Raise MSRV

[alex]

We're raising the MSRV for pyca/cryptography to 1.83. I think historically we have been the most conservative user of pyo3, and thus normally this means pyo3 can also raise its MSRV.

Things I know we'd benefit from with a 1.83 MSRV:

• We can drop memoffset as that functionality is now in std
• We can drop at least some once_cell usage in favor of std

[davidhewitt]

This would also benefit #5438 which proposes using MSRV 1.83 for the experimental-inspect feature to enable processing of some of the type hint information inside const fn. We could probably also simplify a lot of const generic code for similar reasons.

In #5170 we noted our reference Linux distributions have Rust 1.84+, so there would be no concern from our promised policy there.

I will note that Rust 1.76 removed support for Windows 7, which was also last supported in Python 3.8. If you still support Python 3.7 or 3.8 in cryptography you might not want to raise MSRV above 1.75 else you'll probably break any old software running on Windows 7 which attempts to download a new crytography version (Windows 7 is end of life, but breaking it seems unfortunate).

I think we will be pushing the 0.27 release next week to coincide with the Python 3.14.0 release, I don't think we should make this change before then.

When we're ready to drop Python 3.7 and 3.8 support, I think we can bump to 1.83.

[alex]

We still support 3.8, but we dropped Windows 7 support already.

[davidhewitt]

Is there packaging metadata you can use to prevent Windows 7 users downloading an unsupported cryptography? (similar to requires_python)?

[alex]

Alas, no. Just documented in our changelog.

[davidhewitt]

Fair enough!

I guess if that's worked ok and the rest of the ecosystem only supports 3.9 (with no Windows 7 concern) then presumably we're ok to move forward with this bump. 👍