* Reduce Integral to just a single additional abstract method.
* Supply the binary operations as mixins instead of abstract methods.
* Three argument __pow__ is now optional.
* Convert with __int__ instead of __long__.

Is the patch good to go?

Not really, see the reply I sent to the checkin mail.

Not much feedback on comp.lang.python.

Does numbers3.diff look like what you had in mind?

Mostly, but I *really* want to see unittests for this.

Jeffrey, do you have an interest in taking it from here?

Raymond, mind if we roll back your previous (incorrect) changes?  They
really should not have been submitted in the first place.

No problem.  Whatever you think is best.

Still hoping that Jeffrey can take a look at the mixin approach and the 
__rand__ logic.  No one else seems to have an interest and I won't have 
time to write the tests for a few days (they are complicated because 
Integral requires so many abstract methods to be defined before it will 
instantiate).

Yeah, I'll take a look. Feel free to bug me if I haven't gotten to it in
a couple more days.

Ping?

I've long since lost interest in this.
If anyone wants to push it forward,
feel free to re-open.

New changeset 4678259af5a4 by Georg Brandl in branch '2.7':
The Integral class does not contain implementations for the bit-shifting operations. (See #3056.)
http://hg.python.org/cpython/rev/4678259af5a4

New changeset 1d4ba14cc505 by Georg Brandl in branch '3.3':
The Integral class does not contain implementations for the bit-shifting operations. (See #3056.)
http://hg.python.org/cpython/rev/1d4ba14cc505