canrun / src /canrun_engine.py
grasant's picture
Upload 24 files
d86b25e verified
raw
history blame
36.3 kB
"""
CanRun Engine - Main orchestration module for Universal Game Compatibility Checker
Privacy-focused game compatibility analysis for NVIDIA RTX/GTX systems.
"""
import logging
import asyncio
import json
import os
import re
from typing import Dict, List, Optional, Tuple, Any
from dataclasses import dataclass, asdict
from datetime import datetime, timedelta
from src.privacy_aware_hardware_detector import PrivacyAwareHardwareDetector, PrivacyAwareHardwareSpecs
from src.game_requirements_fetcher import GameRequirementsFetcher, GameRequirements
from src.optimized_game_fuzzy_matcher import OptimizedGameFuzzyMatcher
from src.compatibility_analyzer import CompatibilityAnalyzer, CompatibilityAnalysis, ComponentAnalysis, ComponentType, CompatibilityLevel
from src.dynamic_performance_predictor import DynamicPerformancePredictor, PerformanceAssessment, PerformanceTier
from src.rtx_llm_analyzer import GAssistLLMAnalyzer, LLMAnalysisResult
@dataclass
class CanRunResult:
"""Complete CanRun analysis result."""
game_name: str
timestamp: str
hardware_specs: PrivacyAwareHardwareSpecs
game_requirements: GameRequirements
compatibility_analysis: CompatibilityAnalysis
performance_prediction: PerformanceAssessment
llm_analysis: Optional[Dict[str, LLMAnalysisResult]]
cache_used: bool
analysis_time_ms: int
def get_minimum_requirements_status(self) -> Dict[str, Any]:
"""Get clear status about minimum requirements compliance."""
return self.compatibility_analysis.get_minimum_requirements_status()
def get_runnable_status_message(self) -> str:
"""Get simple runnable status message for CANRUN."""
return self.compatibility_analysis.get_runnable_status()
def can_run_game(self) -> bool:
"""Check if the game can run on minimum requirements."""
return self.compatibility_analysis.can_run_minimum
def exceeds_recommended_requirements(self) -> bool:
"""Check if system exceeds recommended requirements."""
return self.compatibility_analysis.can_run_recommended
class CanRunEngine:
"""Main CanRun engine for privacy-aware game compatibility checking."""
def __init__(self, cache_dir: str = "cache", enable_llm: bool = True):
"""Initialize CanRun engine with all components."""
assert isinstance(cache_dir, str), "Cache directory must be a string"
assert isinstance(enable_llm, bool), "LLM enable flag must be boolean"
self.logger = logging.getLogger(__name__)
self.cache_dir = cache_dir
self.cache_duration = timedelta(minutes=15)
self.enable_llm = enable_llm
# Initialize G-Assist LLM analyzer if enabled
self.llm_analyzer = None
if enable_llm:
try:
self.llm_analyzer = GAssistLLMAnalyzer()
self.logger.info("G-Assist LLM analyzer initialized")
except Exception as e:
self.logger.warning(f"LLM analyzer initialization failed: {e}")
# Initialize components with LLM analyzer
self.hardware_detector = PrivacyAwareHardwareDetector()
self.requirements_fetcher = GameRequirementsFetcher(self.llm_analyzer)
self.fuzzy_matcher = OptimizedGameFuzzyMatcher()
self.compatibility_analyzer = CompatibilityAnalyzer()
self.performance_predictor = DynamicPerformancePredictor()
# Create cache directory and validate
os.makedirs(cache_dir, exist_ok=True)
assert os.path.isdir(cache_dir), f"Cache directory creation failed: {cache_dir}"
# Session hardware cache
self._hardware_cache: Optional[PrivacyAwareHardwareSpecs] = None
self.logger.info("CanRun engine initialized successfully")
async def check_game_compatibility(self, game_name: str, use_cache: bool = True) -> CanRunResult:
"""
Main entry point for game compatibility checking.
