"Private" attributes - a possible idea (maybe PEP-worthy).

[Delaney, Timothy]

> From: Jeff Shannon [mailto:jeff at ccvcorp.com]
> Michal Wallace wrote:
> 
> > On Thu, 31 Jan 2002, Delaney, Timothy wrote:
> >
> > > > From: Skip Montanaro [mailto:skip at pobox.com]
> > > >     Tim> a.Klass -> _1__attr1 and _1__attr2
> > > >     Tim> b.Klass -> _2__attr1 and _2__attr2
> > > >
> > > > This would be unworkable.  Private attributes couldn't 
> be pickled.
> > >
> > > Damn - knew there had to be *something* big against it.
> > >
> > > However, I can't help but feel that there must be *some*
> > > way to enforce that a "private" variable won't be
> > > accidentally used by a subclass.
> >
> > But that's exactly what the magical __whatever notation is
> > for int he first place! :)
> 
> Well, yes, but the O.P. was providing an example of two classes in
> different modules with the same name, one of which inherits from the
> other.  (i.e., class SomeClass(othermodule.SomeClass): ...)
> 
> Of course, the solution to this problem is -- don't do that!  :)   It
> seems to me that it'd be confusing anyway.  If nothing else, call the
> derived one My_SomeClass or something (or the base one 
> BaseSomeClass), so

Ah - but my most important point was that sometimes you may not know the
know of one of the classes in the inheritance tree. Say I have a class
called Klass, and I want to create a subclass called TimKlass.

However, unbeknowst to me, the author of the class was *also* called Tim,
and Klass actually inherits from a TimKlass.

Suddenly you may find "private" attributes changing for no apparent reason.

Yes, I know - read the source. However, the whole point of a public API is
that it should not matter how things are composed. Perhaps a better example:
I inherit from Klass to create TimKlass, having ensured that there is no
class in the inheritance tree called TimKlass. I get a new version of the
library. Now Klass *does* inherit from TimKlass, and I happen to be using
the same private attribute names that TimKlass does. My code breaks *badly*
for no apparent reason - and hunting down the problem would be an absolute
nightmare.

Tim Delaney