Add support for __getattr__ (#36)

[stevenfontanella]

I was working on __setattr__ as well, but for some reason I get

  File "/home/steven/code/fypy/forbiddenfruit/tests/unit/test_forbidden_fruit.py", line 357, in test_overriding_setattr
    my_dict.abc = "xyz"
SystemError: error return without exception set

when I run the following test:

@skip_legacy
def test_overriding_setattr():
    """TODO: docstring"""
    def setter(self, x, val):
        super(tuple, self).__setattr__(x, val)
    curse(tuple, "__setattr__", setter)

    tup = ()
    tup.abc = "xyz"
    assert(tup.abc == "xyz")

The code I added for that was:

PyTypeObject._fields_ = [
# ...
    ('tp_setattr', ctypes.CFUNCTYPE(PyObject_p, PyObject_p, PyObject_p, PyObject_p))
]
# ...
override_dict['__setattr__'] = ('tp_setattr', "tp_setattr")

[alexchandel]

You're using the deprecated name, use tp_setattro.

And your signature is wrong. The signature documented here conflicts with here. The question is resolved by the cpython header, where an int (ctypes.c_int) is returned. Never assume, always check the headers.

[stevenfontanella]

Gotcha, I'll give it another look. Thanks for the quick answer!

[stevenfontanella]

Even after changing the name and signature, I get the same error. Is it possible that there's not a good way to implement __setattr__ patching? __setattro__ should return a c_int, but all we can do from Python is return a pointer to c_int (or maybe just a PyObject_p, I'm not entirely sure). Because I'm always returning a non-null pointer, the value is always non-zero, so we get SystemError: error return without exception set when __setattr__ returns. Does that sound plausible?

[alexchandel]

but all we can do from Python is return a pointer to c_int

No. You should return a c_int, not a pointer to anything. Look a few lines up at the types of printfunc and setattrfunc. It's throwing SystemError because (at the very least) your signature is wrong. Instead:

    ('tp_str', ctypes.CFUNCTYPE(PyObject_p, PyObject_p)),
    ('tp_getattro', ctypes.CFUNCTYPE(PyObject_p, PyObject_p, PyObject_p)),
    ('tp_setattro', ctypes.CFUNCTYPE(ctypes.c_int, PyObject_p, PyObject_p, PyObject_p)),
    ('tp_as_buffer', ctypes.c_void_p),
    ('tp_flags', ctypes.c_ulong),
    # ...

[stevenfontanella]

Right, I changed the signature. My point was that when I would like to override __setattr__ in python, I would need to write a function that matches the signature, i.e. returning a c_int. If you look at my commit here, I've done that, however I still get the same error in test_overriding_setattr. I think it's because even though the setter function returns a c_int, this is still a boxed type in python, so it's not the same as returning an int from a C function. In other words it looks like it's returning a c_int but it's really returning a PyObject_p. My question is: is there a way to write a python function that satisfies this signature and returns a c_int instead?

[7UR7L3]

I got the same errors @stevenfontanella mentioned with c_int, but I found that ('tp_setattro', ctypes.CFUNCTYPE(ctypes.c_char_p, PyObject_p, PyObject_p, PyObject_p)) worked for at least dict.__setattr__ without even bothering with a return somehow, in case anybody stumbles across this like me. It was only happy when I replaced __setattr__ before replacing __getattr__ if that means anything.

@ctypes.CFUNCTYPE(ctypes.c_char_p, PyObject_p, PyObject_p, PyObject_p)
def dict_setattr_cfunc(dict, field, val):
    dict[field] = val;

tp_func_dict[(dict, "__setattr__")] = dict_setattr_cfunc;
PyTypeObject.from_address(id(dict)).tp_setattro = dict_setattr_cfunc;