Args:
game_name: Name of the game to check
use_cache: Whether to use cached results
Returns:
Complete CanRun analysis result
"""
# Validate inputs
assert game_name and isinstance(game_name, str), "Game name must be a non-empty string"
assert isinstance(use_cache, bool), "Cache flag must be boolean"
game_name = game_name.strip()
assert len(game_name) > 0, "Game name cannot be empty after strip"
start_time = datetime.now()
self.logger.info(f"Starting compatibility check for: {game_name}")
# Step 1: Centralized Game Name Correction
all_known_games = self.requirements_fetcher.get_all_cached_game_names()
if not all_known_games:
self.logger.warning("No games in local cache. Matching will rely on external sources.")
match_result = await self.fuzzy_matcher.find_best_match(game_name, all_known_games)
if not match_result:
# If no match is found, we proceed with the original game name and let the fetcher handle it.
self.logger.warning(f"No confident match for '{game_name}'. Proceeding with original name.")
corrected_game_name = game_name
else:
corrected_game_name, match_confidence = match_result
self.logger.info(f"Query '{game_name}' matched to '{corrected_game_name}' with confidence {match_confidence:.2f}")
# Step 2: Check Cache with Corrected Name
if use_cache:
normalized_name = self.fuzzy_matcher.normalize_game_name(corrected_game_name)
cache_file = os.path.join(self.cache_dir, f"{normalized_name}.json")
cached_result = self._load_cache_file(cache_file)
if cached_result:
self.logger.info(f"Returning cached result for '{corrected_game_name}'")
return cached_result
# Step 3: Fetch Requirements with Corrected Name
game_requirements = await self._fetch_game_requirements(corrected_game_name)
if game_requirements is None:
raise ValueError(f"Game requirements not found for '{corrected_game_name}'.")
# Step 4: Get Hardware Specifications
hardware_specs = await self._get_hardware_specs()
assert hardware_specs is not None, "Hardware detection failed"
# Step 3: Analyze compatibility
compatibility_analysis = await self._analyze_compatibility(
game_name, hardware_specs, game_requirements
)
assert compatibility_analysis is not None, "Compatibility analysis failed"
# Step 4: Predict performance using S-A-B-C-D-F tier system
hardware_dict = {
"gpu_model": hardware_specs.gpu_model,
"gpu_vram_gb": hardware_specs.gpu_vram_gb,
"cpu_model": hardware_specs.cpu_model,
"ram_total_gb": hardware_specs.ram_total_gb,
"supports_rtx": hardware_specs.supports_rtx,
"supports_dlss": hardware_specs.supports_dlss
}
game_requirements_dict = {
"minimum_gpu": game_requirements.minimum_gpu,
"recommended_gpu": game_requirements.recommended_gpu,
"minimum_cpu": game_requirements.minimum_cpu,
"recommended_cpu": game_requirements.recommended_cpu,
"minimum_ram_gb": game_requirements.minimum_ram_gb,
"recommended_ram_gb": game_requirements.recommended_ram_gb
}
performance_prediction = await asyncio.get_event_loop().run_in_executor(
None, self.performance_predictor.assess_performance, hardware_dict, game_requirements_dict
)
assert performance_prediction is not None, "Performance assessment failed"
# Step 5: Perform LLM analysis if enabled
llm_analysis = None
if self.llm_analyzer:
llm_analysis = await self._perform_llm_analysis(
compatibility_analysis, performance_prediction, hardware_specs
)
# Calculate analysis time
analysis_time = int((datetime.now() - start_time).total_seconds() * 1000)
# Create result
result = CanRunResult(
game_name=corrected_game_name,
timestamp=datetime.now().isoformat(),
hardware_specs=hardware_specs,
game_requirements=game_requirements,
compatibility_analysis=compatibility_analysis,
performance_prediction=performance_prediction,
llm_analysis=llm_analysis,
cache_used=False,
analysis_time_ms=analysis_time
)
# Cache result
if use_cache:
self._save_cached_result(corrected_game_name, result)
self.logger.info(f"Analysis completed for {game_name} in {analysis_time}ms")
return result
async def get_hardware_info(self) -> PrivacyAwareHardwareSpecs:
"""Get current hardware specifications."""
