Sushyy/bug-explainer-ml
0
1# import json2 3# def analyze_code(language, code, tokenizer, model):4# messages = [5# {6# "role": "system",7# "content": (8# "You are a helpful and expert-level AI code reviewer and bug fixer. "9# "Your task is to analyze the given buggy code in the specified programming language, "10# "identify bugs (logical, syntax, runtime, etc.), and fix them. "11# "Return a JSON object with the following keys:\n\n"12# "1. 'bug_analysis': a list of objects, each containing:\n"13# " - 'line_number': the line number (approximate if needed)\n"14# " - 'error_message': a short name of the bug\n"15# " - 'explanation': short explanation of the problem\n"16# " - 'fix_suggestion': how to fix it\n"17# "2. 'corrected_code': the entire corrected code block.\n\n"18# "Respond with ONLY the raw JSON object, no extra commentary or markdown."19# )20# },21# {22# "role": "user",23# "content": f"💻 Language: {language}\n🐞 Buggy Code:\n```{language.lower()}\n{code.strip()}\n```"24# }25# ]26 27# inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)28# attention_mask = (inputs != tokenizer.pad_token_id).long()29 30# outputs = model.generate(31# inputs,32# attention_mask=attention_mask,33# max_new_tokens=1024,34# do_sample=False,35# pad_token_id=tokenizer.eos_token_id,36# eos_token_id=tokenizer.eos_token_id37# )38 39# response = tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True)40 41# # Try parsing response to JSON42# try:43# json_output = json.loads(response)44# return json_output45# except json.JSONDecodeError:46# print("⚠️ Could not decode response into JSON. Here's the raw output:\n")47# print(response)48# return None49# import json50# import logging51# import time52# import torch53 54# # Configure logging55# logger = logging.getLogger(__name__)56 57# def analyze_code(language, code, tokenizer, model):58# """59# Analyze code and return bug analysis with improved logging and error handling60# """61# start_time = time.time()62# logger.info(f"🔍 Starting analysis for {language} code ({len(code)} characters)")63 64# try:65# # Prepare messages66# messages = [67# {68# "role": "system",69# "content": (70# "You are a helpful and expert-level AI code reviewer and bug fixer. "71# "Your task is to analyze the given buggy code in the specified programming language, "72# "identify bugs (logical, syntax, runtime, etc.), and fix them. "73# "Return a JSON object with the following keys:\n\n"74# "1. 'bug_analysis': a list of objects, each containing:\n"75# " - 'line_number': the line number (approximate if needed)\n"76# " - 'error_message': a short name of the bug\n"77# " - 'explanation': short explanation of the problem\n"78# " - 'fix_suggestion': how to fix it\n"79# "2. 'corrected_code': the entire corrected code block.\n\n"80# "Respond with ONLY the raw JSON object, no extra commentary or markdown."81# )82# },83# {84# "role": "user",85# "content": f"💻 Language: {language}\n🐞 Buggy Code:\n```{language.lower()}\n{code.strip()}\n```"86# }87# ]88 89# logger.info("🔧 Applying chat template...")90# inputs = tokenizer.apply_chat_template(91# messages, 92# add_generation_prompt=True, 93# return_tensors="pt"94# ).to(model.device)95 96# attention_mask = (inputs != tokenizer.pad_token_id).long()97 98# logger.info(f"📏 Input length: {inputs.shape[1]} tokens")99# logger.info("🚀 Starting model generation...")100 101# generation_start = time.time()102 103# # Generate with more conservative settings104# with torch.no_grad(): # Ensure no gradients are computed105# outputs = model.generate(106# inputs,107# attention_mask=attention_mask,108# max_new_tokens=512, # Reduced from 1024 for faster inference109# do_sample=False,110# temperature=0.1, # Add temperature for more consistent output111# pad_token_id=tokenizer.eos_token_id,112# eos_token_id=tokenizer.eos_token_id,113# use_cache=True, # Enable KV cache for efficiency114# )115 116# generation_time = time.time() - generation_start117# logger.info(f"⚡ Generation completed in {generation_time:.2f} seconds")118 119# logger.info("📝 Decoding response...")120# response = tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True)121 