Faster Ord when Eq

[Eh2406]

I have a tool for benchmarking Cargoes resolver and a new PubGrub based resolver. Both resolvers spend a lot of time checking whether two Versions are equal using Ord. For example, .contains on a BTreeMap<Versions, .... @arielb1 pointed out that adding a fast path makes a big difference in the runtime of the resolvers. On master the program reports PubGrub CPU time: 86.97min, Cargo CPU time: 190.67min, with this PR PubGrub CPU time: 71.33min, Cargo CPU time: 167.92min. (For posterity the command is cargo r -r -- -t 100 running commit 9dc991f94c45688a6d2d90d29532c5d96fd064b3) there is about +/- 3 min of run-to-run variability.

In most cases Prerelease and BuildMetadata are empty. Prerelease already did not do much work for the empty == empty case, most of the time difference therefore comes from BuildMetadata.

I also considered a simpler

if self.identifier.is_empty() && rhs.identifier.is_empty() {
    return Ordering::Equal;
}

Which is not as big an improvement (PubGrub CPU time: 75.85min).

Other variations that get approximately the same result, at least within the margin of error are:

if self.identifier == rhs.identifier {
    return Ordering::Equal;
}

or

if self.identifier.as_str().as_ptr() == rhs.identifier.as_str().as_ptr() {
    return Ordering::Equal;
}

What are your thoughts about this optimization? Which variation would you prefer? Is there some testing you would like added?

[dtolnay]

This looks great to me. Thank you!

[dtolnay]

Published in 1.0.24.