return await self._get_hardware_specs()
async def batch_check_games(self, game_names: List[str], use_cache: bool = True) -> List[CanRunResult]:
"""Check compatibility for multiple games."""
assert isinstance(game_names, list), "Game names must be a list"
assert all(isinstance(name, str) for name in game_names), "All game names must be strings"
assert len(game_names) > 0, "Game names list cannot be empty"
self.logger.info(f"Starting batch check for {len(game_names)} games")
results = []
for game_name in game_names:
try:
result = await self.check_game_compatibility(game_name, use_cache)
results.append(result)
except Exception as e:
self.logger.error(f"Batch check failed for {game_name}: {e}")
results.append(self._create_error_result(game_name, str(e)))
self.logger.info(f"Batch check completed for {len(game_names)} games")
return results
def clear_cache(self) -> None:
"""Clear all cached results."""
assert os.path.isdir(self.cache_dir), "Cache directory does not exist"
cache_files = [f for f in os.listdir(self.cache_dir) if f.endswith('.json')]
for cache_file in cache_files:
os.remove(os.path.join(self.cache_dir, cache_file))
self.logger.info(f"Cleared {len(cache_files)} cache files")
def get_cache_stats(self) -> Dict[str, int]:
"""Get cache statistics."""
assert os.path.isdir(self.cache_dir), "Cache directory does not exist"
cache_files = [f for f in os.listdir(self.cache_dir) if f.endswith('.json')]
total_size = sum(os.path.getsize(os.path.join(self.cache_dir, f)) for f in cache_files)
return {
'total_files': len(cache_files),
'total_size_bytes': total_size,
'total_size_mb': round(total_size / (1024 * 1024), 2)
}
async def _get_hardware_specs(self) -> PrivacyAwareHardwareSpecs:
"""Get hardware specifications with session caching."""
if self._hardware_cache is None:
# Since get_hardware_specs is now async, we await it directly
self._hardware_cache = await self.hardware_detector.get_hardware_specs()
assert self._hardware_cache is not None, "Hardware detection returned None"
return self._hardware_cache
async def _fetch_game_requirements(self, game_name: str) -> GameRequirements:
"""Fetch game requirements from available sources."""
assert game_name and isinstance(game_name, str), "Game name must be a non-empty string"
requirements = await self.requirements_fetcher.fetch_requirements(game_name)
assert requirements is not None, f"Requirements not found for {game_name}"
return requirements
async def _analyze_compatibility(self, game_name: str,
hardware_specs: PrivacyAwareHardwareSpecs,
game_requirements: GameRequirements) -> CompatibilityAnalysis:
"""Analyze hardware compatibility with game requirements."""
assert all([game_name, hardware_specs, game_requirements]), "All parameters are required"
analysis = await asyncio.get_event_loop().run_in_executor(
None, self.compatibility_analyzer.analyze_compatibility,
game_name, hardware_specs, game_requirements
)
assert analysis is not None, "Compatibility analysis returned None"
return analysis
async def _predict_advanced_performance(self, hardware_specs: Dict, game_requirements: Dict = None) -> Dict:
"""
Predict game performance using the advanced tiered assessment system.
Args:
hardware_specs: Hardware specifications from the detector
game_requirements: Optional game requirements
Returns:
Dict containing advanced performance assessment with tier information
"""
loop = asyncio.get_event_loop()
assessment = await loop.run_in_executor(
None,
self.performance_predictor.predict_advanced_performance,
hardware_specs,
game_requirements
)
# Convert assessment to dict for compatibility
return {
'tier': assessment.tier.name,
'tier_description': assessment.tier_description,
'score': assessment.score,
'expected_fps': assessment.expected_fps,
'recommended_settings': assessment.recommended_settings,
'recommended_resolution': assessment.recommended_resolution,
'bottlenecks': assessment.bottlenecks,
'upgrade_suggestions': assessment.upgrade_suggestions
}
def _get_cached_result(self, game_name: str) -> Optional[CanRunResult]:
"""DEPRECATED: This method is no longer the primary way to get cached results.
