[Python-Dev] __eq__ vs hash

[Jean-Paul Calderone]

On Fri, 4 Apr 2008 07:38:04 -0700, Guido van Rossum <guido at python.org> wrote:
>On Fri, Apr 4, 2008 at 2:46 AM, Ralf Schmitt <schmir at gmail.com> wrote:
>> the news file for python 2.6 does not mention that you need to define
>> __hash__ in case you define __eq__ for a class.
>> This breaks some code (for me: mercurial and pyparsing).
>> Shouldn't this be documented somewhere (I also cannot find it in the
>> whatsnew file).
>
>Well, technically this has always been the requirement.
>
>What specific code breaks? Maybe we need to turn this into a warning
>in order to be more backwards compatible?
>

There was some code in Twisted (one class, specifically) which was broken/
revealed to be broken by this Python 2.6 change.  The code assumed identity
hashing if no __hash__ method was implemented.  This ended up only working
if you only had a singleton instance of the class, but the class also went
out of its way to make sure that was the case.  We have since changed the
code to work on Python 2.6.

If you're curious about the details, here's the code after the fix:

http://twistedmatrix.com/trac/browser/trunk/twisted/web2/dav/element/base.py?rev=22305#L345

Here's the changeset that fixed it:

http://twistedmatrix.com/trac/changeset/22305

And here's the same class before the fix:

http://twistedmatrix.com/trac/browser/trunk/twisted/web2/dav/element/base.py?rev=22304#L344

Jean-Paul