tunacode-cli 0.0.41__py3-none-any.whl → 0.0.43__py3-none-any.whl

tunacode/core/agents/utils.py

@@ -1,13 +1,20 @@
 import asyncio
 import importlib
 import json
+import logging
 import os
 import re
 from collections.abc import Iterator
 from datetime import datetime, timezone
 from typing import Any
 
-from tunacode.constants import READ_ONLY_TOOLS
+from tunacode.constants import (
+    JSON_PARSE_BASE_DELAY,
+    JSON_PARSE_MAX_DELAY,
+    JSON_PARSE_MAX_RETRIES,
+    READ_ONLY_TOOLS,
+)
+from tunacode.exceptions import ToolBatchingJSONError
 from tunacode.types import (
     ErrorMessage,
     StateManager,
@@ -16,6 +23,9 @@ from tunacode.types import (
     ToolName,
 )
 from tunacode.ui import console as ui
+from tunacode.utils.retry import retry_json_parse_async
+
+logger = logging.getLogger(__name__)
 
 
 # Lazy import for Agent and Tool
@@ -167,11 +177,28 @@ async def parse_json_tool_calls(
             if brace_count == 0 and start_pos != -1:
                 potential_json = text[start_pos : i + 1]
                 try:
-                    parsed = json.loads(potential_json)
+                    # Use retry logic for JSON parsing
+                    parsed = await retry_json_parse_async(
+                        potential_json,
+                        max_retries=JSON_PARSE_MAX_RETRIES,
+                        base_delay=JSON_PARSE_BASE_DELAY,
+                        max_delay=JSON_PARSE_MAX_DELAY,
+                    )
                     if isinstance(parsed, dict) and "tool" in parsed and "args" in parsed:
                         potential_jsons.append((parsed["tool"], parsed["args"]))
-                except json.JSONDecodeError:
-                    pass
+                except json.JSONDecodeError as e:
+                    # After all retries failed
+                    logger.error(f"JSON parsing failed after {JSON_PARSE_MAX_RETRIES} retries: {e}")
+                    if state_manager.session.show_thoughts:
+                        await ui.error(
+                            f"Failed to parse tool JSON after {JSON_PARSE_MAX_RETRIES} retries"
+                        )
+                    # Raise custom exception for better error handling
+                    raise ToolBatchingJSONError(
+                        json_content=potential_json,
+                        retry_count=JSON_PARSE_MAX_RETRIES,
+                        original_error=e,
+                    ) from e
                 start_pos = -1
 
     matches = potential_jsons
@@ -220,7 +247,13 @@ async def extract_and_execute_tool_calls(
 
     for match in code_matches:
         try:
-            tool_data = json.loads(match)
+            # Use retry logic for JSON parsing in code blocks
+            tool_data = await retry_json_parse_async(
+                match,
+                max_retries=JSON_PARSE_MAX_RETRIES,
+                base_delay=JSON_PARSE_BASE_DELAY,
+                max_delay=JSON_PARSE_MAX_DELAY,
+            )
             if "tool" in tool_data and "args" in tool_data:
 
                 class MockToolCall:
@@ -240,7 +273,22 @@ async def extract_and_execute_tool_calls(
                 if state_manager.session.show_thoughts:
                     await ui.muted(f"FALLBACK: Executed {tool_data['tool']} from code block")
 
-        except (json.JSONDecodeError, KeyError, Exception) as e:
+        except json.JSONDecodeError as e:
+            # After all retries failed
+            logger.error(
+                f"Code block JSON parsing failed after {JSON_PARSE_MAX_RETRIES} retries: {e}"
+            )
+            if state_manager.session.show_thoughts:
+                await ui.error(
+                    f"Failed to parse code block tool JSON after {JSON_PARSE_MAX_RETRIES} retries"
+                )
+            # Raise custom exception for better error handling
+            raise ToolBatchingJSONError(
+                json_content=match,
+                retry_count=JSON_PARSE_MAX_RETRIES,
+                original_error=e,
+            ) from e
+        except (KeyError, Exception) as e:
             if state_manager.session.show_thoughts:
                 await ui.error(f"Error parsing code block tool call: {e!s}")
 

tunacode/core/recursive/__init__.py

@@ -0,0 +1,18 @@
+"""Module: tunacode.core.recursive
+
+Recursive task execution system for complex task decomposition and execution.
+"""
+
+from .aggregator import ResultAggregator
+from .budget import BudgetManager
+from .decomposer import TaskDecomposer
+from .executor import RecursiveTaskExecutor
+from .hierarchy import TaskHierarchy
+
+__all__ = [
+    "RecursiveTaskExecutor",
+    "TaskDecomposer",
+    "TaskHierarchy",
+    "BudgetManager",
+    "ResultAggregator",
+]

