Version 1.1.0 bumps minimum compiler version

[jethrogb]

Version 1.1.0 bumps the minimum compiler version because of the use of associated constants. This is not a semver-compatible change and should be reverted.

[BurntSushi]

There is no consensus on whether bumping the MSRV is a semver incompatible change or not.

[jethrogb]

My CI started failing overnight because of this release even though I employ lockfiles everywhere.

[sfackler]

What does "employing lockfiles everywhere" mean? If your versions are locked, then you wouldn't pick up a new version with a higher compiler version requirement.

[jethrogb]

Yeah, I suppose that's quite puzzling. I've narrowed it down to an optional dependency that only gets activated through a particular feature. This dependency is not in the lockfile.

[jethrogb]

Caused by rust-lang/cargo#3629

[KodrAus]

Yeh the 1.1.0 release didn't bump the minimum compiler version. The additional build.rs specifically exists to avoid bumping the minimum compiler version.

[jethrogb]

That might not be working correctly then:

02:00:25   --> /home/jenkins/.cargo/registry/src/github.com-1ecc6299db9ec823/lazy_static-1.1.0/src/heap_lazy.rs:17:5
02:00:25    |
02:00:25 17 |     pub const INIT: Self = Lazy(0 as *const T, ONCE_INIT);
02:00:25    |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
02:00:25    |
02:00:25    = help: add #![feature(associated_consts)] to the crate attributes to enable
02:00:25 
02:00:25 error: aborting due to previous error(s)
02:00:25 
02:00:25 error: Could not compile `lazy_static`.

I don't get this error on 1.0.2

[KodrAus]

@jethrogb Oh are you compiling lazy_static on a compiler before 1.21.0? While we didn't bump our documented minimum supported version from 1.21.0 we might've bumped our technically supported minimum version by actually using a feature that stabilized close to 1.21.0?

[jethrogb]

I'm on 1.18

we didn't bump our documented minimum supported version from 1.21.0

Documented where exactly?

[KodrAus]

Documented where exactly?

It's documented as the explicit version we test in Travis through our .travis.yml, but you could reasonably argue that totally doesn't count as documentation. It's clearly not discoverable...

Maybe we should adopt the approach some other crates take and stick a badge in the readme with our minimum version?

[BurntSushi]

Digging a little deeper here...

• Rust 1.21 was added to the CI config on Dec 8, 2017.
• lazy-static 1.0.0 was released on Nov 29, 2017.
• From testing, it looks like lazy_static 1.0.2's actual minimum Rust version is Rust 1.6.0.
• From testing, it looks like lazy_static 1.1.0's actual minimum Rust version is Rust 1.20.0. (The documented minimum Rust version is Rust 1.21.0.) The version increase appears to have been caused by the use of associated constants.

There are no rigid policies on how to handle all of this, but it is a bit of a mess. This also means regex's documented minimum version is technically wrong because it is less than lazy_static's documented minimum version (a dependency of regex). I don't really know what to do about it. I'm not sure it's worth this amount of fuss at this point, at least until we get some sort of LTS release that core crates can rally around.

[KodrAus]

Hm it is a bit of a mess... I think we've made a few assumptions that lead to this specific case:

• I assumed that because we test 1.21.0 explicitly in CI that it's our minimum version and bumping up to 1.21.0 is ok and expected
• Others have assumed that because their library that depends on lazy_static happens to compile on a given version that lazy_static supports that version

Some sort of LTS release would nicely cut through those assumptions in the future. It seems like our efforts right now across the ecosystem to manage minimum versions are a bit too ad-hoc to scale.

[BurntSushi]

It seems like our efforts right now across the ecosystem to manage minimum versions are a bit too ad-hoc to scale.

I agree. I watch the repos of (almost?) every dependency I use, including this one, and specifically look out for this and I still missed it. :-(