On Fri, Dec 18, 2015 at 10:57 PM, Steven D'Aprano steve@pearwood.info wrote:
See my example in the other email (https://mail.python.org/pipermail/python-ideas/2015-December/037498.html). That's a case where the order of comparison matters, so you can't do a conceptual "unordered comparison" without, in the worst case, comparing everything to everything else. This is due to custom __eq__ (by OrderedDict, for irony's sake): a == b and b == c does not mean a == c.
I don't know what Guido means by "values might not be comparable", but your example is lack of transitivity.
Mathematical equality is transitive: if a == b, and b == c, then a == c. But that doesn't extend to non-numeric concepts of equality, e.g. preference ordering, or other forms of ranking. Since Python __eq__ can be overridden, we cannot assume that equality of arbitrary objects is necessarily transitive. And indeed, even with Mappings they are not:
py> from collections import OrderedDict as odict py> a = odict([('a', 1), ('b', 2)]) py> b = dict(a) py> c = odict([('b', 2), ('a', 1)]) py> a == b == c True py> a == c False
Well, that's my point AND my example.
If this were lists, then a lack of transitivity of elements would mean a lack of transitivity for lists of those elements. You get what you put in.
But since dict views are unordered collections, a lack of transitivity for elements would mean INCONSISTENCY: the comparison of two views as multisets would depend on the exact order of comparison. Unless you are willing to compare everything to everything else in the worst case.