tunacode/core/recursive/aggregator.py

@@ -0,0 +1,467 @@
+"""Module: tunacode.core.recursive.aggregator
+
+Result aggregation and context management for recursive task execution.
+"""
+
+import json
+import logging
+from dataclasses import dataclass, field
+from datetime import datetime
+from enum import Enum
+from typing import Any, Dict, List, Optional, Set
+
+from tunacode.core.state import StateManager
+
+logger = logging.getLogger(__name__)
+
+
+class AggregationStrategy(Enum):
+    """Strategies for aggregating results."""
+
+    CONCATENATE = "concatenate"  # Simple concatenation
+    STRUCTURED = "structured"  # Preserve task structure
+    SUMMARY = "summary"  # Generate summary
+    INTELLIGENT = "intelligent"  # Use agent to merge
+
+
+class ConflictResolution(Enum):
+    """Strategies for resolving conflicts in results."""
+
+    LATEST = "latest"  # Use most recent result
+    PRIORITY = "priority"  # Use highest priority result
+    MERGE = "merge"  # Attempt to merge
+    AGENT = "agent"  # Use agent to resolve
+
+
+@dataclass
+class TaskResult:
+    """Result from a single task execution."""
+
+    task_id: str
+    task_title: str
+    result_data: Any
+    status: str  # completed, failed, partial
+    timestamp: datetime = field(default_factory=datetime.now)
+    error: Optional[str] = None
+    metadata: Dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class AggregatedResult:
+    """Aggregated result from multiple tasks."""
+
+    primary_result: Any
+    task_results: List[TaskResult]
+    strategy_used: AggregationStrategy
+    conflicts_resolved: int = 0
+    partial_failures: List[str] = field(default_factory=list)
+    metadata: Dict[str, Any] = field(default_factory=dict)
+    aggregation_time: datetime = field(default_factory=datetime.now)
+
+
+@dataclass
+class ExecutionContext:
+    """Context information during execution."""
+
+    task_id: str
+    parent_context: Dict[str, Any]
+    local_context: Dict[str, Any]
+    files_accessed: Set[str] = field(default_factory=set)
+    tools_used: List[str] = field(default_factory=list)
+    key_findings: List[str] = field(default_factory=list)
+
+
+class ResultAggregator:
+    """Aggregates results from distributed subtask executions."""
+
+    def __init__(self, state_manager: StateManager):
+        """Initialize the ResultAggregator.
+
+        Args:
+            state_manager: StateManager for accessing agents
+        """
+        self.state_manager = state_manager
+        self._context_cache: Dict[str, ExecutionContext] = {}
+
+    async def aggregate_results(
+        self,
+        task_results: List[TaskResult],
+        parent_task: Optional[Dict[str, Any]] = None,
+        strategy: AggregationStrategy = AggregationStrategy.INTELLIGENT,
+    ) -> AggregatedResult:
+        """Aggregate results from multiple subtasks.
+
+        Args:
+            task_results: List of results from subtasks
+            parent_task: Optional parent task information
+            strategy: Aggregation strategy to use
+
+        Returns:
+            AggregatedResult with merged data
+        """
+        if not task_results:
+            return AggregatedResult(
+                primary_result="No results to aggregate", task_results=[], strategy_used=strategy
+            )
+
+        # Separate successful and failed results
+        successful_results = [r for r in task_results if r.status == "completed"]
+        failed_results = [r for r in task_results if r.status == "failed"]
+        partial_results = [r for r in task_results if r.status == "partial"]
+
+        # Handle complete failure
+        if not successful_results and not partial_results:
+            return AggregatedResult(
+                primary_result="All subtasks failed",
+                task_results=task_results,
+                strategy_used=strategy,
+                partial_failures=[r.task_id for r in failed_results],
+            )
+
+        # Apply aggregation strategy
+        if strategy == AggregationStrategy.CONCATENATE:
+            result = await self._aggregate_concatenate(successful_results + partial_results)
+        elif strategy == AggregationStrategy.STRUCTURED:
+            result = await self._aggregate_structured(
+                successful_results + partial_results, parent_task
+            )
+        elif strategy == AggregationStrategy.SUMMARY:
+            result = await self._aggregate_summary(
+                successful_results + partial_results, parent_task
+            )
+        elif strategy == AggregationStrategy.INTELLIGENT:
+            result = await self._aggregate_intelligent(
+                successful_results + partial_results, parent_task
+            )
+        else:
+            result = await self._aggregate_concatenate(successful_results + partial_results)
+
+        return AggregatedResult(
+            primary_result=result,
+            task_results=task_results,
+            strategy_used=strategy,
+            partial_failures=[r.task_id for r in failed_results],
+        )
+
+    async def _aggregate_concatenate(self, results: List[TaskResult]) -> str:
+        """Simple concatenation of results.
+
+        Args:
+            results: List of successful results
+
+        Returns:
+            Concatenated string result
+        """
+        parts = []
+        for i, result in enumerate(results):