It is kept for potential direct cache inspection but should not be used in the main workflow.
The main workflow now fetches requirements first, then checks the cache with the corrected name.
"""
normalized_name = self.fuzzy_matcher.normalize_game_name(game_name)
cache_file = os.path.join(self.cache_dir, f"{normalized_name}.json")
return self._load_cache_file(cache_file)
def _load_cache_file(self, cache_file: str) -> Optional[CanRunResult]:
"""Load and validate a single cache file."""
if not os.path.isfile(cache_file):
return None
try:
mtime = os.path.getmtime(cache_file)
if (datetime.now().timestamp() - mtime) > self.cache_duration.total_seconds():
# Cache expired
return None
with open(cache_file, "r", encoding="utf-8") as f:
data = json.load(f)
# Convert dictionary data back to proper dataclass objects
return self._reconstruct_canrun_result(data)
except Exception as e:
self.logger.warning(f"Failed to load cache file {cache_file}: {e}")
return None
def _reconstruct_canrun_result(self, data: Dict[str, Any]) -> CanRunResult:
"""Reconstruct CanRunResult from dictionary data."""
# Reconstruct nested dataclasses
hardware_specs = PrivacyAwareHardwareSpecs(**data['hardware_specs'])
game_requirements = GameRequirements(**data['game_requirements'])
# Reconstruct compatibility analysis with proper ComponentAnalysis objects
compat_data = data['compatibility_analysis'].copy()
if 'component_analyses' in compat_data:
component_analyses = []
for comp_data in compat_data['component_analyses']:
if isinstance(comp_data, dict):
# Handle enum serialization - extract value from string representation
component_value = comp_data['component']
if isinstance(component_value, str):
# Handle both "ComponentType.GPU" and "GPU" formats
if '.' in component_value:
component_value = component_value.split('.')[-1] # Extract "GPU" from "ComponentType.GPU"
try:
component_type = ComponentType[component_value]
except KeyError:
component_type = ComponentType(component_value)
else:
component_type = component_value
# Convert dictionary back to ComponentAnalysis
component_analyses.append(ComponentAnalysis(
component=component_type,
meets_minimum=comp_data['meets_minimum'],
meets_recommended=comp_data['meets_recommended'],
score=comp_data['score'],
bottleneck_factor=comp_data['bottleneck_factor'],
details=comp_data['details'],
upgrade_suggestion=comp_data.get('upgrade_suggestion')
))
else:
# Already a ComponentAnalysis object
component_analyses.append(comp_data)
compat_data['component_analyses'] = component_analyses
# Convert CompatibilityLevel from string if needed
if isinstance(compat_data.get('overall_compatibility'), str):
compat_data['overall_compatibility'] = CompatibilityLevel(compat_data['overall_compatibility'])
# Convert bottlenecks from strings to ComponentType if needed
if 'bottlenecks' in compat_data:
bottlenecks = []
for bottleneck in compat_data['bottlenecks']:
if isinstance(bottleneck, str):
# Handle both "ComponentType.GPU" and "GPU" formats
if '.' in bottleneck:
bottleneck = bottleneck.split('.')[-1] # Extract "GPU" from "ComponentType.GPU"
try:
bottlenecks.append(ComponentType[bottleneck])
except KeyError:
bottlenecks.append(ComponentType(bottleneck))
else:
bottlenecks.append(bottleneck)
compat_data['bottlenecks'] = bottlenecks
compatibility_analysis = CompatibilityAnalysis(**compat_data)
performance_prediction = PerformanceAssessment(**data['performance_prediction'])
# Handle LLM analysis if present
llm_analysis = None
if data.get('llm_analysis'):
llm_analysis = {}
for key, value in data['llm_analysis'].items():
llm_analysis[key] = LLMAnalysisResult(**value)
return CanRunResult(
game_name=data['game_name'],
timestamp=data['timestamp'],
hardware_specs=hardware_specs,
game_requirements=game_requirements,
compatibility_analysis=compatibility_analysis,
performance_prediction=performance_prediction,
llm_analysis=llm_analysis,
cache_used=data.get('cache_used', True),
analysis_time_ms=data.get('analysis_time_ms', 0)
)
def _save_cached_result(self, game_name: str, result: CanRunResult) -> None:
"""Save analysis result to cache using normalized game name."""
