I also can't wait for this to happen, but I think we might want to wait for the final 0.20.0 release or even 0.20.1 before removing Py2 code to make backports easier.

From experience with IPython/Jupyter and backport: do not rush to remove code, it will make backport way harder, and trip git bisect/git blame.

We still have some py2 compat code here and there close to 18 month after dropping 2.7 and only remove it when it bother us.

Also feel free to use our backport bot for trivial backport.

Should we add some tests to make sure at least the new PRs don't include things such as from __future__ ..., etc.?

If master is Py3only, the change to get python2-ism is slim.

"I'm going to start implementing this feature by making sure it works with Python 2 even if the whole library is Python3 only" said no-one ever.

There are few construct that will likely survive the transition (not making use of general unpacking... etc), but nothing easy to detect.

One thing you might be able to do is turn more deprecation warnings into fatal errors.

Thanks for sharing your experience on this @Carreau !

There are a few further references to Py2:

sklearn/datasets/base.py:    """Ensure different filenames for Python 2 and Python 3 pickles
sklearn/datasets/base.py:    An object pickled under Python 3 cannot be loaded under Python 2. An object
sklearn/datasets/base.py:    pickled under Python 2 can sometimes not be loaded correctly under Python 3
sklearn/datasets/base.py:    because some Python 2 strings are decoded as Python 3 strings which can be
sklearn/datasets/base.py:    problematic for objects that use Python 2 strings as byte buffers for
sklearn/datasets/base.py:    manages by Python 2 and Python 3 in the same SCIKIT_LEARN_DATA folder so as
sklearn/datasets/base.py:      - /path/to/folder/filename.pkl under Python 2
sklearn/datasets/tests/data/svmlight_invalid.txt:python 2:2.5 10:-5.2 15:1.5
sklearn/ensemble/tests/test_forest.py:    # Using a Python 2.x list as the sample_weight parameter used to raise
sklearn/utils/_unittest_backport.py:            # don't switch to '{}' formatting in Python 2.X
sklearn/utils/deprecation.py:        # on function arguments in Python 2 (already works in Python 3)
sklearn/utils/fixes.py:    """Workaround for Python 2 limitations of pickling instance methods
sklearn/utils/testing.py:    Copied from Python 2.7.5 and modified as required.
sklearn/utils/testing.py:    We may not need to do this any more when getting rid of Python 2, not

I am working on sklearn/datasets/base.py

Working on sklearn/ensemble/tests/test_forest.py

Is there any pending task in this issue?

@albertvillanova you could check and see if everything in #11991 (comment) is done, or at least a PR is open for them.

@adrinjalali that is the reason why I was asking if there are any additional tasks, because the ones mentioned in this issue seem to have been already addressed.

Then I think it's safe to close this one. We can always have PRs which do further cleanups, but seems like this is mostly done. Feel free to re-open @jnothman if you see obvious leftovers.

I think all of the points were indeed addressed here.

@albertvillanova thanks! If you are looking for for another issue to contribute, you can also have a look at stalled PRs starting with those that are short and not too controversial as they are more likely to get merged faster.

@rth thanks for the advice. However it's easier said than done. It seems that most of the PR are waiting for reviews, aren't they?

@albertvillanova I've put some effort in checking stalled PR and labelling them as "help wanted": perhaps you will find your way in this pool.

Thank you @cmarmo. That is very useful!