+            parts.append(f"=== Task {i + 1}: {result.task_title} ===")
+            parts.append(str(result.result_data))
+            parts.append("")
+
+        return "\n".join(parts)
+
+    async def _aggregate_structured(
+        self, results: List[TaskResult], parent_task: Optional[Dict[str, Any]]
+    ) -> Dict[str, Any]:
+        """Create structured aggregation preserving task relationships.
+
+        Args:
+            results: List of successful results
+            parent_task: Parent task information
+
+        Returns:
+            Structured dictionary result
+        """
+        structured_result = {
+            "parent_task": parent_task.get("title") if parent_task else "Unknown",
+            "completed_subtasks": len([r for r in results if r.status == "completed"]),
+            "partial_subtasks": len([r for r in results if r.status == "partial"]),
+            "subtask_results": [],
+        }
+
+        for result in results:
+            structured_result["subtask_results"].append(
+                {
+                    "task_id": result.task_id,
+                    "title": result.task_title,
+                    "status": result.status,
+                    "result": result.result_data,
+                    "timestamp": result.timestamp.isoformat(),
+                }
+            )
+
+        return structured_result
+
+    async def _aggregate_summary(
+        self, results: List[TaskResult], parent_task: Optional[Dict[str, Any]]
+    ) -> str:
+        """Generate a summary of results.
+
+        Args:
+            results: List of successful results
+            parent_task: Parent task information
+
+        Returns:
+            Summary string
+        """
+        summary_parts = []
+
+        if parent_task:
+            summary_parts.append(f"Summary for: {parent_task.get('title', 'Unknown Task')}")
+            summary_parts.append("")
+
+        summary_parts.append(f"Completed {len(results)} subtasks:")
+        summary_parts.append("")
+
+        for i, result in enumerate(results):
+            # Extract key information from result
+            result_str = str(result.result_data)
+            preview = result_str[:200] + "..." if len(result_str) > 200 else result_str
+
+            summary_parts.append(f"{i + 1}. {result.task_title}")
+            summary_parts.append(f"   Status: {result.status}")
+            summary_parts.append(f"   Result: {preview}")
+            summary_parts.append("")
+
+        return "\n".join(summary_parts)
+
+    async def _aggregate_intelligent(
+        self, results: List[TaskResult], parent_task: Optional[Dict[str, Any]]
+    ) -> Any:
+        """Use agent to intelligently merge results.
+
+        Args:
+            results: List of successful results
+            parent_task: Parent task information
+
+        Returns:
+            Intelligently merged result
+        """
+        agent = self.state_manager.session.agents.get("main")
+        if not agent:
+            # Fallback to summary
+            return await self._aggregate_summary(results, parent_task)
+
+        # Prepare context for agent
+        context = {
+            "parent_task": parent_task.get("description") if parent_task else "Unknown",
+            "subtask_count": len(results),
+            "subtask_results": [],
+        }
+
+        for result in results:
+            context["subtask_results"].append(
+                {
+                    "title": result.task_title,
+                    "result": str(result.result_data)[:500],  # Limit size
+                }
+            )
+
+        merge_prompt = f"""Intelligently merge and synthesize these subtask results into a coherent response.
+
+Parent Task: {context["parent_task"]}
+
+Subtask Results:
+{json.dumps(context["subtask_results"], indent=2)}
+
+Instructions:
+1. Identify the key achievements from each subtask
+2. Synthesize findings into a unified response
+3. Highlight any important patterns or insights
+4. Present the result in a clear, actionable format
+
+Provide a comprehensive but concise synthesis that addresses the original task."""
+
+        try:
+            result = await agent.run(merge_prompt)
+            return result
+        except Exception as e:
+            logger.error(f"Agent-based aggregation failed: {str(e)}")
+            # Fallback to summary
+            return await self._aggregate_summary(results, parent_task)
+
+    def create_context(
+        self, task_id: str, parent_context: Optional[Dict[str, Any]] = None
+    ) -> ExecutionContext:
+        """Create an execution context for a task.
+
+        Args:
+            task_id: Task identifier
+            parent_context: Optional parent context to inherit
+
+        Returns:
+            New execution context
+        """
+        context = ExecutionContext(
+            task_id=task_id, parent_context=parent_context or {}, local_context={}
+        )
+
+        self._context_cache[task_id] = context
+        return context
+
+    def get_context(self, task_id: str) -> Optional[ExecutionContext]:
+        """Get execution context for a task.
+
+        Args:
+            task_id: Task identifier
+
+        Returns:
+            Execution context or None
+        """
+        return self._context_cache.get(task_id)
+
+    def update_context(
+        self,
+        task_id: str,
+        updates: Dict[str, Any],
