So, after the post on events PJE suggested I try using generic functions for events. I'm still not sure that's where I want to start, since there's some more conservative options out there to address this. However, I have long meant to use his generic functions, as I think they seem pretty neat.

Installation was easy; following what the PEAK homepage says, I ran:

easy_install -Zf http://peak.telecommunity.com/snapshots/ RuleDispatch

It was a little problematic, because PyDispatcher also implements a dispatch package. I ended up just uninstalling that, though there's probably a better way to do it.

My first test was an experiment in converting objects to a JSONable form (result here). This worked very easily, and with no surprises. Of course, it's an easy problem. But still, it's not a given that introducing a new tool to solve an easy problem will give an easy result.

Using this, you can fiddle with the JSON representation of nearly any object. For instance, lets say you get Binary objects (like from a database BLOB column), and you want to turn them into base64-encoded strings:

@jsonify.when('isinstance(obj, Binary)')
def jsonify_binary(obj):
    return str(obj).encode('base64')

The only issue I have with the implementation is that it's hard to say apply this rule only in this context, because jsonify() doesn't have any argument that implies context. So if you wnat to change the representation of an object, but only in the context of one controller in your web app, then that doesn't fit into this system. Dynamically scoped variables would probably address this case.

Anyway, in my second test I threw in a context argument parameter, just for this kind of thing. Especially because in my second test I wanted to allow something most other adaptation systems can't -- detailed tweaking of specific cases based on arbitrary rules that don't have any overriding sense to them. These are exactly the rules that make up good UI design, and my second test is for generating forms from objects.

For the second test I added some (experimental) modules to FormEncode (new website, BTW): fields, formgen, and sqlformgen (the SQLObject-specific form generation routines), and a test.

And while the result is by no means complete, and there's still parts of the design I don't understand well, again the generic functions didn't introduce any surprises. I suspect when I try different combinators or some other tricky thing they will surprise me eventually; but of all the parts of this (rather difficult) problem, the generic functions seem the easiest.