Daniel Roy Greenfeld

Daniel Roy Greenfeld

About | Articles | Books | Jobs | News | Tags

Thoughts on Python interfaces

This was originally posted on blogger here.

Python does not have interfaces. Yet people have implemented interfaces in Python via some really innovative code (PEAK and Zope 3 comes to mind, but Trac also has them). Those people have great arguments for interfaces, claiming documentation and enhancement of system structure. From what I gather the theory is that if you use interfaces its easy to create truly componentized architectures because you know what to expect from a component.

Now that said, I find it really amusing how often interfaces end up being just so much boilerplate. By this, I mean an empty, (or marker), interface. We are given to understand that one can do so much more, but sometimes a framework demands an interface in a particular place, and often that interface is just plain empty.

I've never played with PEAK elements beyond easy_install. I've toyed with the innards of Trac and been shocked by what makes up the core of that so important software tool. Zope 3 is weel organized and I've done some shockingly fun stuff there after I got over the Zope 3 ZCML hump. And in all of that, I barely saw the need for interfaces. So often I wonder if interfaces are needed.

Well, upon reflection for those systems interfaces work surprising well. The underlying code for Trac might be questionable but anyone can make a plugin by following rules obviously managed by the interface system. Zope 3 is really nice once you get past the curve because you can make components and tie them easily in knots with ZCML (or with python in the case of Grok).

The point of these thoughts? Nothing really. I can live with or without interfaces, and use them in the frameworks that need them.

Tags: grok python interfaces zope legacy-blogger
← Back to home