[Python-Dev] Re: PEP 218 (sets); moving set.py to Lib

[Guido van Rossum]

> Just to make sure, why not move sort_repr=False out of the parameter
> list and into the code body.

Good idea.  It can be a non-public class variable.

> [FP]
> > By the way (this was discussed on Python list a while ago), it might
> > be worth stressing in the official documentation that dicts, and maybe
> > Sets as well, all have a "natural" iteration order which remains fixed at
> > least while the dict or Set does not loose or acquire keys, and that this
> > same fixed order is used for .items(), .keys(), .values(), and all three
> > .iter* flavours.  It is sometimes useful being able to rely on this fact,
> > especially if Python clearly commits itself through the documentation.
> 
> Just like stability for the new list.sort(), this promise ought to remain
> a hidden, undocumented implementation detail.  Because of collision
> resolution, the "natural" order can vary depending on the order that
> the keys are inserted.  While the ordering stays constant until there
> is a change, it is fragile and could be changed by a resize operation
> even if the keys remain the same.  Let's keep the options open here
> in case someday we want GC or a memory manager to rebuild the
> dictionary at an arbitrary time.

I don't think François was stating that the order was only dependent
on the inserted keys.  I believe he was merely referring to the fact
that the order doesn't change as long as you don't mutate a dict, and
that it's the same for items(), keys(), values(), iterators, and
display order.  There's no reason to keep that hidden, and I believe
it's documented (though François didn't find it).

--Guido van Rossum (home page: http://www.python.org/~guido/)