bump coco to 0.3

[ignatenkobrain]

No description provided.

[cuviper]

If stjepang/coco#20 gets merged and published, we won't have to worry about bumping rayon-core's minimum rustc for this.

[stjepang]

@cuviper Done! Merged and published.

[ignatenkobrain]

@stjepang https://crates.io/crates/coco still doesn't list 0.3.2 ;)

[stjepang]

Oops, sorry! Now it does. :)

[cuviper]

Here are some cargo-chrono plots before and after:

Mostly wins, but two big regressions:

• fibonacci_split_recursive: this is a kind of weird one, that I'm willing to write off. FWIW, the related fibonacci_join benchmarks (not shown here) are about twice as fast now!
• parallel_find_last: the really weird part about this is that it's now performing even worse than parallel_find_missing, which itself stayed about the same. Parallel finding the last element should be able to short-circuit and return sooner than with a missing element that has to check everything. I'm really baffled by this one!

[cuviper]

On a whim, I just tried compiling rayon-demo with all test modules except find commented out, and now parallel_find_last does perform better than parallel_find_missing as expected, and about the same as with the old coco. But parallel_find_middle got worse.

My best guess is that these are just sensitive to changes in overall code generation and/or layout. Meh.

bors r+

[bors]

Build succeeded

• continuous-integration/appveyor/branch
• continuous-integration/travis-ci/push