AscendGYM / eval /bare_harness /Backend /utils /error2func_new.py
SLAI-AITP's picture
Add files using upload-large-folder tool
ef881ce verified
Raw History Blame Contribute Delete
11.5 kB
import re
import os
# 定义C++关键字集合,用于过滤无效函数名
CPP_KEYWORDS = {
'if', 'else', 'for', 'while', 'switch', 'case', 'return', 'break', 'continue',
'try', 'catch', 'throw', 'class', 'struct', 'union', 'enum', 'namespace', 'template',
'typename', 'static', 'const', 'virtual', 'inline', 'explicit', 'friend', 'public',
'private', 'protected', 'new', 'delete', 'operator', 'sizeof', 'typedef', 'typeid',
'using', 'alignas', 'alignof', 'and', 'and_eq', 'asm', 'auto', 'bitand', 'bitor',
'bool', 'char', 'char8_t', 'char16_t', 'char32_t', 'compl', 'concept', 'const_cast',
'constexpr', 'decltype', 'default', 'do', 'double', 'dynamic_cast', 'false', 'float',
'goto', 'int', 'long', 'mutable', 'noexcept', 'not', 'not_eq', 'nullptr', 'or', 'or_eq',
'register', 'reinterpret_cast', 'requires', 'short', 'signed', 'static_assert',
'static_cast', 'thread_local', 'true', 'typedef', 'typeid', 'typename', 'union',
'unsigned', 'void', 'volatile', 'wchar_t', 'xor', 'xor_eq'
}
def find_cpp_function_name_at_line(file_content, error_line_num):
"""
从C++源代码中查找指定行号所在的函数名
特别处理类成员函数的识别
Args:
file_content: 文件的完整内容
error_line_num: 出错的行号(1-based)
Returns:
函数名,如果找不到返回None
"""
lines = file_content.split('\n')
# 1. 首先验证行号是否有效
if error_line_num < 1 or error_line_num > len(lines):
print(f"警告:行号 {error_line_num} 超出文件范围 (1-{len(lines)})")
return None
# 2. 从错误行向上搜索,寻找函数定义
current_line_idx = error_line_num - 1 # 转换为0-based索引
# 3. 改进的函数签名识别逻辑
# 先构建一个包含错误行附近上下文的代码段
context_start = max(0, current_line_idx - 100) # 向上搜索100行
context_lines = lines[context_start:current_line_idx + 10] # 包含错误行后10行以便识别函数结束
# 尝试1:精确匹配类成员函数定义(处理多行情况)
# 构建完整的上下文字符串
context_text = '\n'.join(context_lines)
# 模式1:匹配类成员函数定义(返回类型 类名::函数名(...) {...})
# 这个模式更宽松,可以处理跨多行的函数签名
member_func_pattern = r'([\w:<>]+\s+[\w\d_:~]+\s*::\s*[\w\d_:~]+)\s*\([^)]*\)\s*(?:const\s*)?(?:->\s*[\w:<>]+)?\s*\{'
matches = list(re.finditer(member_func_pattern, context_text, re.DOTALL))
if matches:
# 取最接近错误行的匹配
best_match = None
best_distance = float('inf')
for match in matches:
# 计算匹配位置到错误行的距离
match_line_num = context_start + context_text[:match.start()].count('\n') + 1
distance = abs(error_line_num - match_line_num)
if distance < best_distance:
best_distance = distance
best_match = match
if best_match:
full_func_name = best_match.group(1).strip()
# 提取函数名部分(最后一个::之后的部分)
if '::' in full_func_name and not full_func_name.startswith('~'):
func_name = full_func_name.split('::')[-1]
# 验证函数名是否有效
if func_name and func_name not in CPP_KEYWORDS and len(func_name) > 1:
return func_name
# 模式2:匹配普通函数定义
func_pattern = r'([\w:<>]+\s+[\w\d_:~]+)\s*\([^)]*\)\s*(?:const\s*)?(?:->\s*[\w:<>]+)?\s*\{'
matches = list(re.finditer(func_pattern, context_text, re.DOTALL))
if matches:
# 取最接近错误行的匹配
best_match = None
best_distance = float('inf')
for match in matches:
# 计算匹配位置到错误行的距离
match_line_num = context_start + context_text[:match.start()].count('\n') + 1
distance = abs(error_line_num - match_line_num)
if distance < best_distance:
best_distance = distance
best_match = match
if best_match:
full_func_name = best_match.group(1).strip()
# 验证函数名是否有效
if full_func_name and full_func_name not in CPP_KEYWORDS and len(full_func_name) > 1:
return full_func_name
# 尝试3:精确的逐行搜索,更灵活的模式
search_limit = max(0, current_line_idx - 200)
best_candidate = None
best_distance = float('inf')
for line_idx in range(current_line_idx, search_limit - 1, -1):
line = lines[line_idx].strip()
# 跳过空行和注释
if not line or line.startswith('//') or line.startswith('/*') or line.startswith('*') or line.startswith('#'):
continue
# 跳过明显不是函数定义的控制结构
control_keywords = ['if', 'else', 'for', 'while', 'switch', 'case', 'return', 'break', 'continue', 'try', 'catch']
if any(line.startswith(kw) for kw in control_keywords):
continue
# 检查是否可能是函数定义
# 模式1:类成员函数
member_match = re.search(r'([\w:<>]+\s+)?([\w\d_:~]+::[\w\d_:~]+)\s*\([^)]*\)\s*(?:const\s*)?(?:->\s*[\w:<>]+)?\s*\{?', line)
if member_match:
full_name = member_match.group(2).strip()
