PySequence_Check is always true on instances?

[Tim Peters]

[Walter Moreira]
> Hi. If I define a simple class
>
>    >>> class eggs:
>    ...     pass
>    >>> a = eggs()
>
> and I call PySequence_Check on the instance 'a', at C level, it
> returns 1.

Right.  This is an unfortunate consequence of the "type/class split" before
2.2:  all classic classes fill in all type method slots "just in case", so
pass all tests that just look for the existence of non-NULL method slots:

int
PySequence_Check(PyObject *s)
{
	return s != NULL && s->ob_type->tp_as_sequence &&
		s->ob_type->tp_as_sequence->sq_item != NULL;
}

> If the class is a new style class:
>
>    >>> class eggs(object):
>    ...     pass
>    >>> b = eggs()
>
> then PySequence_Check on 'b' returns 0. I proved this on Python
> 2.2 and 2.1 (couldn't download 2.2.1 yet).

That won't change in 2.2.1.

All old-style instances return true for these; whether a new-style instance
returns true depends on whether it inherits or directly defines the type
slots being examined.

> Is this a bug?

Not in the sense you're hoping for <wink>.

> If it is not a bug, which is the standard way to check that an
> object is sequence, in C?

There isn't one:  it depends on exactly what *you* mean by "a sequence".
It's a fuzzy concept in Python, and you'll find that PySequence_Check() is
rarely used in the core.  In most cases for 2.2, it's not a question worth
even trying to answer:  most "sequence contexts" in 2.2 were generalized to
allow any iterable object, and the core usually determines whether an object
is iterable by seeing whether PyObject_GetIter() returns NULL or not.