[Python-ideas] incremental hashing in __hash__

[Paul Moore]

On 6 January 2017 at 07:26, Neil Girdhar <mistersheik at gmail.com> wrote:
> On Fri, Jan 6, 2017 at 2:07 AM Stephen J. Turnbull
> <turnbull.stephen.fw at u.tsukuba.ac.jp> wrote:
>>
>> Neil Girdhar writes:
>>
>>  > I don't understand this?  How is providing a default method in an
>>  > abstract base class a pessimization?  If it happens to be slower
>>  > than the code in the current methods, it can still be overridden.
>>
>> How often will people override until it's bitten them?  How many
>> people will not even notice until they've lost business due to slow
>> response?  If you don't have a default, that's much more obvious.
>> Note that if there is a default, the collections are "large", and
>> equality comparisons are "rare", this could be a substantial overhead.
>
>
> I still don't understand what you're talking about here.  You're saying that
> we shouldn't provide a __hash__ in case the default hash happens to be
> slower than what the user wants and so by not providing it, we force the
> user to write a fast one?  Doesn't that argument apply to all methods
> provided by abcs?

The point here is that ABCs should provide defaults for methods where
there is an *obvious* default. It's not at all clear that there's an
obvious default for __hash__.

Unless I missed a revision of your proposal, what you suggested was:

> A better option is to add collections.abc.ImmutableIterable that derives from Iterable and provides __hash__.

So what classes would derive from ImmutableIterable? Frozenset? A
user-defined frozendict? There's no "obvious" default that would work
for both those cases. And that's before we even get to the question of
whether the default has the right performance characteristics, which
is highly application-dependent.

It's not clear to me if you expect ImmutableIterable to provide
anything other than a default implementation of hash. If not, then how
is making it an ABC any better than simply providing a helper function
that people can use in their own __hash__ implementation? That would
make it explicit what people are doing, and avoid any tendency towards
people thinking they "should" inherit from ImmutableIterable and yet
needing to override the only method it provides.

Paul