# Normalize game name for consistent caching
# This ensures "Diablo 4" and "Diablo IV" use the same cache file
normalized_name = self.fuzzy_matcher.normalize_game_name(game_name)
cache_file = os.path.join(self.cache_dir, f"{normalized_name}.json")
# Ensure cache directory exists
os.makedirs(self.cache_dir, exist_ok=True)
try:
# Convert dataclass to dict recursively, handling nested dataclasses
result_dict = asdict(result)
# Update the result to use the normalized name for consistency
result_dict['game_name'] = normalized_name
with open(cache_file, "w", encoding="utf-8") as f:
json.dump(result_dict, f, indent=2, default=str)
self.logger.debug(f"Cached result for '{game_name}' as '{normalized_name}'")
except Exception as e:
self.logger.warning(f"Failed to save cache for {game_name}: {e}")
async def _perform_llm_analysis(self, compatibility_analysis: CompatibilityAnalysis,
performance_prediction: PerformanceAssessment,
hardware_specs: PrivacyAwareHardwareSpecs) -> Optional[Dict[str, LLMAnalysisResult]]:
"""Perform LLM analysis if G-Assist is available."""
if not self.llm_analyzer:
return None
try:
# Create analysis context for LLM
context = {
'compatibility': compatibility_analysis,
'performance': performance_prediction,
'hardware': hardware_specs
}
# Perform LLM analysis
llm_result = await self.llm_analyzer.analyze_bottlenecks(context)
return {'analysis': llm_result} if llm_result else None
except Exception as e:
self.logger.warning(f"LLM analysis failed: {e}")
return None
def _create_error_result(self, game_name: str, error_message: str) -> CanRunResult:
"""Create an error result for failed analysis."""
from datetime import datetime
# Create minimal error hardware specs
error_hardware = PrivacyAwareHardwareSpecs(
gpu_model="Unknown",
gpu_vram_gb=0,
cpu_name="Unknown",
cpu_cores=0,
cpu_threads=0,
ram_gb=0,
is_nvidia_gpu=False,
supports_rtx=False,
supports_dlss=False,
nvidia_driver_version="Unknown",
os_name="Unknown",
directx_version="Unknown"
)
# Create minimal error requirements
error_requirements = GameRequirements(
game_name=game_name,
minimum_cpu="Unknown",
minimum_gpu="Unknown",
minimum_ram_gb=0,
minimum_vram_gb=0,
minimum_storage_gb=0,
recommended_cpu="Unknown",
recommended_gpu="Unknown",
recommended_ram_gb=0,
recommended_vram_gb=0,
recommended_storage_gb=0,
supports_rtx=False,
supports_dlss=False,
directx_version="Unknown"
)
# Create error compatibility analysis
error_compatibility = CompatibilityAnalysis(
game_name=game_name,
overall_compatibility="incompatible",
cpu_compatibility="error",
gpu_compatibility="error",
ram_compatibility="error",
vram_compatibility="error",
storage_compatibility="error",
overall_score=0,
bottlenecks=[f"Error: {error_message}"],
recommendations=[]
)
# Create error performance assessment
error_performance = PerformanceAssessment(
score=0,
tier=PerformanceTier.F,
tier_description="Error occurred during analysis",
expected_fps=0,
recommended_settings="Unable to determine",
recommended_resolution="Unknown",
bottlenecks=[],
upgrade_suggestions=["Please retry the analysis"]
)
return CanRunResult(
game_name=game_name,
timestamp=datetime.now().isoformat(),
hardware_specs=error_hardware,
game_requirements=error_requirements,
compatibility_analysis=error_compatibility,
performance_prediction=error_performance,
llm_analysis=None,
cache_used=False,
analysis_time_ms=0
)
def _parse_ram_value(self, ram_str: str) -> int:
"""Parse RAM value from string to integer GB."""
