Yes. So what we're talking about is a better way to do indexing all together, all types in one shot. ('indexes', 'slices' and 'collections' ? I'm not sure how to name the first and the last one, if not that.) And not just indexing, but setting and deleting at the same time, right ?
Yes.
So the most preferred way is to pass a single argument which is either:
- A single object (directly what was passed in) for a single index. - A slice-object if any kind of single-slice was passed in, which holds the zero-to-three objects that the slice was 'created' with. - A tuple of zero or more of either of the above two.
(Make that a tuple of two or more!) Indexing like a[] is illegal, and indexing like a[i] passes a non-tuple.
While I'm writing this down, I'm not sure if it's such a good idea. Isn't this placing a tad too much into one function ? It might require some severe logic to support this all, especially if you give 'special' meanings to some indexes. And we're inserting a new catch-all method -- a set of them, actually: get, set and del -- and that while Paul is trying to solve the other catch-all Python has, __getattr__/__setattr__.
Actually, __getitem__ *already* has to deal with -- the only case it doesn't get currently is the step-less single slice, which gets passed to __getslice__ -- and dies if it's not there.
And lets not forget the convenience of writing those methods: __getitem__ is intuitive, both in name and semantics. So is __getslice__. The passing of a slice object to __getitem__ is a tad less intuitive, and a tuple of index/slice objects even less.
Don't worry, that's already the case.
I'm tempted to suggest a single change: when __getslice__ is not defined, pass a slice object (with no step, as if the slice had a trailing ':') to __getitem__, and document it properly.
Great! That's *exactly* what I've been proposing.
(and make builtin objects accept sliceobjects too !)
That gets a +1 from me too, in general (there's some doubt about the usefulness of things like ``L1[lo:ho:step] = L2'').
Perhaps try to slowly deprecate getslice__.
Yes.
Give plenty __of examples of using __getitem__ and slice objects in the standard documentation.
Sure.
Also, I think it makes sense to make slice objects indexable, so that you can do:
start, end, step = sliceobj
instead of the less intuitive
start, end, step = sliceobj.start, sliceobj.end, sliceobj.step
But I've never used slice objects myself, so I can't really say whether it's a good idea.
Maybe that's a good idea.
I suspect this is all for 2.1 or later, though.
Actually, the actual change (fall back on __*item__ with a slice object when __*slice__ doesn't exist) should be simple to add for 2.0. Just submit to the SF PM! --Guido van Rossum (home page: http://www.pythonlabs.com/~guido/)