chore: update to virtualenv<20.22.0

[mayeut]

I'd like not to drop 3.6 right away (& possibly up-to centos 7 EOL or OpenSSL 1.1.1 EOL) as it is still used in around 15% of wheel builds (on linux at least).

[henryiii]

Don’t we need the new version of virtualenv for python 3.12? pypa/virtualenv#2558

Can’t users needing an old version of Python use an old version of cibuildwheel, an old version of manylinux, etc? How many of that 15% are actually wheels that are still being updated, and how many are old wheels for projects that have already dropped 3.6?

[mayeut]

How many of that 15% are actually wheels that are still being updated

That's for the projects published in the past 28 days (I rebuilt the graph for https://mayeut.github.io/manylinux-timeline/ changing the sliding window to 28 days).

Don’t we need the new version of virtualenv for python 3.12 ?

We can revisit at this time if needed (admittedly, very soon).

Just for comparison, when 2.7 & 3.5 were dropped in July 2021, the usage was around 6% & 9% (of around 700 packages for a 28 days period at that time vs around 1300 packages nowadays for a 28 days period)
If we look at the consumer side (which does not take into account wheels still being updated), it was 5% & 2% vs 10%

[mayeut]

Don’t we need the new version of virtualenv for python 3.12 ?

We can revisit at this time if needed (admittedly, very soon).

The answer is we don't need it: https://github.com/mayeut/cibuildwheel/actions/runs/4850698405/jobs/8643858588

[joerick]

I'm okay with holding on a little longer with Python 3.6 if you feel it's important @mayeut . But I wonder if there might be a bit of a chicken-egg situation here, where some users are gonna keep using 3.6 until they realise they need to upgrade for a package, and packagers are gonna keep releasing 3.6 wheels while cibuildwheel supports it, and cibuildwheel/manylinux aren't gonna drop 3.6 while users... you get the picture :)

[henryiii]

See also #1375. I do know there's a strong desire to support "everything cibuildwheel/manylinux supports" (I've been influenced by that before). But it's okay for users who really need to support old versions to use old versions of things, including old cibuildwheel/manylinux.

Windows working might not be a guarantee of what will happen on Linux or macOS.

We should continue to support whatever manylinux does, I think, but if it starts forcing caps on users who don't care about 3.6, that could start increasing the burden of supporting it.

[henryiii]

PS: Not sure about "right away" - we are already a year and four months after 3.6's EoL. :)