if not ram_str or ram_str == "Unknown":
return 0
# Extract number from strings like "8 GB", "16GB", "8192 MB", etc.
ram_str = str(ram_str).upper()
# Match number followed by optional space and unit
match = re.search(r'(\d+)\s*(GB|MB|G|M)?', ram_str)
if match:
value = int(match.group(1))
unit = match.group(2) or 'GB'
# Convert MB to GB
if unit in ['MB', 'M']:
value = max(1, value // 1024) # Convert MB to GB, minimum 1GB
return value
return 0
async def analyze_multiple_games(self, game_names: List[str], use_cache: bool = True) -> Dict[str, Optional[CanRunResult]]:
"""Analyze multiple games and convert the results to a dictionary format expected by tests.
Args:
game_names: List of game names to analyze
use_cache: Whether to use cached results
Returns:
Dictionary containing compatibility and performance analysis in the format expected by tests
"""
results = {}
for game_name in game_names:
try:
result = await self.check_game_compatibility(game_name, use_cache)
results[game_name] = result
except Exception as e:
self.logger.error(f"Failed to analyze {game_name}: {e}")
results[game_name] = None
# Return the dictionary of results
return results
async def get_system_info(self) -> Dict[str, Any]:
"""Get comprehensive system information."""
hardware_specs = await self._get_hardware_specs()
return {
'cpu': {
'name': hardware_specs.cpu_name,
'cores': hardware_specs.cpu_cores,
'threads': hardware_specs.cpu_threads
},
'gpu': {
'name': hardware_specs.gpu_model,
'vram_gb': hardware_specs.gpu_vram_gb,
'supports_rtx': hardware_specs.supports_rtx,
'supports_dlss': hardware_specs.supports_dlss,
'driver_version': hardware_specs.nvidia_driver_version
},
'memory': {
'total': hardware_specs.ram_gb
},
'system': {
'os': hardware_specs.os_name,
'directx': hardware_specs.directx_version
}
}
async def get_optimization_suggestions(self, game_name: str, settings: str, resolution: str) -> List[Dict[str, str]]:
"""Get optimization suggestions for specific game and settings."""
try:
# Get game requirements and hardware specs
hardware_specs = await self._get_hardware_specs()
game_requirements = await self._fetch_game_requirements(game_name)
if not game_requirements:
return [{'type': 'error', 'description': f'Game requirements not found for {game_name}'}]
# Analyze compatibility to get recommendations
compatibility_analysis = await self._analyze_compatibility(
game_name, hardware_specs, game_requirements
)
# Convert recommendations to optimization format
optimizations = []
for rec in compatibility_analysis.recommendations:
optimizations.append({
'type': 'settings',
'description': rec
})
# Add resolution-specific optimizations
if resolution == '4K':
optimizations.append({
'type': 'resolution',
'description': 'Consider using DLSS Quality mode for better 4K performance'
})
elif resolution == '1440p':
optimizations.append({
'type': 'resolution',
'description': 'DLSS Balanced mode recommended for optimal 1440p experience'
})
# Add RTX-specific optimizations
if hardware_specs.supports_rtx and game_requirements.supports_rtx:
optimizations.append({
'type': 'rtx',
'description': 'Enable RTX ray tracing for enhanced visual quality'
})
return optimizations
except Exception as e:
self.logger.error(f"Failed to get optimization suggestions: {e}")
return [{'type': 'error', 'description': str(e)}]
async def analyze_game_compatibility(self, game_name: str, settings: str = "Medium", resolution: str = "System Default") -> Optional[Dict[str, Any]]:
"""Legacy method for backward compatibility with tests."""
