = []
+
+        # Initialize decomposer and hierarchy manager
+        self.decomposer = TaskDecomposer(state_manager)
+        self.hierarchy = TaskHierarchy()
+        self.budget_manager = BudgetManager(
+            total_budget=default_iteration_budget * 10,  # Scale based on default
+            min_task_budget=2,
+        )
+        self.aggregator = ResultAggregator(state_manager)
+
+    async def execute_task(
+        self,
+        request: str,
+        parent_task_id: Optional[str] = None,
+        depth: int = 0,
+        inherited_context: Optional[Dict[str, Any]] = None,
+    ) -> Tuple[bool, Any, Optional[str]]:
+        """Execute a task, potentially decomposing it into subtasks.
+
+        Args:
+            request: The task request/description
+            parent_task_id: ID of parent task if this is a subtask
+            depth: Current recursion depth
+            inherited_context: Context inherited from parent task
+
+        Returns:
+            Tuple of (success, result, error_message)
+        """
+        # Check recursion depth
+        if depth >= self.max_depth:
+            logger.warning(f"Max recursion depth {self.max_depth} reached")
+            return False, None, f"Maximum recursion depth ({self.max_depth}) exceeded"
+
+        # Create task node
+        task_node = TaskNode(
+            title=request[:100],  # First 100 chars as title
+            description=request,
+            parent_id=parent_task_id,
+            depth=depth,
+            context=inherited_context or {},
+        )
+
+        self._task_hierarchy[task_node.id] = task_node
+        self._execution_stack.append(task_node.id)
+
+        try:
+            # Analyze task complexity
+            complexity_result = await self._analyze_task_complexity(request)
+            task_node.complexity_score = complexity_result.complexity_score
+
+            # Decide whether to decompose or execute directly
+            if (
+                complexity_result.is_complex
+                and complexity_result.complexity_score >= self.min_complexity_threshold
+                and depth < self.max_depth - 1
+            ):
+                # Decompose into subtasks
+                if self.state_manager.session.show_thoughts:
+                    logger.info(
+                        f"Decomposing complex task (score: {complexity_result.complexity_score:.2f})"
+                    )
+                return await self._execute_with_decomposition(task_node, complexity_result)
+            else:
+                # Execute directly
+                if self.state_manager.session.show_thoughts:
+                    logger.info(
+                        f"Executing task directly (score: {complexity_result.complexity_score:.2f})"
+                    )
+                return await self._execute_directly(task_node)
+
+        except Exception as e:
+            logger.error(f"Error executing task {task_node.id}: {str(e)}")
+            task_node.status = "failed"
+            task_node.error = str(e)
+            return False, None, str(e)
+        finally:
+            self._execution_stack.pop()
+
+    async def _analyze_task_complexity(self, request: str) -> TaskComplexityResult:
+        """Analyze task complexity using the main agent.
+
+        Args:
+            request: The task description
+
+        Returns:
+            TaskComplexityResult with analysis
+        """
+        # Get the main agent from state manager
+        agent = self.state_manager.session.agents.get("main")
+        if not agent:
+            # Simple heuristic if agent not available
+            word_count = len(request.split())
+            has_multiple_parts = any(
+                word in request.lower() for word in ["and", "then", "also", "plus"]
+            )
+
+            complexity_score = min(1.0, (word_count / 50) + (0.3 if has_multiple_parts else 0))
+            is_complex = complexity_score >= self.min_complexity_threshold
+
+            return TaskComplexityResult(
+                is_complex=is_complex,
+                complexity_score=complexity_score,
+                reasoning="Heuristic analysis based on request length and structure",
+            )
+
+        # Use agent to analyze complexity
+        complexity_prompt = f"""Analyze the complexity of this task and determine if it should be broken down into subtasks.
+
+Task: {request}
+
+Provide:
+1. A complexity score from 0.0 to 1.0 (0 = trivial, 1 = extremely complex)
+2. Whether this task should be decomposed (true/false)
+3. Brief reasoning for your assessment
+4. If complex, suggest 2-5 subtasks
+
+Consider factors like:
+- Number of distinct operations required
+- Dependencies between operations
+- Technical complexity
+- Estimated time/effort needed
+
+Respond in JSON format:
+{{
+    "is_complex": boolean,
+    "complexity_score": float,
+    "reasoning": "string",
+    "suggested_subtasks": ["subtask1", "subtask2", ...]
