Test i686-unknown-linux-gnu and aarch64-unknown-linux-gnu on CI

[taiki-e]

Tests i686-unknown-linux-gnu and aarch64-unknown-linux-gnu on CI by using cross.
Closes #598

I'm surprised that it looks like we don't have to ignore channel tests.

r? @jeehoonkang

[jeehoonkang]

I wonder if we can maintain only one set of crates for both x86-64 and the cross-target platforms. Maybe later we will add more crates, and I'd wish we should change the CI script in only one place...

[taiki-e]

Agreed. A single 'crates' set is certainly preferable. (It makes a matrix a little more complicated, but it should be possible.)

[taiki-e]

It makes a matrix a little more complicated, but it should be possible.

Hmm, the way I thought would work didn't really work. I'll look for other ways.

[taiki-e]

I addressed this by testing all crates at once in cross tests. (07a1366. The way of adjusting the matrix did not work.)

This should make non-x86-64 platform test time a bit longer than x86-64 platform, at this time. However, eventually, x86-64 platform test will take more time because sanitizers (#591), miri (#578), etc. will be added to x86-64 platform test. So I think this is fine.

[jeehoonkang]

Thank you for the revised PR. I just want to suggest another (potential) alternative: can we also test x86-64 platform in cross? If so, we can uniformly treat all platforms using cross.

[taiki-e]

cross doesn't provide images for windows-msvc and darwin, so I don't think using cross on those platforms will no benefit.
However, I'll take a look for a way to merge the test and cross jobs into one job.

[taiki-e]

@jeehoonkang Updated to test all crates at once.

• Tests that were previously run per crate are run at once using --all flag.
• Use cross if the target is set, otherwise, use cargo and test with the host target. See Test i686-unknown-linux-gnu and aarch64-unknown-linux-gnu on CI #613 (comment) for context.
  For now, the non-host target only runs tests.
• Replaced sanitize feature with cfg(crossbeam_sanitize).
  Feature flag can be accidentally enabled by --all-features, so use cfg instead. (This issue actually occurred when I tried to use --all-features instead because --features nightly cannot be used together with --all.)
  sanitize feature is a feature for testing, not a feature intended to be used as a public API, so it should be okay to remove it.
• Separated check for docs into its own job.
  This is in a different area than the test and I think it makes sense to run it as a separate job.

[jeehoonkang]

Thanks for the cleanup! I have a few comments.

[jeehoonkang]

Thanks!

bors r+

[bors]

Build succeeded:

• ci