Incomparable abominations (was: python-dev Summary)

Tim Peters tim.one at comcast.net
Sun Mar 23 05:44:24 CET 2003


[Donn Cave]
> What about
>
>       1 == "2"
>
> , should it also be cause for an exception?  Sorry if I missed the
> earlier posts in this thread that would explain how == differs from
> < in this application.

It's arguable both ways, of course.  After rich comparisons appeared, people
played with both approaches in their own types.  What Guido seems happiest
with now (and which I not coincidentally implemented for the 2.3 datetime
module's types, after real-world problems with "always raise an exception"
on meaningless comparisons):

class SomeType:
    def __eq__(self, other):
        if not_a_type_i_know_about(type(other)):
            return NotImplemented
        return True or False appropriately

and similarly for __ne__.  The others raise TypeError instead if they don't
know about other.

The implications for __eq__:  if type(other) is not one SomeType knows
about, returning NotImplemented gives other a chance at providing a
meaningful result.  If other.__eq__(self) also returns NotImplemented,
Python falls back to comparing self and other by memory address.  Since self
is not other, this must return False.

Similarly, if __ne__ falls back to comparing by memory address, that must
return True.

In a world where this was consistently applied,

    1 == "2"

would return False, and

    1 != "2"

True, and

    1 <= "2"
    1 <  "2"
    1 >  "2"
    1 >= "2"

would raise TypeError.






More information about the Python-list mailing list