+}}"""
+
+        try:
+            await agent.run(complexity_prompt)
+            # Parse the response (simplified - in production would use proper JSON parsing)
+            # For now, return a default result
+            return TaskComplexityResult(
+                is_complex=True,
+                complexity_score=0.8,
+                reasoning="Task requires multiple distinct operations",
+                suggested_subtasks=[],
+            )
+        except Exception as e:
+            logger.error(f"Error analyzing task complexity: {str(e)}")
+            # Fallback to heuristic
+            return TaskComplexityResult(
+                is_complex=False, complexity_score=0.5, reasoning="Error in analysis, using default"
+            )
+
+    async def _execute_with_decomposition(
+        self, task_node: TaskNode, complexity_result: TaskComplexityResult
+    ) -> Tuple[bool, Any, Optional[str]]:
+        """Execute a task by decomposing it into subtasks.
+
+        Args:
+            task_node: The task node to execute
+            complexity_result: The complexity analysis result
+
+        Returns:
+            Tuple of (success, aggregated_result, error_message)
+        """
+        task_node.status = "in_progress"
+
+        # Generate subtasks
+        subtasks = await self._generate_subtasks(task_node, complexity_result)
+
+        # Allocate iteration budgets
+        subtask_budgets = self._allocate_iteration_budgets(
+            task_node.iteration_budget, len(subtasks)
+        )
+
+        # Execute subtasks
+        results = []
+        errors = []
+
+        # Show UI feedback if thoughts are enabled
+        if self.state_manager.session.show_thoughts:
+            from tunacode.ui import console as ui_console
+            from tunacode.ui.recursive_progress import show_recursive_progress
+
+            await show_recursive_progress(
+                ui_console.console,
+                f"Executing {len(subtasks)} subtasks",
+                task_id=task_node.id,
+                depth=task_node.depth,
+            )
+
+        for i, (subtask_desc, budget) in enumerate(zip(subtasks, subtask_budgets)):
+            if self.state_manager.session.show_thoughts:
+                logger.info(f"Executing subtask {i + 1}/{len(subtasks)}: {subtask_desc[:50]}...")
+
+            # Create subtask context
+            subtask_context = {
+                **task_node.context,
+                "parent_task": task_node.description,
+                "subtask_index": i,
+                "total_subtasks": len(subtasks),
+            }
+
+            # Execute subtask recursively
+            success, result, error = await self.execute_task(
+                subtask_desc,
+                parent_task_id=task_node.id,
+                depth=task_node.depth + 1,
+                inherited_context=subtask_context,
+            )
+
+            if success:
+                results.append(result)
+            else:
+                errors.append(f"Subtask {i + 1} failed: {error}")
+
+            # Show completion status
+            if self.state_manager.session.show_thoughts:
+                from tunacode.ui import console as ui_console
+                from tunacode.ui.recursive_progress import show_task_completion
+
+                await show_task_completion(
+                    ui_console.console,
+                    task_node.id + f".{i + 1}",
+                    success,
+                    depth=task_node.depth + 1,
+                )
+
+        # Aggregate results
+        if errors:
+            task_node.status = "failed"
+            task_node.error = "; ".join(errors)
+            return False, results, f"Some subtasks failed: {'; '.join(errors)}"
+
+        task_node.status = "completed"
+        aggregated_result = await self._aggregate_results(task_node, results)
+        task_node.result = aggregated_result
+
+        return True, aggregated_result, None
+
+    async def _execute_directly(self, task_node: TaskNode) -> Tuple[bool, Any, Optional[str]]:
+        """Execute a task directly using the main agent.
+
+        Args:
+            task_node: The task node to execute
+
+        Returns:
+            Tuple of (success, result, error_message)
+        """
+        task_node.status = "in_progress"
+
+        # Get the main agent
+        agent = self.state_manager.session.agents.get("main")
+        if not agent:
+            error_msg = "Main agent not available"
+            task_node.status = "failed"
+            task_node.error = error_msg
+            return False, None, error_msg
+
+        try:
+            # Execute with the agent
+            result = await agent.run(task_node.description)
+            task_node.status = "completed"
+            task_node.result = result
+            return True, result, None
+
+        except Exception as e:
+            error_msg = f"Error executing task: {str(e)}"
+            logger.error(error_msg)
+            task_node.status = "failed"
+            task_node.error = error_msg
+            return False, None, error_msg
+
+    async def _generate_subtasks(
+        self, task_node: TaskNode, complexity_result: TaskComplexityResult
+    ) -> List[str]:
+        """Generate subtasks for a complex task.
