"""LLM-based fallback analyzer for languages without native linters.""" import json from groq import Groq from app.modules.analyzer_base import CodeAnalyzer, AnalysisResult from app.config import GROQ_API_KEY, MODEL_NAME, MAX_TOKENS, TEMPERATURE class LLMFallbackAnalyzer(CodeAnalyzer): def __init__(self, language: str): super().__init__(language) self._client = None def _get_client(self) -> Groq: if self._client is None: self._client = Groq(api_key=GROQ_API_KEY) return self._client def analyze(self, code: str) -> AnalysisResult: try: client = self._get_client() prompt = f"""Analyze this {self.language} code for issues and return ONLY a JSON array with objects containing: - line: estimated line number (1-based) - symbol: issue code/identifier - message: issue description - severity: 'high', 'medium', or 'low' - category: 'correctness', 'performance', 'security', or 'style' Limit to top 5 issues. Return ONLY valid JSON array, no markdown. Code: {code} JSON array:""" resp = client.chat.completions.create( model=MODEL_NAME, messages=[{"role": "user", "content": prompt}], max_tokens=MAX_TOKENS, temperature=TEMPERATURE, ) response_text = resp.choices[0].message.content.strip() try: start = response_text.find("[") end = response_text.rfind("]") + 1 if start >= 0 and end > start: issues_data = json.loads(response_text[start:end]) else: issues_data = [] except json.JSONDecodeError: return AnalysisResult(self.language, error="Failed to parse LLM response") issues = [] for issue in issues_data[:5]: formatted = self.format_issue( line=issue.get("line", 0), symbol=issue.get("symbol", "llm-issue"), message=issue.get("message", ""), category=issue.get("category", "style"), severity=issue.get("severity", "low"), ) issues.append(formatted) return AnalysisResult(self.language, issues) except Exception as e: return AnalysisResult(self.language, error=str(e))