Fuzzyman wrote: > That means making keys, values, and items custom objects. > Creating a new instance would have the overhead of creating 4 new > objects instead of just 1. Is the added convenience worth it ? (Plus > the extra layers of method calls for each access).
I'm not sure about that either. But since you are using odict for convenience reasons anyway, and not for performance reasons, it would be consequential. Performance testing should be made anyway, so this could be tested as well. I think that creating these 3 additional objects will not matter much if the dict has more than a handful of items. And the extra layers will be only called if you really access these keys, values or items attributes which will not happen very frequently. Normally, you just loop over an ordered directory and acess keyed values. > I'm not sure. It's a nice idea in terms of using it (we could just > leave the sequence attribute as an alias for the new keys attribute - > for backwards compatibility). Yes, you could make it a deprecated feature. -- Christoph -- http://mail.python.org/mailman/listinfo/python-list
