Allow some deprecated items

[killercup]

These were deprecated recently but we don't want to use them yet to not
break compatibility with older rustc versions unnecessary. Explicitly
allowing them should fix the build for nightly with #[deny(warnings)].

[JohnTitor]

mem::initialized will be deprecated in 1.39.0, not 1.38.0 by this PR but it's reasonable to allow that before it is too late.

[killercup]

Rebased on top of your branch, @JohnTitor, so it now includes the Once::new(). Let's see what CI says.

[JohnTitor]

Hmm, an error unrelated to this PR has occurred in macOS mysql beta.

[killercup]

Yeah, the tests even passed but

Hosted Agent lost communication with the server

when running rustdoc.

I'll read this as "CI is green" :D

[JohnTitor]

lgtm, nice work!

[sgrif]

Given that many folks on the unsafe code guidelines feel that mem::uninitialized can never be used safely without UB, should we just switch to using MaybeUninit instead of allowing deprecated here? Or is it not stable on the current stable yet?

[JohnTitor]

The problem with using MaybeUninit is that it has recently been stabilized (1.36.0). As mentioned above, we need to bump up the minimal version. If bumping up it isn't a big problem, it's good to replace uninitialized() with MaybeUninit.

[sgrif]

Our policy is that we only bump if we have a good reason (e.g. not just for a convenience function). I think that meets this bar. We usually want to have some amount of lag so folks using their distro's Rust have a chance of being able to get the right version, but I think right now we're at least a few months from 2.0 shipping so it's fine.

[JohnTitor]

Yeah, it seems we have a good reason to bump up the version.

[weiznich]

I would vote for closing this PR.
#2131 fixes the CI by allowing all those warnings on nightly builds.
#2108 Fixes the warning regarding to ONCE_INIT
The deprecation of mem::uninitialized should be handled in a different PR

[killercup]

👍