We can now be fairly confident that our naive HashMap has the same behavior as the standard library for the properties we've checked. But how does the runtime performance of naive HashMap stack up? To answer this, we'll write a benchmark, but not a benchmark using the unstable Bencher subsystem that's available in the nightly channel. Instead, we'll use Jorge Aparicio's criterion—inspired by the Haskell tool of the same name by Bryan O'Sullivan—which is available on stable and does statistically valid sampling of runs. All Rust benchmark code lives under the top-level benches/ directory and criterion benchmarks are no different. Open benches/naive.rs and give it this preamble:
#[macro_use] extern crate criterion; ...
