homepage

From http://docs.python.org/py3k/reference/datamodel.html#object.__hash__
-----------------------------------------------------------------------
Classes which inherit a __hash__() method from a parent class but change the meaning of __eq__() such that the hash value returned is no longer appropriate (e.g. by switching to a value-based concept of equality instead of the default identity based equality) can explicitly flag themselves as being unhashable by setting __hash__ = None in the class definition. Doing so means that not only will instances of the class raise an appropriate TypeError when a program attempts to retrieve their hash value, but they will also be correctly identified as unhashable when checking isinstance(obj, collections.Hashable) (unlike classes which define their own __hash__() to explicitly raise TypeError).
If a class that overrides __eq__() needs to retain the implementation of __hash__() from a parent class, the interpreter must be told this explicitly by setting __hash__ = <ParentClass>.__hash__. Otherwise the inheritance of __hash__() will be blocked, just as if __hash__ had been explicitly set to None.
-----------------------------------------------------------------------
The first paragraph says the user has to change __hash__ if it's different because of changes to __eq__, the second paragraph says __hash__ is automatically removed if __eq__ is changed; the second paragraph reflects reality.
Proposed change:
-----------------------------------------------------------------------
Classes which change the meaning of __eq__() (thus losing automatic delegation to the parent class' __hash__) can explicitly flag themselves as being unhashable by setting __hash__ = None in the class definition (which is otherwise done implicity). Having __hash__ set to None, either explicitly or implicitly, means that not only will instances of the class raise an appropriate TypeError when a program attempts to retrieve their hash value, but they will also be correctly identified as unhashable when checking isinstance(obj, collections.Hashable) (unlike classes which define their own __hash__() to explicitly raise TypeError).
If a class that overrides __eq__() needs to retain the implementation of __hash__() from a parent class, the interpreter must be told this explicitly by setting __hash__ = <ParentClass>.__hash__.
-----------------------------------------------------------------------
Patch attached.

Some of the sentence phrasing still sounds a bit awkward to me ("[...], means that not only will instances" for example, and I would also remove the parens at the end), but globally I think this is an improvement.

Éric Araujo wrote:
> Changes by Éric Araujo <merwok@netwok.org>:
> 
> 
> ----------
> versions: +Python 2.7 -Python 3.1
The docs for 2.7 are correct, as __hash__ is not automatically 
suppressed in that version.
--
~Ethan~

More re-writing...

Are the changes good? Can they be commited?

Newest changes uploaded.

Any problems with the current doc patch? If not, can it be applied before RC1?

RC2 has just been released. Any chance of this getting in to the final release? Nobobdy has pointed out any problems with the last update...

New changeset 957e1eef3296 by R David Murray in branch '3.2':
#14617: clarify discussion of interrelationship of __eq__ and __hash__.
http://hg.python.org/cpython/rev/957e1eef3296
New changeset c8d60d0c736b by R David Murray in branch 'default':
Merge #14617: clarify discussion of interrelationship of __eq__ and __hash__.
http://hg.python.org/cpython/rev/c8d60d0c736b 

I rewrote the section a bit differently than you did in your patch...if you think my changes are not an improvement please let me know.

R. David Murray wrote:
> I rewrote the section a bit differently than you did in your patch...if you think my changes are not an improvement please let me know.
This line is incorrect:
A class which defines its own :meth:`__hash__` that explicitly raises a 
:exc:`TypeError` would be incorrectly identified as hashable by an 
``isinstance(obj, collections.Hashable)`` call.
It should be "would not be correctly identified as hashable".
Otherwise, looks great.

Ethan Furman wrote:
> Ethan Furman added the comment:
> 
> R. David Murray wrote:
>> I rewrote the section a bit differently than you did in your patch...if you think my changes are not an improvement please let me know.
> 
> This line is incorrect:
> 
> A class which defines its own :meth:`__hash__` that explicitly raises a 
> :exc:`TypeError` would be incorrectly identified as hashable by an 
> ``isinstance(obj, collections.Hashable)`` call.
> 
> It should be "would not be correctly identified as hashable".
> 
> Otherwise, looks great.
Argh. Nevermind! It all looks good. :)

This can wait until after the release. We will have periodic point releases where further doc updates can go in.