The uubench from akdubya

uubench's People

Stargazers

Watchers

uubench's Issues

numbers off?

hey, im doing

    var persec = 1000 / stats.elapsed
      , ops = stats.iterations * persec;

where i'd expect setTimeout(next, 1000) to give 1 op/s, however it seems to be around 10 etc

The implementation of "minimum runtime" is inaccurate

@akdubya

According to README.md:

var suite = new uubench.Suite({
  min: 1000, // each benchmark should run for at least 1000ms
  ...
 });

I looked up the source code:

function adaptive() {
    if (--pend === 0) {
      var elapsed = new Date() - start;
      if (elapsed < min) {
        self.run(iter*2);
      } else {
        self.callback({iterations: iter, elapsed: elapsed});//since "start" is reset whenever "Bench.run" is invoked, then the stats is inaccurate
      }
    }
  }

The key point is that the value of "start" is reset whenever "Bench.run" is invoked:

Bench.prototype.run = function(iter) {
  var self = this, fn = self.test,
      checkfn = self.options.type === "adaptive" ? adaptive : fixed,
      i = iter, pend = i,
      min = self.options.min, start;

  if (self.loop) {
    pend = 1;
    start = new Date();  //start is reset
    fn(checkfn, i);
  } else {
    start = new Date();   //start is reset
    while (i--) {
      fn(checkfn);
    }
  }

So the stats is inaccurate, I think the value of "start" should be set at the very begining of "Suite.prototype.runOne", and don't reset it in Bench.run:

Suite.prototype.runOne = function(idx) {
  var self = this;
  setTimeout(function() {
    self.tests[idx].start = new Date().getTime();  //set "start" value here, so "min" is the accumulative runtime for benchmark
    self.tests[idx].run(self.options.iterations);
  }, self.options.delay);
}

I'm very curious to this issue. Please give me a hint!

Finally, many thanks to you for your great work and the sharing :)

Recommend Projects