Are dicts supposed to raise comparison errors

[Chris Angelico]

On Wed, Aug 1, 2018 at 6:36 PM, Robin Becker <robin at reportlab.com> wrote:
> On 31/07/2018 16:52, Chris Angelico wrote:
>>
>> On Wed, Aug 1, 2018 at 1:28 AM, MRAB <python at mrabarnett.plus.com> wrote:
>>>
>>> On 2018-07-31 08:40, Robin Becker wrote:
>>>>
>>>>
>>>> A bitbucket user complains that python 3.6.6 with -Wall -b prints
>>>> warnings
>
> .............
>>>
>>> The warning looks wrong to be.
>>>
>>> In Python 2, u'a' and b'a' would be treated as the same key, but in
>>> Python 3
>>> they are distinct and can co-exist.
>>>
>>> Something for Python's bug tracker, I think!
>>
>>
>> It's a warning specifically requested by the -b option. The two keys
>> in question have the same hash, which means they have to be compared
>> directly; they will compare unequal, but because of the -b flag, the
>> comparison triggers a warning. If that warning is spurious, *don't use
>> the -b option*.
>>
>> ChrisA
>>
>
> I looked at the language documentation for 3.7 mappings
>
>> These represent finite sets of objects indexed by nearly arbitrary values.
>> The only types of values not acceptable as keys are values containing lists
>> or dictionaries or other mutable types that are compared by value rather
>> than by object identity, the reason being that the efficient implementation
>> of dictionaries requires a key’s hash value to remain constant. Numeric
>> types used for keys obey the normal rules for numeric comparison: if two
>> numbers compare equal (e.g., 1 and 1.0) then they can be used
>> interchangeably to index the same dictionary entry.
>
>
>
>
> it says explicitly that numeric keys will use numeric comparison, but no
> mention is made of strings/bytes etc etc and there's an implication that
> object identity is used rather than comparison. In python 3.x b'a' is not
> the same as 'a' either the documentation is lacking some reference to
> strings/bytes or the warning is wrong. Using the excuse that normal
> comparison is being used seems a bit feeble. It would clearly improve speed
> if object identity were to be used, but I suppose that's not the case for
> other reasons.
>

Technically, the comparison used is:

a is b or a == b

in other words, identity will match, but mainly, equality is used. The
identity check improves performance in many common cases, and also
avoids pathological cases involving float("nan"), but in general
discussion, it's assumed that value rather than identity is the
comparison used.

ChrisA