The collections module will be offering new datatypes, including a double-ended queue (deque).
Right now, these are new elemental types, inheriting directly from object. I believe they should be subclassed from existing classes where this makes sense. Even if no actual methods are inherited, the API is valuable.
For instance, should:
list([1,2,3]) == deque([1,2,3]) deque() < list([1,2]) < deque()
If deque inherits directly from object, then the only comparison available is based on address. If deque subclasses list (or at least pretends to), then sensible comparisons can be made.
All list operations make sense on a deque (and vice versa). Operations may be more (or less) efficient depending on the type, but they do make sense. Even if the concern is to avoid encouraging inefficient code, the methods should still be available.
d = deque(...) d[4:11] = ('x', 'y')
is inefficient, but still better than
d = deque(...) temp = list(d) temp[4:11] = ('x', 'y') d = deque(temp)
As an additional benefit, making deque a subclass of list with the same API gives later python implementations more freedom; in theory, the compiler could determine which internal representation made more sense for each particular case.