Use Edition 2024, MSRV 1.85

[dhardy]

The MSRV bump is required if we want to make getrandom depend on rand_core (see rust-random/rand#1675). Unless we use a split MSRV depending on whether rand_core is a dependency.

It is however a significant bump, so we might prefer not to.

[newpavlov]

I am in favor of bumping MSRV and releasing v0.4 (see #722) regardless of whether OsRng will be part of getrandom or not.

(I will wait for CI to be fixed before approving)

[newpavlov]

Ideally, we should extract MSRV from the rust-version field instead of manually providing it.

[dhardy]

Releasing v0.4 works for me. Okay, I'll try to get this fixed.

[dhardy]

So yesterday CI showed this error:

error[E0133]: call to unsafe function `core::arch::x86_64::_rdrand64_step` is unsafe and requires unsafe block
  --> src/backends/rdrand.rs:34:12
   |
34 |         if rdrand_step(&mut val) == 1 {
   |            ^^^^^^^^^^^^^^^^^^^^^ call to unsafe function

Today after adding unsafe, the RDRAND UEFI job fails:

error: unnecessary `unsafe` block
  --> src/backends/rdrand.rs:34:12
   |
34 |         if unsafe { rdrand_step(&mut val) } == 1 {
   |            ^^^^^^ unnecessary `unsafe` block
   |
   = note: `-D unused-unsafe` implied by `-D warnings`
   = help: to override `-D warnings` add `#[allow(unused_unsafe)]`

This is using nightly rustc. So why does it not think unsafe is necessary here?

[dhardy]

Note: core::error::Error is stable since 1.81 so we may be able to drop the std feature after this. (New PR.)

[dhardy]

Got a better idea?

At least allow(unused_unsafe) is fairly harmless.

[newpavlov]

It's not quite a false positive... On Nightly the RDRAND intrinsics are marked as safe (e.g. see here). I am not sure why it's still marked as unsafe on the stable toolchain, since most intrinsics were changed to safe in 1.87 (after 1.86 relaxed unsafety requirements on functions marked with #[target_feature(enable="..")]).

For now, I would just mark the whole function with allow(unused_unsafe) without gating it on the target with a comment stating that in new Rust versions RDRAND intrinsics are safe.

[newpavlov]

we may be able to drop the std feature after this

We also use std for the From<Error> for io::Error impl and in the fmt::Debug/Display impls.

I don't have a strong opinion on whether we should keep them or remove the std feature completely.

[newpavlov]

Suggested change

	### Changes
	### Changed