Issue 15475: Correct __sizeof__ support for itertools

Here is a patch that implements correct __sizeof__ for some itertools iterators: product(), combinations(), combinations_with_replacement() and permutations().

Patches updated on Meador's remarks. Tests updated to use new test.support helpers.

Sorry, on Martin's remarks.

The clean-up patch looks much nicer.  Thanks.

I'm not sure why anyone would really want this functionality for itertools, but I don't see any harm in adding it as long as the other implementations aren't required to follow.

Also, it could be viewed as being a new feature, so I'm not sure backports are warranted.  Be sure to ask the respective release managers before applying.

On Wed, Aug 8, 2012 at 5:41 AM, Raymond Hettinger <report@bugs.python.org> wrote:

> I'm not sure why anyone would really want this functionality for itertools,
> but I don't see any harm in adding it as long as the other implementations
> aren't required to follow.

The general expection is that __sizeof__ is for CPython.  We tag the tests
with '@cpython_only'.

> Also, it could be viewed as being a new feature, so I'm not sure backports
> are warranted. 

So far, we have been applying the __sizeof__ patches to all branches.

> Be sure to ask the respective release managers before applying.

I will check with Georg for 3.3.  Thanks.

Are there any objections?

Certainly not for 3.3.0.  I'm also not sure this counts as a bugfix in any case.

Georg, sorry, I don't understood you. You have no objection to the inclusion 
of this patch in 3.3.0 or you say that this patch should not be in 3.3.0?

Ah, sorry, that was unclear.  I meant "this is certainly not going into 3.3.0."

Unassigning from myself.  I thought I would have more time to review and push this through.

Serhiy, you can make the decision on whether this goes into 3.4 or gets closed.

I don't think any actual user has ever expressed interest in this functionality and it is hard to meaningfully reason about the real memory cost because we don't have a way to include the memory of the referred-to objects.  To me, it seems like unnecessary code bloat.

New changeset 9bce03920afe by Serhiy Storchaka in branch 'default':
Issue #15475: Add __sizeof__ implementations for itertools objects.
http://hg.python.org/cpython/rev/9bce03920afe

messages: + msg166780