122# logger.info(f"📄 Response length: {len(response)} characters")123# logger.info(f"🔍 First 100 chars: {response[:100]}...")124 125# # Try parsing response to JSON126# logger.info("🔍 Attempting to parse JSON...")127# try:128# # Clean up response - remove any markdown formatting129# cleaned_response = response.strip()130# if cleaned_response.startswith('```json'):131# cleaned_response = cleaned_response[7:]132# if cleaned_response.startswith('```'):133# cleaned_response = cleaned_response[3:]134# if cleaned_response.endswith('```'):135# cleaned_response = cleaned_response[:-3]136 137# cleaned_response = cleaned_response.strip()138 139# json_output = json.loads(cleaned_response)140 141# total_time = time.time() - start_time142# logger.info(f"✅ Analysis completed successfully in {total_time:.2f} seconds")143 144# # Validate the JSON structure145# if not isinstance(json_output, dict):146# raise ValueError("Response is not a dictionary")147 148# if 'bug_analysis' not in json_output:149# logger.warning("⚠️ Missing 'bug_analysis' key, adding empty list")150# json_output['bug_analysis'] = []151 152# if 'corrected_code' not in json_output:153# logger.warning("⚠️ Missing 'corrected_code' key, adding original code")154# json_output['corrected_code'] = code155 156# return json_output157 158# except json.JSONDecodeError as e:159# logger.error(f"❌ JSON decode error: {e}")160# logger.error(f"📄 Raw response: {repr(response)}")161 162# # Return a fallback structure with the raw response163# fallback_response = {164# "bug_analysis": [{165# "line_number": 1,166# "error_message": "Analysis parsing failed",167# "explanation": "The AI model returned a response that couldn't be parsed as JSON",168# "fix_suggestion": "Please try again or check the code format"169# }],170# "corrected_code": code,171# "raw_output": response,172# "parsing_error": str(e)173# }174 175# return fallback_response176 177# except Exception as e:178# total_time = time.time() - start_time179# logger.error(f"❌ Analysis failed after {total_time:.2f} seconds: {str(e)}")180# logger.error(f"💥 Exception type: {type(e).__name__}")181 182# # Return error response183# return {184# "bug_analysis": [{185# "line_number": 1,186# "error_message": "Analysis failed",187# "explanation": f"An error occurred during analysis: {str(e)}",188# "fix_suggestion": "Please try again or contact support"189# }],190# "corrected_code": code,191# "error": str(e),192# "error_type": type(e).__name__193# }194 195# analyzer.py196# analyzer.py197 198import torch199import json200import time201import logging202 203# Configure logger204logger = logging.getLogger("CodeAnalyzer")205logger.setLevel(logging.INFO)206handler = logging.StreamHandler()207formatter = logging.Formatter("[%(asctime)s] [%(levelname)s] - %(message)s")208handler.setFormatter(formatter)209logger.addHandler(handler)210 211def analyze_code(tokenizer, model, language, code):212 """213 Analyze and fix buggy code using CodeT5+ model with 'fix:' prompt prefix.214 Works across multiple programming languages.215 """216 start_time = time.time()217 218 # Prepare prompt in CodeT5+ style219 prompt = f"fix: {code.strip()}"220 221 logger.info(f"🔍 Starting analysis for language: {language}")222 logger.info(f"🧾 Prompt: {prompt[:80]}...")223 224 try:225 # Tokenize and generate response226 inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512).to(model.device)227 with torch.no_grad():228 output = model.generate(**inputs, max_new_tokens=1024)229 230 # Decode output231 response = tokenizer.decode(output[0], skip_special_tokens=True).strip()232 233 elapsed = round(time.time() - start_time, 2)234 logger.info(f"✅ Inference completed in {elapsed}s")235 236 return {237 "bug_analysis": [], # Optional: You could add heuristics here238 "corrected_code": response239 }240 241 except Exception as e:242 logger.error(f"❌ Error during analysis: {e}")243 return {244 "bug_analysis": [{245 "line_number": 0,246 "error_message": "Inference failed",247 "explanation": str(e),248 "fix_suggestion": "Try again with simpler code or retry later"249 }],250 "corrected_code": code251 }252