+
+        Args:
+            task_node: The parent task node
+            complexity_result: The complexity analysis result
+
+        Returns:
+            List of subtask descriptions
+        """
+        # If we have suggested subtasks from complexity analysis, use them
+        if complexity_result.suggested_subtasks:
+            return complexity_result.suggested_subtasks
+
+        # Otherwise, use agent to generate subtasks
+        agent = self.state_manager.session.agents.get("main")
+        if not agent:
+            # Fallback to simple decomposition
+            return [f"Part 1 of: {task_node.description}", f"Part 2 of: {task_node.description}"]
+
+        decompose_prompt = f"""Break down this task into 2-5 logical subtasks that can be executed independently:
+
+Task: {task_node.description}
+
+Provide a list of clear, actionable subtasks. Each subtask should:
+- Be self-contained and executable
+- Have clear success criteria
+- Contribute to completing the overall task
+
+Return ONLY a JSON array of subtask descriptions:
+["subtask 1 description", "subtask 2 description", ...]"""
+
+        try:
+            await agent.run(decompose_prompt)
+            # Parse subtasks from result (simplified)
+            # In production, would parse actual JSON response
+            return [
+                f"Step 1: Analyze requirements for {task_node.title}",
+                f"Step 2: Implement core functionality for {task_node.title}",
+                f"Step 3: Test and validate {task_node.title}",
+            ]
+        except Exception as e:
+            logger.error(f"Error generating subtasks: {str(e)}")
+            return [f"Execute: {task_node.description}"]
+
+    def _allocate_iteration_budgets(self, total_budget: int, num_subtasks: int) -> List[int]:
+        """Allocate iteration budget across subtasks.
+
+        Args:
+            total_budget: Total iteration budget available
+            num_subtasks: Number of subtasks
+
+        Returns:
+            List of budgets for each subtask
+        """
+        if num_subtasks == 0:
+            return []
+
+        # Simple equal allocation with remainder distribution
+        base_budget = total_budget // num_subtasks
+        remainder = total_budget % num_subtasks
+
+        budgets = [base_budget] * num_subtasks
+        # Distribute remainder to first tasks
+        for i in range(remainder):
+            budgets[i] += 1
+
+        return budgets
+
+    async def _aggregate_results(self, task_node: TaskNode, subtask_results: List[Any]) -> Any:
+        """Aggregate results from subtasks into a final result.
+
+        Args:
+            task_node: The parent task node
+            subtask_results: List of results from subtasks
+
+        Returns:
+            Aggregated result
+        """
+        if not subtask_results:
+            return "Task completed with no results"
+
+        # Convert subtask results to TaskResult objects
+        task_results = []
+        for i, (subtask, result) in enumerate(zip(task_node.subtasks, subtask_results)):
+            task_results.append(
+                TaskResult(
+                    task_id=subtask.id,
+                    task_title=subtask.title,
+                    result_data=result,
+                    status=subtask.status,
+                    error=subtask.error,
+                )
+            )
+
+        # Use intelligent aggregation
+        aggregated = await self.aggregator.aggregate_results(
+            task_results=task_results,
+            parent_task={
+                "id": task_node.id,
+                "title": task_node.title,
+                "description": task_node.description,
+            },
+            strategy=AggregationStrategy.INTELLIGENT,
+        )
+
+        return aggregated.primary_result
+
+    def get_task_hierarchy(self) -> Dict[str, TaskNode]:
+        """Get the current task hierarchy.
+
+        Returns:
+            Dictionary mapping task IDs to TaskNode objects
+        """
+        return self._task_hierarchy.copy()
+
+    def get_execution_stack(self) -> List[str]:
+        """Get the current execution stack of task IDs.
+
+        Returns:
+            List of task IDs in execution order
+        """
+        return self._execution_stack.copy()
+
+    def clear_hierarchy(self):
+        """Clear the task hierarchy and execution stack."""
+        self._task_hierarchy.clear()
+        self._execution_stack.clear()