etime extend golang's time package, to provide more and better features.

Moved to https://github.com/go-eden/common

go get github.com/go-eden/etime

This section shows all features, and their usage.

This feature works like Java's System.currentTimeMillis(), it will return int64 value directly:

• CurrentSecond: obtain current second, use syscall for better performance
• CurrentMicrosecond: obtain current microsecond, use syscall for better performance
• CurrentMillisecond: obtain current millisecond, use syscall for better performance

For better performance, Current* didn't use time.Now(), because it's a bit slow.

package main

import (
	"github.com/go-eden/etime"
	"time"
)

func main() {
	println(etime.NowSecond())
	println(etime.NowMillisecond())
	println(etime.NowMicrosecond())

	println(etime.CurrentSecond())
	println(etime.CurrentMillisecond())
	println(etime.CurrentMicrosecond())

	println(time.Now().Unix())
	println(time.Now().UnixNano() / 1e6)
	println(time.Now().UnixNano() / 1e3)
}

In my benchmark, the performance of etime.CurrentSecond was about 40 ns/op, the performance of time.Now() was about 68 ns/op:

BenchmarkCurrentSecond-12         	30000000	        40.8 ns/op	       0 B/op	       0 allocs/op
BenchmarkCurrentMillisecond-12    	50000000	        39.7 ns/op	       0 B/op	       0 allocs/op
BenchmarkCurrentMicrosecond-12    	30000000	        39.7 ns/op	       0 B/op	       0 allocs/op
BenchmarkNowSecond-12             	20000000	        67.5 ns/op	       0 B/op	       0 allocs/op
BenchmarkNowMillisecond-12        	20000000	        68.1 ns/op	       0 B/op	       0 allocs/op
BenchmarkNowMicrosecond-12        	20000000	        67.8 ns/op	       0 B/op	       0 allocs/op

Under the same hardware environment, Java's System.currentTimeMillis() was like this:

Benchmark               Mode  Cnt   Score   Error  Units
TimestampBenchmark.now  avgt    9  25.697 ± 0.139  ns/op

Some library may be sensitive to this 28ns optimization, like slf4go.

By the way, System.currentTimeMillis()'s implementation was similar with etime.CurrentSecond:

jlong os::javaTimeMillis() {
  timeval time;
  int status = gettimeofday(&time, NULL);
  assert(status != -1, "bsd error");
  return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
}

So, there should have room for improvement.

MIT