app/services/reviewer_service.py

import json
import logging
import re
from typing import List, Dict
from app.core.model_loader import llm_engine

logger = logging.getLogger(__name__)

class AIReviewerService:
    def __init__(self):
        pass

    def review_batch_code(self, files: list) -> list:
        results = []
        
        # Process in batches of 5 to stay within Gemini Free Tier limits (15 RPM)
        batch_size = 5
        
        for i in range(0, len(files), batch_size):
            batch = files[i : i + batch_size]
            combined_code = ""
            file_names = []
            
            for f in batch:
                # Minify code to save tokens
                raw_content = f.content or ""
                # Limit to 6k chars per file to stay within context window safely
                minified_content = self._minify_code(raw_content[:6000], f.fileName)
                
                combined_code += f"\n--- FILE: {f.fileName} ---\n{minified_content}\n"
                file_names.append(f.fileName)

            # Updated prompt to explicitly request suggestions and improvements
            prompt = f"""
Analyze the following {len(batch)} source code files for security vulnerabilities and code quality.
{combined_code}

Task:
1. Detect severe security/logic issues (Vulnerabilities).
2. If a vulnerability is found, provide a concise 'suggestion' on how to fix it.
3. If NO vulnerabilities are found in a file, provide a list of 'improvement_suggestions' (clean code, performance, or architecture tips).
4. Provide metrics for complexity and maintainability (scale 1-10).

Output a JSON array (exactly 1 object per file):
[
  {{
    "fileName": "exact/path/from/header",
    "vulnerabilities": [
      {{
        "type": "SQLi/Logic/etc",
        "line": 10,
        "description": "Short explanation",
        "suggestion": "Specific code fix" 
      }}
    ],
    "improvement_suggestions": ["Tip 1", "Tip 2"],
    "metrics": {{"complexity": 3, "maintainability": 8}}
  }}
]
"""
            try:
                # 8k output tokens for the batch analysis
                response_text = llm_engine.generate(prompt, max_tokens=8192)
                batch_results = self._parse_json(response_text)
                
                # Map results by fileName for easy lookup
                processed_map = {item.get('fileName'): item for item in batch_results if isinstance(item, dict)}
                
                for fn in file_names:
                    if fn in processed_map:
                        res = processed_map[fn]
                        
                        # Ensure all required keys exist to prevent Android serialization errors
                        res.setdefault("vulnerabilities", [])
                        res.setdefault("improvement_suggestions", [])
                        res.setdefault("metrics", {"complexity": 5, "maintainability": 5})
                        
                        # Ensure every vulnerability has a suggestion field
                        for vuln in res["vulnerabilities"]:
                            if "suggestion" not in vuln:
                                vuln["suggestion"] = "Review the implementation logic for improved safety."
                        
                        results.append(res)
                    else:
                        # Fallback if the AI skipped a file in its response
                        results.append({
                            "fileName": fn, 
                            "vulnerabilities": [], 
                            "improvement_suggestions": ["No immediate improvements identified."],
                            "metrics": {"complexity": 1, "maintainability": 10}
                        })
                        
            except Exception as e:
                logger.error(f"Batch processing error: {e}")
                for fn in file_names:
                    results.append({
                        "fileName": fn, 
                        "vulnerabilities": [], 
                        "improvement_suggestions": [], 
                        "metrics": {"complexity": 0, "maintainability": 0}
                    })

        return results

    def _minify_code(self, code: str, filename: str) -> str:
        """
        Removes comments, empty lines, and logs to optimize token usage.
        """
        lines = code.split('\n')
        cleaned_lines = []
        
        is_python = filename.endswith('.py')
        is_js_style = filename.endswith(('.js', '.ts', '.jsx', '.tsx', '.java', '.kt', '.c', '.cpp'))
        
        for line in lines:
            stripped = line.strip()
            
            if not stripped: continue
            
            # Skip comments
            if is_python and stripped.startswith('#'): continue
            if is_js_style and stripped.startswith(('//', '/*', '*')): continue
            
            # Skip common logging
            if any(log in stripped for log in ['console.log', 'print(', 'logger.', 'Log.d', 'Log.e']):
                continue
            
            # Strip inline comments
            if is_python and ' #' in line: line = line.split(' #', 1)[0]
            if is_js_style and ' //' in line: line = line.split(' //', 1)[0]
                
            if not line.strip(): continue

            cleaned_lines.append(line.rstrip())
            
        return '\n'.join(cleaned_lines)

    def _parse_json(self, text: str):
        """
        Cleans and parses the LLM response into a Python list/dict.
        """
        text = text.strip()
        if not text:
            return []
            
        # Clean markdown formatting if present
        text = re.sub(r"^[^[]*\[", "[", text) 
        text = re.sub(r"\][^]]*$", "]", text) 
        text = text.replace("```json", "").replace("```", "").strip()
        
        try:
            return json.loads(text)
        except json.JSONDecodeError as e:
            logger.warning(f"JSON Decode Error: {e}")
            return []

# Instantiate the service
service = AIReviewerService()