| msg265110 - (view) |
Author: Joshua Morton (Joshua Morton) |
Date: 2016-05-08 04:24 |
| Following the comments in python ideas [1], I'm submitting a bug report. In python 3, dictionary views (KeysView and ItemsView specifically) do not adhere to the same interface as Sets. Specifically, the __and__, __or__, __xor__, and __sub__ methods on Views will accept a non-set type (`{}.keys() | []`), while the same throws a TypeError on a Set. The suggested, not-backwards-compatible solution, was to have dictviews raise errors in the same way. [1] https://mail.python.org/pipermail/python-ideas/2016-April/039469.html |
|
|
| msg265114 - (view) |
Author: Steven D'Aprano (steven.daprano) *  |
Date: 2016-05-08 06:00 |
| On Sun, May 08, 2016 at 04:24:59AM +0000, Joshua Morton wrote: > Following the comments in python ideas [1] [...] > > [1] https://mail.python.org/pipermail/python-ideas/2016-April/039469.html Am I missing something? I don't see anything resembling a consensus that the behaviour of dict views is a bug in that thread, in fact I hardly see any discussion about this specific behaviour. (Other related topics are discussed in more depth.) I don't think it is a bug for views to permit non-set arguments with binops: py> {'a': 1, 'b': 2}.keys() & ['a', 'c'] {'a'} I think it is a matter of taste. My own taste tells me that sets should be more restrictive, only accepting other [frozen]sets or subclasses, but views should be less restrictive, and perform more duck-typing of "set-like" objects, including lists. No, I can't justify it, except by an appeal to status quo: that's how it is now, and I don't think that changing it is worth breaking backwards compatibility. |
|
|
| msg265117 - (view) |
Author: Joshua Morton (Joshua Morton) |
Date: 2016-05-08 07:08 |
| There (seemed to be) consensus between the one or two on topic commenters that something was off, although much of the discussion was on a tangent. Although on looking back, there was even less discussion than I originally thought. Heh. My response was going to be very different, however I did some additional digging: the `collections.abc.Set` class can be used as a mixin for implementing set-likes. In its default implementations of the various binops, they will attempt to convert the right hand argument of an operator to the correct type, via _from_iterable. This in effect makes the set the special case that is more permissive than Set is. This strikes me as strange, I'd expect set and Set to be 'the same', or in other words a Set to act like a set and vice versa. But it seems that python swings toward your intuition and not mine. And the opposite, making set *more* permissive also doesn't sit well, nor does it feel necessary. |
|
|
| msg265158 - (view) |
Author: R. David Murray (r.david.murray) * |
Date: 2016-05-08 19:06 |
| In any case I think the backward compatibility argument torpedoes making the views more restrictive. Since you agree there's no real consensus on python-ideas, I'm going to close this. If there is some other consistency change that seems to make sense and gains consensus on python-ideas, you can open another issue. |
|
|