• Focus on one-shot benchmark
• RAII-style
• Statistics (Average, Median, 95% and 99% percentile)
• Colored output
• Tagging
• Nesting

I guess there are two types of benchmarks.

One is a benchmark of a small and fast function in which we want the statistics from a million of iterations. For this type of benchmark, Criterion.rs is a good fit.

Another type is what I call one-shot benchmark.

You may have wanted to write a benchmark program like this.

let mut db = DB::new();

let t = Instant::now();
db.write(...);
println!("write: {:?}", t.elapsed());

let t = Instant::now();
db.read(...);
println!("read: {:?}", t.elapsed());

According to Criterion.rs #531, this type of benchmark is infeasible with Criterion.rs because Criterion is focusing on the first type.

That's why I started to create benchman.

RAII is a good technique to manage resource access. My idea behind designing benchman is that stopwatch is like a resource because it is like a producer of a benchmark result that sends the result to the single central consumer and there is a strict rule that stopwatch shouldn't send the result twice.

With this idea, the library is designed like this.

let stopwatch = benchman.get_stopwatch("some_tag");
do_something();
drop(stopwatch);

// or

{
    let _sw = benchman.get_stopwatch("some_tag");
    do_something();
}

When the stopwatch is dropped, the measurement result is sent to the central database.

Akira Hayakawa (@akiradeveloper)