try:
result = await self.check_game_compatibility(game_name)
if not result:
return None
# Check if result is already a dictionary (from cache) or CanRunResult object
if isinstance(result, dict):
# Result is already in dictionary format from cache
return result
# Use LLM to estimate missing values intelligently
llm_estimates = {}
if self.llm_analyzer:
try:
# Get LLM estimates for component scores and performance metrics
llm_estimates = await self.llm_analyzer.estimate_compatibility_metrics(
game_name,
result.hardware_specs,
result.compatibility_analysis,
result.performance_prediction
)
except Exception as e:
self.logger.warning(f"LLM estimation failed, using fallback: {e}")
# Convert CanRunResult to dictionary format with LLM estimates
return {
'compatibility': {
'compatibility_level': result.compatibility_analysis.overall_compatibility,
'overall_score': result.compatibility_analysis.overall_score,
'bottlenecks': result.compatibility_analysis.bottlenecks,
'component_analysis': {
'cpu': {
'status': next(('Excellent' if comp.meets_recommended else 'Good' if comp.meets_minimum else 'Poor'
for comp in result.compatibility_analysis.component_analyses
if comp.component.name.lower() == 'cpu'), 'Unknown'),
'score': llm_estimates.get('cpu_score', next((int(comp.score * 100)
for comp in result.compatibility_analysis.component_analyses
if comp.component.name.lower() == 'cpu'), 75))
},
'gpu': {
'status': next(('Excellent' if comp.meets_recommended else 'Good' if comp.meets_minimum else 'Poor'
for comp in result.compatibility_analysis.component_analyses
if comp.component.name.lower() == 'gpu'), 'Unknown'),
'score': llm_estimates.get('gpu_score', 80)
},
'memory': {
'status': next(('Excellent' if comp.meets_recommended else 'Good' if comp.meets_minimum else 'Poor'
for comp in result.compatibility_analysis.component_analyses
if comp.component.name.lower() == 'ram'), 'Unknown'),
'score': llm_estimates.get('memory_score', 85)
},
'storage': {
'status': next(('Excellent' if comp.meets_recommended else 'Good' if comp.meets_minimum else 'Poor'
for comp in result.compatibility_analysis.component_analyses
if comp.component.name.lower() == 'storage'), 'Unknown'),
'score': llm_estimates.get('storage_score', 90)
}
}
},
'performance': {
'fps': result.performance_prediction.expected_fps if hasattr(result.performance_prediction, 'expected_fps') else 0,
'performance_level': result.performance_prediction.tier.value if hasattr(result.performance_prediction, 'tier') else 'Unknown',
'stability': llm_estimates.get('stability', 'stable'),
'optimization_suggestions': result.performance_prediction.upgrade_suggestions if hasattr(result.performance_prediction, 'upgrade_suggestions') else []
},
'optimization_suggestions': result.performance_prediction.upgrade_suggestions if hasattr(result.performance_prediction, 'upgrade_suggestions') else [],
'hardware_analysis': {
'gpu_tier': llm_estimates.get('gpu_tier', 'high-end'),
'bottleneck_analysis': result.compatibility_analysis.bottlenecks
}
}
except Exception as e:
self.logger.error(f"Legacy compatibility analysis failed: {e}")
return None
def _parse_ram_value(self, ram_str: str) -> int:
"""Parse RAM value from string to integer GB with proper unit handling."""
if not ram_str or ram_str == "Unknown":
return 0
# Convert to uppercase for consistency
ram_str = str(ram_str).upper()
# Check if explicitly specified as MB
if 'MB' in ram_str:
# Extract number
mb_match = re.search(r'(\d+\.?\d*)\s*MB', ram_str)
if mb_match:
# Convert MB to GB (rounded up to 0.5 GB minimum for values under 512MB)
mb_value = float(mb_match.group(1))
if mb_value < 512:
return 0.5 # Minimum 0.5GB for small values
else:
return max(1, int(mb_value / 1024)) # Convert MB to GB, minimum 1GB
# Default GB matching - more flexible pattern to match various formats
gb_match = re.search(r'(\d+\.?\d*)\s*G?B?', ram_str)
if gb_match:
return int(float(gb_match.group(1)))
# Last resort fallback - just try to extract any number
number_match = re.search(r'(\d+\.?\d*)', ram_str)
if number_match:
return int(float(number_match.group(1)))
return 0