I'm writing a C extension module and discovered that Py_CLEAR() causes a
crash if the programmer happened to name their variable "tmp".  The
Py_CLEAR() macro internally uses the name "tmp" in a new scope, hiding
the callers "tmp", and calling Py_DECREF() on an essentially random bit
of memory.

I suggest changing Py_CLEAR() to use something a little less common than
"tmp".  Perhaps "_py_tmp".

For easy reference, here's how Py_CLEAR() is defined now:

#define Py_CLEAR(op)				\
        do {                            	\
                if (op) {			\
                        PyObject *tmp = (PyObject *)(op);	\
                        (op) = NULL;		\
                        Py_DECREF(tmp);		\
                }				\
        } while (0)

A better option may be to append _tmp to the passed in token:

#define Py_CLEAR(op)				\
        do {                            	\
                if (op) {			\
                        PyObject *op##_tmp = (PyObject *)(op);	\
                        (op) = NULL;		\
                        Py_DECREF(op##_tmp);		\
                }				\
        } while (0)

Appending _tmp is a good idea, but it won't work when the parameter
isn't a simple symbol.  For example, there's a line in cPickle.c like
this: Py_CLEAR(*p).

Committed the fix r64927. Thanks.