[Python-Dev] Do we need __length_hint__ at all? (Was PEP 0424: A method for exposing a length hint)

[Nick Coghlan]

On Tue, Jul 17, 2012 at 12:01 AM, Antoine Pitrou <solipsis at pitrou.net> wrote:
> On Mon, 16 Jul 2012 23:23:18 +1000
> Nick Coghlan <ncoghlan at gmail.com> wrote:
>>
>> - distinguishing between different reasons for saying "don't
>> preallocate any space" (i.e. returning zero). I still haven't heard a
>> convincing rationale for this one
>
> The point is that zero is a valid value for a length hint. By making it
> a special value for "don't know", you are making the protocol
> potentially confusing, and you are also departing from the current
> semantics.

No, it just means the default estimate is always zero. If you don't do
that, then *every* client of __length_hint__ has to check for the
magic value. It's making the API harder to use for no good reason.

> (and, yes, I think distinguishing between zero and "don't know" is
> useful: imagine a container that would preallocate 256 entries by
> default when the answer is "don't know")

Such a container has to already deal with the case when it
overestimates severely. The only cost of using zero as a default
estimate is that such hypothetical containers will overallocate when
they technically didn't need to, which will already happen for any
empty iterator that doesn't provide __length_hint__ at all.

Given that all standard library containers default to assuming a size
of zero (absent information indicating otherwise), and a special value
would need to be special cased by *every* client of the API (and
almost always defaulted to zero), it's simply not a good trade-off.

>> The PEP itself already has this general tone, but I think that it
>> should be even more emphatic that it's about codifying the status quo,
>> *not* about modifying it with ideas haven't been proven useful through
>> past experience.
>
> Then the PEP shouldn't address infinite iterators at all.

Noting that infinite iterators are free to define __length_hint__ to
always throw an exception *is* the status quo. We just haven't done it
for the stdlib ones.

Cheers,
Nick.

-- 
Nick Coghlan   |   ncoghlan at gmail.com   |   Brisbane, Australia