Make once_cell optional to meet reasonable MSRV

[TheBlueMatt]

once_cell switched to "no MSRV at all" in matklad/once_cell#201, which makes it rather annoying to use aHash in any project that supports anything other than "latest rustc", which makes it annoying to use hashbrown in anything other than "latest rustc". Given the relatively little use of once_cell in ahash, it should be rather trivial to simply remove it entirely, which would be much appreciated.

[notgull]

I was under the impression that once_cell's MSRV was bumped to 1.56.0, rather than being abandoned entirely?

[TheBlueMatt]

I don't believe so, no, that's just the current effective MSRV, quoting the linked issue:

We support workflows which target up-to-date, supported versions of Rust compiler, but we also give somewhat generous grace period, as keeping perfectly up-to-date is hard.

We explicitly do not support workflows which depend on using the old compiler. Making sure that the latest once_cell can be compiled with rustc packaged with debian stable is a non-goal.

[matklad]

There's a big range of possibilities between "debian stable" and "latest stable". once_cell is using "about a year old MSRV". This is very comfortable for folks who try to generally keep up to date. That is not enough for various LTSes, which are typically more-than-a-year long.

[TheBlueMatt]

Yep, its totally a reasonable decision...maintainers get to decide what they want to support. Doesn't mean downstream projects shouldn't/can't decide to switch if they want to support more :). In this case we're switching, and it'd be easiest for us if ahash also switches, otherwise we'll swap ahash for a different hasher in our (nostd) hashtables...hence this issue/question.

[striezel]

That is not enough for more-than-a-year LTSes.

Anything below a year is not really worth calling it LTS, is it? After all, the L in LTS stands for "Long", and something like six months is not really long.

Don't get me wrong here: Code maintainers / owners always get to decide what they support and for how long they are willing to support it. But everything below a year does not really deserve to be called LTS, in my opinion.

[matklad]

Thanks, my wording was ambiguous, clarified.

[tkaitchuck]

Some features like: #136 do impose version requirements. (For example that feature won't wore before 1.54) I don't know a way to do conditional compilation based on the Rust version number. So removing onceCell may not be the only obstacle.

I am using use once_cell::race::OnceBox which is a very small amount of code. Rather than try to find a way around it, it might be better to break that out into its own crate.

[TheBlueMatt]

At least for my use-case (no-std hashmaps by directly depending on hashbrown, which disables ahash default-features), I'm not worried about the std feature. Ideally, of course, that too would have a strong MSRV guarantee, but providing a strong MSRV guarantee for some feature subset is still a good step forward.

[tkaitchuck]

A PR is open to add MSRV: #181