[Python-ideas] lazy tuple unpacking

[Ian Cordasco]

On Fri, Jul 4, 2014 at 7:59 PM, Paul Tagliamonte <paultag at gmail.com> wrote:
> Given:
>
>     >>> def g_range(n):
>     ...     for y in range(n):
>     ...         yield y
>     ...
>
> I notice that:
>
>     >>> a, b, c, *others = g_range(100)
>
> Works great. Super useful stuff there. Looks good.
>
> I also notice that this causes *others to consume the generator
> in a greedy way.
>
>     >>> type(others)
>     <class 'list'>
>
> And this makes me sad.
>
>     >>> a, b, c, *others = g_range(10000000000)
>     # will also make your machine very sad. Eventually resulting
>     # (ok, unless you've got a really fancy bit of kit) in:
>     Killed
>
> Really, the behavior (I think) should be more similar to:
>
>     >>> _x = g_range(1000000000000)
>     >>> a = next(_x)
>     >>> b = next(_x)
>     >>> c = next(_x)
>     >>> others = _x
>     >>>
>
>
> Of course, this leads to all sorts of fun errors, like the fact you
> couldn't iterate over it twice. This might not be expected. However, it
> might be nice to have this behavior when you're unpacking a generator.
>
> Thoughts?

I agree that the behaviour is suboptimal, but as Chris already pointed
out it would introduce a significant inconsistency in the API of
unpacking. I'm struggling to see a *good* way of doing this. My first
instinct was that we could make something like this do what you
expect:

>>> a, b, c, others = g_range(some_really_big_number)
>>> others
<generator ...>

But this doesn't work like this currently because Python currently
raises a ValueError because there were too many values to unpack. I'm
also against introducing some new syntax to add the behaviour.