PyObject_DEL() should not be used to destroy an object because it will break the linked list of allocated objects, list used in pydebug mode to detect bugs. cElementTree should use Py_DECREF() instead of PyObject_DEL() to destroy an objects.

Attached patch fixes that:
 * Replace PyObject_Del() by Py_DECREF()
 * Catch element_new_extra() errors
 * parser dealloc: replace Py_DECREF() by Py_XDECREF() because the pointer may be NULL (error in the constructor)
 * set all parser attributes to NULL at the beginning of the constructor to be able to call safetly the destructor
 * element_new(): define tag, text, tail attributes before calling element_new_extra() to be able to call the destructor
 * raise a MemoryError on element_new_extra() failure. element_new() didn't raise any error on element_new_extra() failure. Other functions just forget to catch element_new_extra() error.

See #3299 for the whole story.

Looks reasonable. Are there any tests that could be easily added to the test suite?

> Are there any tests that could be easily added to the test suite?

I don't know an easy way to simulate malloc failure. There is the http://www.nongnu.org/failmalloc/ library but I never used it, and I don't think that it's widely used. You can just test that the patch doesn't make things worse :-)