+        files: Optional[Set[str]] = None,
+        tools: Optional[List[str]] = None,
+        findings: Optional[List[str]] = None,
+    ) -> None:
+        """Update execution context for a task.
+
+        Args:
+            task_id: Task identifier
+            updates: Context updates
+            files: Files accessed
+            tools: Tools used
+            findings: Key findings
+        """
+        context = self._context_cache.get(task_id)
+        if not context:
+            logger.warning(f"No context found for task {task_id}")
+            return
+
+        context.local_context.update(updates)
+
+        if files:
+            context.files_accessed.update(files)
+        if tools:
+            context.tools_used.extend(tools)
+        if findings:
+            context.key_findings.extend(findings)
+
+    def synthesize_contexts(self, task_ids: List[str]) -> Dict[str, Any]:
+        """Synthesize contexts from multiple tasks.
+
+        Args:
+            task_ids: List of task IDs
+
+        Returns:
+            Synthesized context dictionary
+        """
+        synthesized = {
+            "files_accessed": set(),
+            "tools_used": [],
+            "key_findings": [],
+            "combined_context": {},
+        }
+
+        for task_id in task_ids:
+            context = self._context_cache.get(task_id)
+            if context:
+                synthesized["files_accessed"].update(context.files_accessed)
+                synthesized["tools_used"].extend(context.tools_used)
+                synthesized["key_findings"].extend(context.key_findings)
+                synthesized["combined_context"].update(context.local_context)
+
+        # Convert set to list for JSON serialization
+        synthesized["files_accessed"] = list(synthesized["files_accessed"])
+
+        # Remove duplicates from tools while preserving order
+        seen = set()
+        unique_tools = []
+        for tool in synthesized["tools_used"]:
+            if tool not in seen:
+                seen.add(tool)
+                unique_tools.append(tool)
+        synthesized["tools_used"] = unique_tools
+
+        return synthesized
+
+    async def resolve_conflicts(
+        self,
+        conflicting_results: List[TaskResult],
+        strategy: ConflictResolution = ConflictResolution.AGENT,
+    ) -> TaskResult:
+        """Resolve conflicts between contradictory results.
+
+        Args:
+            conflicting_results: List of conflicting results
+            strategy: Resolution strategy
+
+        Returns:
+            Single resolved result
+        """
+        if not conflicting_results:
+            raise ValueError("No results to resolve conflicts for")
+
+        if len(conflicting_results) == 1:
+            return conflicting_results[0]
+
+        if strategy == ConflictResolution.LATEST:
+            # Return most recent result
+            return max(conflicting_results, key=lambda r: r.timestamp)
+
+        elif strategy == ConflictResolution.PRIORITY:
+            # Use metadata priority if available
+            def get_priority(result: TaskResult) -> int:
+                return result.metadata.get("priority", 0)
+
+            return max(conflicting_results, key=get_priority)
+
+        elif strategy == ConflictResolution.MERGE:
+            # Simple merge attempt
+            merged_data = {}
+            for result in conflicting_results:
+                if isinstance(result.result_data, dict):
+                    merged_data.update(result.result_data)
+                else:
+                    # Can't merge non-dict results
+                    return conflicting_results[-1]
+
+            return TaskResult(
+                task_id="merged",
+                task_title="Merged Result",
+                result_data=merged_data,
+                status="completed",
+            )
+
+        elif strategy == ConflictResolution.AGENT:
+            # Use agent to resolve
+            agent = self.state_manager.session.agents.get("main")
+            if not agent:
+                # Fallback to latest
+                return max(conflicting_results, key=lambda r: r.timestamp)
+
+            conflict_data = []
+            for result in conflicting_results:
+                conflict_data.append(
+                    {
+                        "task": result.task_title,
+                        "result": str(result.result_data)[:300],
+                        "timestamp": result.timestamp.isoformat(),
+                    }
+                )
+
+            resolution_prompt = f"""Resolve conflicts between these task results:
+
+{json.dumps(conflict_data, indent=2)}
+
+Analyze the differences and provide a single, coherent result that best represents the correct outcome."""
+
+            try:
+                resolved = await agent.run(resolution_prompt)
+                return TaskResult(
+                    task_id="resolved",
+                    task_title="Conflict Resolved",
+                    result_data=resolved,
+                    status="completed",
+                )
+            except Exception as e:
+                logger.error(f"Agent conflict resolution failed: {str(e)}")
+                # Fallback to latest
+                return max(conflicting_results, key=lambda r: r.timestamp)
+
+        # Default to latest
+        return max(conflicting_results, key=lambda r: r.timestamp)