[VOTE] Ending Python 2.x support in next release (0.83)

[hcho3]

Given that Python 2 will reach its end of life at the end of this year, we should drop Python 2 support. That is, contributors would no longer be required to write Python code that's compatible with Python 2.7. Also, Python distributions of XGBoost on PyPI will be Python 3 exclusive.

Indicate your support by replying "+1" to this thread. I want to get 4-5 votes from the committers to proceed with this decision. (Others, feel free to voice support with +1 reply as well, although only committers will have binding votes.)

@dmlc/xgboost-committer

[terrytangyuan]

Definitely +1

[trivialfis]

+1.

We need to upgrade Python on test farm then.

[RAMitchell]

+1

[CodingCat]

+1

[hcho3]

+1

[pommedeterresautee]

+1

[tqchen]

+1

[hcho3]

With 7 votes from committers, it is decided that the upcoming release (0.83) will support Python 3 exclusively. I filed #4381 to migrate pylint to Python 3, and #4380 will use Python 3 exclusively for all CI tests. Thanks!

[CAM-Gerlach]

Given the major status of this project and the major importance of this project to the scientific Python community, you should consider adding your name and logo to Python3Statement.org of projects planning to drop support for Python 2 on or before Jan 1 2020, to maximize public visibility of your support timeline in a central location, and help send the message that remaining Python 2 users should start transitioning now to minimize the pain later. To add XGBoost, you can just submit a simple PR to the Python3Statement repo with your name and logo—a SVG or transparent PNG of this would be ideal:

Or, if you like, I can create a PR for you there and @hcho3 or whomever responsible can sign off, if you want. Thanks!

[hcho3]

@CAM-Gerlach Yes, we'd love to put our logo there. Can you go ahead and create a PR on our behalf? I'll sign it off.

[CAM-Gerlach]

@hcho3 Sure; I'll go ahead and tag you there. Thanks!

[CAM-Gerlach]

@hcho3 Particularly since it is ending Python 2 support before 2020, it would be really nice to have the project in the timeline. What date should I put for the beginning of Python 3 support (or has it always supported Python 3 since the first release in 2014)? I couldn't find any info on this in the obvious places (Trove classifiers, .travis.yml/appveyor.yml, release notes, readme, etc).

Also, is there an approx release date for 0.90 (it doesn't have to be exact, just the likely date of release assuming Spark 2.4.3 drops on as planned on or about May 5 (which it looks it will) and implementing the remaining items for #4389 go according to plan; due to the long time span of the charge, a difference of up to a couple weeks from the actual won't really be visible. Thanks!

[hcho3]

@CAM-Gerlach XGBoost has been tested with Python 3.x for a long while. At PyPI, we have been distributing "universal" wheels of XGBoost that are both Python 2 and 3 compatible. See https://pypi.org/project/xgboost/#files.

Only change with 0.90 release is that we will no longer make guarantee about supporting Python 2. We hope to release 0.90 by middle of May, assuming we can address the items in the Roadmap.

[CAM-Gerlach]

XGBoost has been tested with Python 3.x for a long while.

Okay, so about what rough timeframe should I put down for that, then?

[hcho3]

At least since Nov 2015, based on this commit: 25c4fbd. This commit adds a compatibility layer to support Python 2 and 3.