Maybe also bump crate versions to make it clear that next releases will be breaking ones?

Unless I forgot something we didn't yet make any breaking changes to rand_core, so non-breaking releases of this plus the RNG crates would still be possible. So I guess we should ask: do we want to make a breaking release to those too?

• For: keeping a single MSRV for everything here is slightly easier for us (and one less CI test)
• For: we have some pending breaking changes: fill_via_chunks: mutate src on BE (small optimisation) #1182, Add read_adapter to avoid dynamic dispatch #1267
• Against: avoiding a breaking release to rand_core makes migration to Rand v0.9 easier
• Some ideas for breaking changes to RngCore have been floated but no PR yet and quite possibly we want new language features first, hence we expect a breaking release in the future anyway (possibly years away so could be 2.0)
• We can't release 1.0 without getrandom 1.0 anyway

So maybe I should revert the MSRV bump to core + RNG crates and add another CI MSRV test?

I'll cross-post to the tracker for anyone watching that.

test-avr fails due to using too old pinned version of nightly:

toolchain: nightly-2021-01-07 # Pinned compiler version due to rust-lang/compiler-builtins#400

AVR failure is rust-lang/rust#102278 ?

Latest AVR failure:

error: ran out of registers during register allocation

At this point I'm tempted to remove AVR support. CC @Palladinium who added this CI test in #1144.

Tests requiring unstable features are expected to fail from time to time, so maybe we should just ignore this failure? I expect it will be fixed eventually.

Is there any point having a test if we ignore failures?

Anyway, I was planning on leaving this open for a few more days.

I added another PR to the list of pending breaking changes for rand_core, so maybe we should just go ahead with a breaking release? I'm still on the fence about this question.

Ideally, we would notice when it works again. However, in this case we pin the nightly, so I guess it won't fix itself.

Does it make sense to keep the test as a comment?

Okay, I commented the AVR test out.

Lets go with a breaking release for rand_core since we have a few proposed changes and no objections.

Apparently AVR works on nightly-2022-12-09 works.

I won't bother re-enabling AVR (no motivation) but if someone else wants to, go ahead and make a PR.