vibe_data_ana/tests/test_task_execution_properties.py

"""Property-based tests for task execution engine."""

import pytest
import pandas as pd
from hypothesis import given, strategies as st, settings

from src.engines.task_execution import (
    execute_task,
    call_tool,
    extract_insights,
    _fallback_task_execution
)
from src.models.analysis_plan import AnalysisTask
from src.data_access import DataAccessLayer
from src.tools.stats_tools import CalculateStatisticsTool


# Feature: true-ai-agent, Property 13: 任务执行完整性
@given(
    task_name=st.text(min_size=5, max_size=50),
    task_description=st.text(min_size=10, max_size=100)
)
@settings(max_examples=10, deadline=None)
def test_task_execution_completeness(task_name, task_description):
    """
    Property 13: For any valid analysis plan and tool set, the task execution
    engine should be able to execute all non-skipped tasks and generate an
    analysis result (success or failure) for each task.
    
    Validates: 场景1验收.3, FR-5.1
    """
    # Create sample data
    sample_data = pd.DataFrame({
        'value': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
        'category': ['A', 'B', 'A', 'B', 'A', 'B', 'A', 'B', 'A', 'B']
    })
    
    # Create sample tools
    sample_tools = [CalculateStatisticsTool()]
    
    # Create task
    task = AnalysisTask(
        id="test_task",
        name=task_name,
        description=task_description,
        priority=3,
        required_tools=['calculate_statistics']
    )
    
    # Create data access
    data_access = DataAccessLayer(sample_data)
    
    # Execute task (using fallback to avoid API dependency)
    result = _fallback_task_execution(task, sample_tools, data_access)
    
    # Verify: Should return AnalysisResult
    assert result is not None
    assert result.task_id == task.id
    assert result.task_name == task.name
    
    # Verify: Should have success status
    assert isinstance(result.success, bool)
    
    # Verify: Should have execution time
    assert result.execution_time >= 0
    
    # Verify: If failed, should have error message
    if not result.success:
        assert result.error is not None
    
    # Verify: Should have insights (even if empty)
    assert isinstance(result.insights, list)


# Feature: true-ai-agent, Property 14: ReAct 循环终止
def test_react_loop_termination():
    """
    Property 14: For any analysis task, the ReAct execution loop should
    terminate within a finite number of steps (either complete the task
    or reach maximum iterations), and should not loop infinitely.
    
    Validates: FR-5.1
    """
    sample_data = pd.DataFrame({
        'value': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
        'category': ['A', 'B', 'A', 'B', 'A', 'B', 'A', 'B', 'A', 'B']
    })
    sample_tools = [CalculateStatisticsTool()]
    
    task = AnalysisTask(
        id="test_task",
        name="Test Task",
        description="Calculate statistics",
        priority=3,
        required_tools=['calculate_statistics']
    )
    
    data_access = DataAccessLayer(sample_data)
    
    # Execute with limited iterations
    result = _fallback_task_execution(task, sample_tools, data_access)
    
    # Verify: Should complete (not hang)
    assert result is not None
    
    # Verify: Should have finite execution time
    assert result.execution_time < 60, "Execution should complete within 60 seconds"


# Feature: true-ai-agent, Property 15: 异常识别
def test_anomaly_identification():
    """
    Property 15: For any data containing obvious anomalies (e.g., a category
    accounting for >80% of data, or values exceeding 3 standard deviations),
    the task execution engine should be able to mark the anomaly in the
    analysis result insights.
    
    Validates: 场景4验收.1
    """
    # Create data with anomaly (category A is 90%)
    anomaly_data = pd.DataFrame({
        'value': list(range(100)),
        'category': ['A'] * 90 + ['B'] * 10
    })
    
    task = AnalysisTask(
        id="test_task",
        name="Anomaly Detection",
        description="Detect anomalies in data",
        priority=3,
        required_tools=['calculate_statistics']
    )
    
    data_access = DataAccessLayer(anomaly_data)
    tools = [CalculateStatisticsTool()]
    
    result = _fallback_task_execution(task, tools, data_access)
    
    # Verify: Should complete successfully
    assert result.success or result.error is not None
    
    # Verify: Should have insights
    assert isinstance(result.insights, list)


# Test tool calling
def test_call_tool_success():
    """Test successful tool calling."""
    sample_data = pd.DataFrame({
        'value': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
        'category': ['A', 'B', 'A', 'B', 'A', 'B', 'A', 'B', 'A', 'B']
    })
    
    tool = CalculateStatisticsTool()
    data_access = DataAccessLayer(sample_data)
    
    result = call_tool(tool, data_access, column='value')
    
    # Should return result dict
    assert isinstance(result, dict)
    assert 'success' in result


# Test insight extraction
def test_extract_insights_without_ai():
    """Test insight extraction without AI."""
    history = [
        {'type': 'thought', 'content': 'Analyzing data'},
        {'type': 'action', 'tool': 'calculate_statistics'},
        {'type': 'observation', 'result': {'data': {'mean': 5.5}}}
    ]
    
    insights = extract_insights(history, client=None)
    
    # Should return list of insights
    assert isinstance(insights, list)
    assert len(insights) > 0


# Test execution with empty tools
def test_execution_with_no_tools():
    """Test execution when no tools are available."""
    sample_data = pd.DataFrame({
        'value': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
        'category': ['A', 'B', 'A', 'B', 'A', 'B', 'A', 'B', 'A', 'B']
    })
    
    task = AnalysisTask(
        id="test_task",
        name="Test Task",
        description="Test",
        priority=3,
        required_tools=['nonexistent_tool']
    )
    
    data_access = DataAccessLayer(sample_data)
    
    result = _fallback_task_execution(task, [], data_access)
    
    # Should fail gracefully
    assert not result.success
    assert result.error is not None