Python Ctypes - String to c_char_p to c_void

I'm being unable to create certain C data structures in Python using Ctypes and then read them back properly. Consider the following:

Setup:

from ctypes import Structure, c_void_p, c_char_p, POINTER, cdll, cast

class GSLIST(Structure):
    """GLib's GSList."""
    pass

GSLIST._fields_ = [
    ('data', c_void_p),
    ('next', POINTER(GSLIST))
]

lib = cdll.LoadLibrary('/path/to/mylib.o')

lib.g_slist_prepend.argtypes = [POINTER(GSLIST), c_char_p]
lib.g_slist_prepend.restype = POINTER(GSLIST)

Example:

py_str_in = bytes('hi', 'ascii')

agslist_ptr = POINTER(GSLIST)()  # NULL
agslist_ptr = lib.g_slist_prepend(agslist_ptr, c_char_p(py_str_in))

print(cast(agslist_ptr.contents.data, c_char_p).value.decode())
# 'hi'

The example may be a bit more complicated than it needs to be, but it's been hard to replicate at a smaller scale.

What the example does is to save the "hi" string in the data field of the first element of a GSList (the most common list type in the GLib library). This field is of type void* so that it can support whatever type. So, first we convert to a c_char_p to make it compatible with void*, and then to read it back we cast() it.

The example as shown works as expected. It breaks if we do instead:

agslist_ptr = lib.g_slist_prepend(agslist_ptr, c_char_p(bytes('hi', 'ascii')))

Simply skipping the assignment to the py_str_in variable results in just '' for an output.

Also, if instead of 'hi' I use any single character, it also works fine.

I can't really explain what is going on and the documentation doesn't go into too much depth. If I had to guess, something with string literals or garbage collection. But it could also be just buggy code.

Thanks for any insight.

CodePudding user response：

The 2nd instance that fails is due to c_char_p(bytes('hi', 'ascii') being the only reference to the object. Python doesn't know that C saved that pointer, so after that line executes the object is destroyed and the pointer that C saved is now invalid. Referencing it is undefined behavior.

In the 1st case the py_str_in reference still exists so the pointer C saved is still valid.