if '::' in full_name and not full_name.startswith('~'):
func_name = full_name.split('::')[-1]
else:
func_name = full_name
if func_name and func_name not in CPP_KEYWORDS and len(func_name) > 1:
distance = current_line_idx - line_idx
if distance < best_distance:
best_distance = distance
best_candidate = func_name
# 模式2:普通函数
func_match = re.search(r'([\w:<>]+\s+)?([\w\d_:~]+)\s*\([^)]*\)\s*(?:const\s*)?(?:->\s*[\w:<>]+)?\s*\{?', line)
if func_match:
func_name = func_match.group(2).strip()
if func_name and func_name not in CPP_KEYWORDS and len(func_name) > 1:
distance = current_line_idx - line_idx
if distance < best_distance:
best_distance = distance
best_candidate = func_name
if best_candidate:
return best_candidate
# 尝试4:最后的尝试 - 检查整个文件内容,寻找函数定义
full_text = file_content
# 查找错误行号附近的函数开始位置
error_pos = 0
for i in range(error_line_num - 1):
error_pos = full_text.find('\n', error_pos) + 1
# 向上搜索,寻找最近的函数开始
search_pos = error_pos - 1
function_start = None
function_name = None
while search_pos > 0:
# 寻找可能的函数定义
match = re.search(r'([\w:<>]+\s+)?([\w\d_:~]+::[\w\d_:~]+|\w+)\s*\([^)]*\)\s*(?:const\s*)?(?:->\s*[\w:<>]+)?\s*\{',
full_text[max(0, search_pos-200):search_pos+200])
if match:
if '::' in match.group(2) and not match.group(2).startswith('~'):
function_name = match.group(2).split('::')[-1]
else:
function_name = match.group(2)
if function_name and function_name not in CPP_KEYWORDS and len(function_name) > 1:
return function_name
search_pos -= 1
return None
def locate_cpp_error_function(error_detail):
"""
从C++错误详情中定位出错的函数
Args:
error_detail: 包含错误信息的字符串
Returns:
(file_name, function_name) 元组,或 (None, None) 如果无法定位
"""
# 1. 从错误信息中提取文件路径和行号
file_match = re.search(r'([\w\d_./\\-]+\.cpp):(\d+):', error_detail)
if not file_match:
print("无法从错误信息中提取文件路径和行号")
return None, None
file_path = file_match.group(1)
line_number = int(file_match.group(2))
file_name = os.path.basename(file_path)
# 2. 检查文件是否存在
if not os.path.exists(file_path):
print(f"警告:文件不存在: {file_path}")
# 尝试在当前目录或常见位置查找
possible_paths = [
file_name,
os.path.join('op_host', file_name),
os.path.join('src', file_name),
os.path.join('src', 'op_host', file_name)
]
found_path = None
for path in possible_paths:
if os.path.exists(path):
found_path = path
break
if found_path:
file_path = found_path
print(f"在替代位置找到文件: {file_path}")
else:
print("无法找到源代码文件,无法定位函数")
return file_name, None
# 3. 读取文件内容
try:
with open(file_path, 'r', encoding='utf-8') as f:
file_content = f.read()
except Exception as e:
print(f"读取文件失败: {e}")
return file_name, None
# 4. 查找函数名
function_name = find_cpp_function_name_at_line(file_content, line_number)
# 5. 后处理:对函数名进行清理
if function_name:
# 移除可能的析构函数标记
function_name = re.sub(r'^~', '', function_name)
# 移除可能的参数列表残留
function_name = re.sub(r'\s*\(.*$', '', function_name)
# 移除可能的返回类型残留
function_name = function_name.split()[-1]
if function_name:
return file_name, function_name
else:
print("无法定位到具体函数,使用默认名称")
return file_name, "unknown_function"
if __name__ == "__main__":
# 测试用例
error_msg1 = "${ASCEND_GYM_DATA_ROOT}/models/pangu7b_new/ModelLink/ModelLink/user/GTSOpsPlus_0122/GTSOpsPlus/op-contrib/ops/built-in/ascendc/add_test_custom_test_new/op_host/add_test_custom_test_new_tiling_calculator.cpp:36:26: error: 'GetTypeSize' was not declared in this scope; did you mean 'GetShapeSize'?\n 36 | uint32_t dtypeSize = GetTypeSize(inputDesc0->GetDataType());\n | ^~~~~~~~~~~\n | GetShapeSize"
error_msg2 = "${ASCEND_GYM_DATA_ROOT}/models/pangu7b_new/ModelLink/ModelLink/user/GTSOpsPlus_0122/GTSOpsPlus/op-contrib/ops/built-in/ascendc/add_test_custom_test_new/op_host/add_test_custom_test_new_tiling_calculator.cpp:20:36: error: 'class optiling::AddTestCustomTestNewTilingCalculator' has no member named 'GetInputDesc'; did you mean 'GetInputTensor'?\n 20 | const auto* inputDesc0 = this->GetInputDesc(0);\n | ^~~~~~~~~~~~\n | GetInputTensor"
# 测试error_msg1
file1, func1 = locate_cpp_error_function(error_msg1)
print("\nerror_msg1结果:")
print(f"文件名: {file1}")
print(f"函数名: {func1}")
# 测试error_msg2
file2, func2 = locate_cpp_error_function(error_msg2)
print("\nerror_msg2结果:")
print(f"文件名: {file2}")
print(f"函数名: {func2}")