The documentation for PyBytes_FromObject states:

.. cfunction:: PyObject* PyBytes_FromObject(PyObject *o)

   Return the bytes representation of object *o* that implements
   the buffer protocol.

However, there exists a special-case for integer object in
PyBytes_FromObject that makes the function return a null-initialized
bytes object. Currently, this is only used for handling `bytes(10)'.

I don't like changing APIs after a stable release, but I believe this
behaviour is error-prone and surprising (and darn right annoying even).
So, I believe the special-case should be made specific to the bytes
constructor.

Thus, here is the fine patch.