Sunsetting official support for Python 2

[sethmlarson]

urllib3 1.x has supported Python 2 for over 4 years since the end of Python 2's security release cycle. Since that time, download numbers indicate that Python 2 makes up around 2% of our total downloads across all versions.

Today it's quite difficult to actually test Python 2 on our 1.26.x branch due to dependencies and GitHub Actions moving on from Python 2 so long ago. I propose we update the minimum Python version on the 1.26.x branch to be "3" in the packaging metadata and remove any Python 2 CI/testing to avoid further blocking that branch. We won't intentionally break Python 2 support, but in the future a security fix may do so.

We'll also keep App Engine support around, but will be dropping the requirement to test it since it requires Python 2.

[pquentin]

I propose we update the minimum Python version on the 1.26.x branch to be "3" in the packaging metadata

Doing this will effectively block Python 2 users to a specific 1.26.x version - if we don't want to break Python 2 and make sure our fixes reach those users, we can't do that specific part.

and remove any Python 2 CI/testing to avoid further blocking that branch.

While this seems workable, I don't really trust myself to keep Python 2 support working without CI keeping me honest. The ideal situation as far as I'm concerned would be to at some point completely drop Python 2. (And other Python EOL versions, possibly keeping the latest EOL for longer).

Since that time, download numbers indicate that Python 2 makes up around 2% of our total downloads across all versions.

In June as you mentioned we were around 2% across all versions (~1.7% during the week, ~2.1% during the weekend). In October, usage has dropped again, around ~1% (~0.9% during the week, ~1.0% during the weekend). (We've always had different number during the week and the weekend, I suspect hobbyists are more represented during the weekend, and they tend to stick to older versions of Python and urllib3.)

Talking about numbers, here are the current numbers on urllib3 1.26.x. Using the following query on BigQuery:

SELECT
  REGEXP_EXTRACT(details.python, r"[0-9]+\.[0-9]+") AS python_version,
  COUNT(*) AS num_downloads,
FROM `bigquery-public-data.pypi.file_downloads`
WHERE
  DATE(timestamp)
    BETWEEN DATE_TRUNC(DATE_SUB(CURRENT_DATE(), INTERVAL 1 WEEK), DAY)
    AND CURRENT_DATE()
  AND file.project = 'urllib3'
  AND file.version LIKE '1.26.%'
GROUP BY `python_version`
ORDER BY `num_downloads` DESC

I'm getting the following chart showing that Python 2.7 is 1.27% of 1.26.x download numbers over the past week, not really higher than numbers across all versions:

I'd suggest dropping Python 2 support in January 2025, five years after the CPython EOL date. I don't think any Tidelift subscriber cares about Python 2.7 anymore.

[pquentin]

Closing now that we dropped support for urllib3 1.26.x and Python 2.7.