Concrete classes -- stylistic question

[Alex Martelli]

On Friday 11 October 2002 02:25, Ian Bicking wrote:
> On Thu, 2002-10-10 at 16:29, Alex Martelli wrote:
> > Gerhard Häring wrote:
> >         ...
> >
> > > Now who will come up with a neat example using metaclasses?
> >
> > I borrowed Guido's time machine to post about that:
> >
> > http://tinyurl.com/1wq1
>
> I should really learn to understand what's going on with metaclasses,
> but here's my attempt at similar functionality without metaclasses...

Pretty good!  A metaclass that only defines __new__, as metaMetaBunch
does, is roughly equivalent to a factory function, such as yours.  Even 
then, of course, using a metaclass lets you use normal class syntax rather
than function-call syntax  --  class XX(blah): etc  rather than
XX = blah(etc)  --  but metaclasses do offer more.

Say that you have metaMetaBunch in some library, and a new
requirement comes up -- in some cases you want repr(X), where
X is a class with that meta, to follow the usual convention, so that
eval(repr(X)) == X.  If type(X) is forced upon you, i.e. if you do NOT
use a custom meta, you just can't -- you don't have any control on
the results of repr(X) nor on the comparison of two such classes.

With the custom metaclass, it becomes easy -- even if you need to
leave the original metaclass alone, you can reuse by inheritance, as
usual: subclass the meta and add __repr__ and __eq__ to it -- such
methods defined in the metaclass control behavior of the metaclass's
instances, i.e., the classes that instantiate it, just as for other cases
of class / instance relationships.


Alex

to