container.___le___ can use only <=?

[cerutti at tds.net]

On Dec 14, 11:04 pm, Steven D'Aprano <st... at REMOVE-THIS-
cybersource.com.au> wrote:
> On Fri, 14 Dec 2007 21:15:44 +0000, Neil Cerutti wrote:
> > When implementing the rich comparison operators for some sort of
> > container, it's tempting to save code by doing something like:
>
> >  class LarchTree:
> >    ...
> >    def __gt__(self, other):
> >      # A lot of code to traverse the tree
> >    def __le__(self):
> >      return not self > other
>
> > However, if I'm thinking correctly, this is a bad idea. The reasoning
> > being that > and <= are not required to be each others opposite for
> > arbitrary value types, and may not even both be supplied by the
> > contained objects.
>
> > If a LarchTree user stores objects that don't support __gt__, will he
> > have a legitimate complaint when using LarchTree.__le__ results in an
> > attribute error?
>
> I'm not sure that an AttributeError is the right exception to raise, but
> in general I'd say that if the contained objects don't support "normal"
> comparisons (that is, if < and >= aren't opposites, etc.) then all bets
> are off. "Behaviour is undefined" time.

I wasn't going to raise the exception, I was thinking about calling a
function that raised the exception.

Assume for a moment that list's __le__ were implemented as above.

class BadGT(object):
  def __gt__(self, o):
    raise RuntimeError("I just hate this operation")
  def __le__(self, o):
    return id(self) <= id(other)

>>> x = [BadGT(), BadGT()]
>>> x <= []

If lists implementation of __le__ calls __gt__ on its constituents
*then* it'll be a RuntimeError.

I've tested list, and it safely compares lists containing BadGT
objects, as along I don't call __gt__ myself. So I was contemplating
implementing a phalanx of tests along these lines, which the current
version of my container will surely fail.

> BTW, what is a LarchTree? Googling just brings up the actual botanical
> tree, a type of conifer.
>
> http://en.wikipedia.org/wiki/Larch

It was a Monty Python reference. I'm actually still fiddling around
with a doubly-linked list implementation. It's kinda embarrassing so I
hid that fact, but actually I'm learning a bunch of Python and getting
design practice from this exercise (it turns out linked lists have
keys!).

--
Neil Cerutti