为何libclang无法找到函数参数的声明？

我正在开发一款用于查找C函数依赖关系的工具。比如在tcpdump项目的print-ppp.c文件中有这样一个函数：

static void ppp_hdlc(netdissect_options *ndo, const u_char *p, int length)

其中netdissect_options定义在netdissect.h头文件里，但用libclang查找该头文件中对应代码时失败了。运行自己写的代码后，definition.spelling、definition.location.file、definition.location.line的值都为空。

import sys
import clang
from clang.cindex import CursorKind
import json
import os

def load_compile_commands(compile_commands_file):
    with open(compile_commands_file, "r") as f:
        compile_commands = json.load(f)
    return compile_commands


def get_translation_unit(compile_commands, source_file):
    if clang.cindex.Config.library_path is None:
        clang.cindex.Config.set_library_file("/opt/homebrew/opt/llvm/lib/libclang.dylib")

    index = clang.cindex.Index.create()
    for command in compile_commands:
        if command["file"] == source_file:
            args = command["arguments"][1:-2]
            args.append("-I" + os.path.dirname(source_file))
            args.append("-I/opt/homebrew/opt/llvm/lib/clang/16/include/")
            
            return index.parse(source_file, args)
    raise Exception(f"No translation unit found for source file {source_file}")


def function_node(cursor, function_name):

    for child in cursor.get_children():
        if (
            child.kind == clang.cindex.CursorKind.FUNCTION_DECL
            and child.spelling == function_name
        ):
            return child
        
    return None

def get_function_arguments(cursor):
    if cursor.kind != clang.cindex.CursorKind.FUNCTION_DECL:
        raise ValueError("Cursor must point to a function declaration")

    arguments = []
    for child in cursor.get_children():
        if child.kind == clang.cindex.CursorKind.PARM_DECL:
            arguments.append(child)

    return arguments


def find_arg_declatation(project_root:str, file_name: str, function:str):
    

    compile_commands_file = project_root + "/compile_commands.json"
    source_file = project_root +"/"+ file_name

    compile_commands = load_compile_commands(compile_commands_file)
    tu = get_translation_unit(compile_commands, source_file)

    for diagnostic in tu.diagnostics:
        print(diagnostic)
        
    function = function_node(tu.cursor, function_name)
    declarations = []
    types = []
    if function is not None:
        arguments = get_function_arguments(function)
        print(f"Nodes dependent on function '{function_name}':")
        for argument in arguments:
            definition = argument.type.get_declaration()
            print(definition.spelling, definition.location.file, definition.location.line)
            
            
if __name__ == "__main__":
    find_arg_declatation("<args go here>")

1. 编译参数截断问题 #

原代码中args = command["arguments"][1:-2]会随意截断原始编译参数，可能丢失tcpdump项目本身的头文件路径配置，导致libclang找不到netdissect.h。应该保留完整编译参数，仅移除源文件自身的参数：

# 替换原args处理逻辑
args = command["arguments"].copy()
# 移除末尾的源文件参数（index.parse会自动传入source_file）
if args[-1] == source_file:
    args.pop()
# 补充必要头路径
args.append("-I" + os.path.dirname(source_file))
args.append("-I/opt/homebrew/opt/llvm/lib/clang/16/include/")

2. 函数查找逻辑局限性 #

当前function_node仅遍历直接子节点，无法找到嵌套在其他作用域里的函数，改成递归遍历：

def function_node(cursor, function_name):
    if (cursor.kind == clang.cindex.CursorKind.FUNCTION_DECL
        and cursor.spelling == function_name):
        return cursor
    for child in cursor.get_children():
        result = function_node(child, function_name)
        if result is not None:
            return result
    return None

3. 指针类型解析问题 #

argument.type返回的是指针类型，而非原始结构体类型，需要先解引用：

for argument in arguments:
    arg_type = argument.type
    # 解引用指针获取底层类型
    if arg_type.kind == clang.cindex.TypeKind.POINTER:
        arg_type = arg_type.get_pointee()
    definition = arg_type.get_declaration()
    # 检查声明有效性
    if definition.spelling and definition.location.file:
        print(f"类型: {definition.spelling}, 文件: {definition.location.file}, 行号: {definition.location.line}")
    else:
        print(f"无法获取参数类型的声明: {arg_type.spelling}")

4. 脚本入口语法错误 #

原代码中if __name__ == "main":是错误写法，必须改为if __name__ == "__main__":，否则脚本无法正确执行。

5. 诊断信息排查 #

务必关注tu.diagnostics输出的警告或错误，比如头文件未找到、语法错误等，这些都会直接导致libclang无法解析类型声明。

内容的提问来源于stack exchange，提问作者Brian