Pre-PEP: Dictionary accumulator methods

[Jeremy Bowers]

On Sat, 19 Mar 2005 20:07:40 -0800, Kay Schluehr wrote:
> It is bad OO design, George. I want to be a bit more become more
> specific on this and provide an example:

Having thought about this for a bit, I agree it is a powerful
counter-argument and in many other languages I'd consider this a total win.

But this is Python, and frankly, I've overridden dict more than once and
violated the Liskov substitution principle without thinking twice. (Once,
oh yes, but not twice.) Of course, all the code was under my control then.

I think the tipping point for me depends on how the interfaces in Python
are going to be implemented, which I haven't dug into. If the dict class
gets an interface definition, can I subclass from dict and "cancel" (or
some other term) the interface I inherited? 

If so, then this might still be OK, although if interfaces aren't going to
confuse newbies enough, wait 'till we try to explain that their code is
blowing up because they forgot to "cancel" their interface, and they
can't *really* pass their subclass in to something expecting a dict
interface. If you *can't* cancel or downgrade the interface, then I'd say
this argument is still good; dict should be kept minimal and this should
go in a subclass.