[libclang] Python binding detects parsed function's arguments incorrectly

[0x4b6579]

Hi everyone,

I'm trying to parse some header files to extract signatures using clang.cindex in Python. Sometimes it detects types as names wrongly.

The following code from musl is an example:

ssize_t recvfrom (int, void *__restrict, size_t, int, struct sockaddr *__restrict, socklen_t *__restrict);

from: https://git.musl-libc.org/cgit/musl/tree/include/sys/socket.h#n397

And I get size_t as the name of the next argument! ->

ssize_t recvfrom(int , void *restrict , int size_t, int , struct sockaddr *restrict , socklen_t *restrict )

My code:

import sys
import clang.cindex


header_file = sys.argv[1]

index = clang.cindex.Index.create()
tu = index.parse(header_file)

for cursor in tu.cursor.get_children():
    if cursor.kind == clang.cindex.CursorKind.FUNCTION_DECL:

        print(cursor.result_type.spelling, cursor.spelling, end='(')

        is_first_arg = False
        for arg in cursor.get_arguments():
            if not is_first_arg:
                is_first_arg = True
            else:
                print(', ', end='')

            print(arg.type.spelling, arg.spelling, end='')

        print(')')

My command:

python poc.py musl-1.2.5/include/sys/socket.h

The output:

__uint16_t __bswap_16(__uint16_t __bsx)
__uint32_t __bswap_32(__uint32_t __bsx)
__uint64_t __bswap_64(__uint64_t __bsx)
__uint16_t __uint16_identity(__uint16_t __x)
__uint32_t __uint32_identity(__uint32_t __x)
__uint64_t __uint64_identity(__uint64_t __x)
int select(int __nfds, fd_set *restrict __readfds, fd_set *restrict __writefds, fd_set *restrict __exceptfds, struct timeval *restrict __timeout)
int pselect(int __nfds, fd_set *restrict __readfds, fd_set *restrict __writefds, fd_set *restrict __exceptfds, const struct timespec *restrict __timeout, const __sigset_t *restrict __sigmask)
struct cmsghdr * __cmsg_nxthdr(struct msghdr * __mhdr, struct cmsghdr * __cmsg)
int socket(int , int , int )
int socketpair(int , int , int , int[2] )
int shutdown(int , int )
int bind(int , const struct sockaddr * , socklen_t )
int connect(int , const struct sockaddr * , socklen_t )
int listen(int , int )
int accept(int , struct sockaddr *restrict , socklen_t *restrict )
int accept4(int , struct sockaddr *restrict , socklen_t *restrict , int )
int getsockname(int , struct sockaddr *restrict , socklen_t *restrict )
int getpeername(int , struct sockaddr *restrict , socklen_t *restrict )
ssize_t send(int , const void * , int size_t, int )
ssize_t recv(int , void * , int size_t, int )
ssize_t sendto(int , const void * , int size_t, int , const struct sockaddr * , socklen_t )
ssize_t recvfrom(int , void *restrict , int size_t, int , struct sockaddr *restrict , socklen_t *restrict )
ssize_t sendmsg(int , const struct msghdr * , int )
ssize_t recvmsg(int , struct msghdr * , int )
int getsockopt(int , int , int , void *restrict , socklen_t *restrict )
int setsockopt(int , int , int , const void * , socklen_t )
int sockatmark(int )

The result of clang looks to be fine when I try this command:

clang -Xclang -ast-dump=json -fsyntax-only include/sys/socket.h

The issue seems to be in clang_getCursorSpelling() in clang/tools/libclang/CIndex.cpp not the Python binding.