testcase-data/Ours/get_check_wrong_code.py

import sys
sys.path.append("/home/i-luoxianzhen/data/TestCase-Gen/methods/utils")
from dataset_all import get_datasets_by_name

import os
import uuid
import subprocess
import tempfile
import json


import re
import subprocess
import tempfile
import os
import resource
import uuid
import json

import random
from decimal import Decimal

import re

def is_decimal(s):
    try:
        a = float(s)
    except:
        return False
        
    return bool(re.match(r"^-?\d+\.\d+$", s))


def run_cpp_code_linux(code, input_string, std, memory_limit, time_limit):
    res = {}
    with tempfile.TemporaryDirectory() as tmpdirname:
        unique_id = uuid.uuid4()
        cpp_file = os.path.join(tmpdirname, f"{unique_id}.cpp")
        exe_file = os.path.join(tmpdirname, f"{unique_id}.out")

        # Write C++ code to file
        with open(cpp_file, "w") as f:
            f.write(code)

        # Compile the C++ code
        compile_result = subprocess.run(
            ["g++", cpp_file, "-o", exe_file, f"-std={std}"],
            capture_output=True,
            text=True
        )

        if compile_result.returncode != 0:
            res["status"] = "CE"
            res["details"] = compile_result.stderr
            return res

        memory_kb = int(memory_limit) * 1024
        time_limit_int = int(time_limit) // 1000 + 1
        cmd = f"ulimit -t {time_limit_int} && ulimit -v {memory_kb} && {exe_file}"
        error = ""
        try:
            result = subprocess.run(
                    cmd,
                    input=input_string,
                    text=True,
                    capture_output=True,
                    shell=True,
                    timeout=time_limit_int
                )
            
            # 检查返回码
            if result.returncode != 0:
                if result.returncode == 137:  # SIGKILL - 通常是内存超限
                    error = "MLE"
                elif result.returncode == 124:  # timeout命令的超时返回码
                    error = "TLE"
                else:
                    error = "RE"
            if not error and result.stderr:
                error = "RE"
            
            actual_lines = [line.strip() for line in result.stdout.splitlines()]
            actual_lines = [line for line in actual_lines if line]
            actual_lines = (" ".join(actual_lines)).strip()

            ### 小数保留 6 位
            if is_decimal(actual_lines):
                actual_lines = float(actual_lines)
            res['stdout'] = actual_lines
            res['status'] = "AC"
        except subprocess.TimeoutExpired:
            error = "TLE"
        except Exception as e:
            error = "RE"     

        if error:
            res['status'] = error   
            res['stdout'] = ""

        return res

import json
new_wrong_code = json.load(open("/home/i-luoxianzhen/data/TestCase-Gen/data/Ours/balance_v1.0.dev_with_code.json", "r", encoding="utf-8"))

compileAndRunOptions = {
    "compiler": "g++",
    "std": "c++11",
    "O": "2",
    "m": "x32"
}

save_dict = {}
i = 0

for k, v in new_wrong_code.items():

    print(f"idx: {i} Start")
    i+=1

    execute_status = []

    wrong_codes = v["cowrong_codes"]
    if len(v['sample']) >= 1:
        input_string = v['sample'][0]['inputData']
        output_string = v['sample'][0]['outputData']
    else:
        continue

    pass_code_list = []
    for item in wrong_codes:
        
        result = run_cpp_code_linux(item['code'], input_string, item['compileAndRunOptions']['std'], v['memoryLimit'], v['timeLimit'])
        pass_code_list.append(result['status'])
        
    save_dict[k] = pass_code_list
    break
    
json.dump(save_dict, open(f"/home/i-luoxianzhen/data/TestCase-Gen/data/Ours/wrong_code_check.json", "w", encoding="utf-8"), indent=4, ensure_ascii=False)