Release 1.15.0

[matklad]

bors r+

[bors[bot]]

Canceled.

[matklad]

bors r+

[matklad]

NB: this bump MSRV to 1.56.0. I think this should be fine, but, if you have specific objection to that, I'd love to hear.

This PR pushes 1.15.0-pre.1, I'll upgrade that to 1.15.0 proper next week.

[bors[bot]]

Build succeeded:

• Rust

[notgull]

My main objection is that Debian stable still uses Rust 1.48, so this would prevent crates I maintain that are dependent on this crate from building on there.

[matklad]

re-reading https://github.com/rust-lang/libs-team/issues/72, I think my stance on that is that once-cell intends to support projects which generally keep with up-to-date Rust, while also providing generous comparability window. It is a non-goal to support use-cases which have to use old compilers without intention to upgrade.

For debian-stable-like specifically, I think the reasonable choices are:

• install a more recent Rust using rustup
• if using old compiler is required, use old version of libraries as well
• if combination of old compiler and new library is required, it's on the consumer to patch the library on their side

This is based primarily on two observations:

• overall Rust ecosystem policy is closer to "live at head". When Rust policy clashes with distro policy, Rust policy is preferred
• practically, it seems relatively few people are stuck with as old Rust as debian stable, so this shouldn't have a big practical impact. And, arguably, those people who are impacted could be advised to find a way to use a supported version of Rust

[matklad]

From tokio discussion:

Tokio has MSRV of 6 month, but it also has a 12 month LTS, which means that effective MSRV for tokio is 18.

I don't think it makes sense for once_cell to chase downstream LTS promises (which is a restatement of the policy above that says that once_cell generally caters to users who aspire to use a somewhat up-to-date compiler).

So, I'll cut a release!