Joe Rickerby

Fixed in #1889.

for reference: [delvewheel](https://github.com/adang1345/delvewheel)

I can't think what cibuildwheel would be doing that would affect this. We don't vary how docker is invoked between GHA and Azure - cibuildwheel treats them the same. So...

I think removing these variables for the test step makes sense in general. Probably the right approach would be to start again from `os.environ` rather than copying `env` here: https://github.com/pypa/cibuildwheel/blob/7c3b0fbee8d7cb8d065a50d12b206978204a8238/cibuildwheel/macos.py#L675...

Cirrus builds on Windows are failing atm. https://cirrus-ci.com/task/5500867840835584?logs=run_cibuildwheel_tests#L775 I'm guessing it's due to a space at the end of the PATH variable- https://cirrus-ci.com/task/5500867840835584?logs=run_cibuildwheel_tests#L550-L553

I thought that if you called `/bin/sh` on Unix you were getting a [POSIX-compliant shell](https://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html#tag_18), which can be implemented by a number of shells - for example, if that's symlinked...

It looks like the 'brace expansion' feature is actually a 'bashism'. So now I wonder why bash in POSIX mode still expands it.

This was fixed in musllinux by choosing `bash` as the default shell. So it'll be fixed in the next release.

Agreed! If we can do it in a way that unifies accounts between the old way and the new way that'll be the trick. The obvious way to do that...

The user model is like this- https://github.com/nordprojects/sirius/blob/main/sirius/models/user.py#L23-L33 So we have no email addresses for existing users. That means the twitter auth can't go away, else we lock users out of...