This won't get built on travis-ci because only master branch gets built (this includes target branch for PR)

Huh, why do we only build the master branch? Seems a leftover from @joerick trying to stop his own PRs building twice (because it being a PR and it being a commit): 9762867. But this seems unintended, no?

Apart from that, thanks for noticing that s390x is now working!

Any reason to update the other versions, if we're not making a release? I'd somehow say it would be nice to have these the same as the last true release, such that trying out Python 3.9 and our branch wouldn't get any unexpected consequences/errors?

Can you not see a Travis build? I see it - https://travis-ci.org/github/joerick/cibuildwheel/builds/704961809

I think it's triggered by #396, and Travis just confuses them because they are the same commit, no?

EDIT: This one, #395, doesn't show up here: https://travis-ci.org/github/joerick/cibuildwheel/pull_requests

I think it's triggered by #396, and Travis just confuses them because they are the same commit, no?

Yes exactly.

Any reason to update the other versions, if we're not making a release? I'd somehow say it would be nice to have these the same as the last true release, such that trying out Python 3.9 and our branch wouldn't get any unexpected consequences/errors?

I wanted to check for unexpected errors with other updates at first (and because running python bin/update_dependencies.py is easy and updates all). I now dropped the other updates and just update the docker images.

Edit: updating other dependencies had for effect what seems to be a deadlock on macOS on travis-ci...

Ah, I see - #396 is just to trigger CI. Looks like it's green, so feel free to merge into python3.9!

Edit: updating other dependencies had for effect what seems to be a deadlock on macOS on travis-ci...

Huh, good to know, thanks! We shouldn't do this update too last minute before the next update, then, @joerick!

Meanwhile, let's get this updated. Thanks again! :-)