Recall sorted...
sorted(iterable, cmp=None, key=None, reverse=False) --> new sorted
list
So why not construct sorteddicts using the same idea of sortedness?
sorteddict((mapping | sequence | nothing), cmp=None, key=None,
reverse=None)
Then we can specify the exact behaviour of sorteddicts: it's the same
as a regular dict, except when the order of keys is important they're
ordered as if they were sorted by sorted(keys, cmp=sd._cmp,
key=sd._key, reverse=sd._reverse). Here I imagine cmp, key and reverse
are stored in the new sorteddict as attributes _cmp, _key, _reverse -
obviously the actual implementation may differ.
This has the benefit of making sorteddict's behaviour explicit and
easy to understand, and doesn't introduce a new sorting API when we
already have a perfectly decent one.
The only problem here is the **kwargs form of the dict constructor
doesn't translate as we're using keyword args to pass in the sort
criterion. Perhaps someone has an idea of how this can be solved. If
nothing else, this constructor could be dropped for sorteddicts, and
sorteddict(dict(**kwargs), cmp=..., key=..., reverse=...) when that
behaviour is wanted.
I don't like the integral indexing idea or the slicing: indexing and
slicing are already part of python and it's bad to have different
syntax for the same concept on sorteddicts. I'd say it's not an
important enough for sorteddicts anyway.
--
Paul Hankin
--
http://mail.python.org/mailman/listinfo/python-list
