ntime provides a monotonic time, expressed as an int64 nanosecond count. It is intended for applications which only require relative time measurements, and wish to optimize memory usage and (likely) speed.
go get github.com/clipperhouse/ntimeimport "github.com/clipperhouse/ntime"
type MyThing struct {
Time ntime.Time
}
thing := MyThing{
Time: ntime.Now(),
}
// Time passes...
time.Sleep(100 * time.Millisecond)
now := ntime.Now()
age := now.Sub(thing.Time)
if age > 100*time.Millisecond {
//... do a thing with the thing
}
The Go stdlib time package offers monotonic time, which is a nice bit of robustness against a changing system clock. You should usually just use that.
If you're an optimizer like me, and are building something that primarily cares about relative time between two operations, then ntime.Time offers an 8-byte type, vs the 24-byte time.Time, while still offering monotonicity.
ntime also offers many of the convenience methods from the stdlib time, such as Sub, After, etc.
ntime.Time is a relative time against an arbitrary constant epoch, and is meant only for comparisons (i.e. durations). Do not mistake it for system ("wall") time. It is only useful in-process; multiple processes will have different epochs. There is a ToTime() method to convert to system time, which should be accurate in the absence of clock skew.
I naïvely began with the simple idea of using Nanoseconds() / UnixNano() from the stdlib time package. Those are integers!
But then I had the good sense to wonder if they are monotonic in the way that time.Now() is. Turns out they are not.
A changing system clock is unlikely, until it isn't, and I would like to eliminate that class of surprise.
One can get both (monotonic + integer) by capturing an epoch via time.Now() at process start, and then using time.Since(). This package makes that convenient.
ntime is this package and time is the standard library. Looks like ntime.Now() is about 2x faster than time.Now() on my machine. Other operations are about the same. See BENCHMARKS.txt.
goos: darwin
goarch: arm64
pkg: github.com/clipperhouse/ntime
cpu: Apple M2
Benchmark_Now/ntime/Serial-8 15.91 ns/op 0 B/op 0 allocs/op
Benchmark_Now/ntime/Parallel-8 3.334 ns/op 0 B/op 0 allocs/op
Benchmark_Now/time/Serial-8 32.25 ns/op 0 B/op 0 allocs/op
Benchmark_Now/time/Parallel-8 10.86 ns/op 0 B/op 0 allocs/op
Benchmark_Since/ntime/Serial-8 15.90 ns/op 0 B/op 0 allocs/op
Benchmark_Since/ntime/Parallel-8 3.425 ns/op 0 B/op 0 allocs/op
Benchmark_Since/time/Serial-8 15.50 ns/op 0 B/op 0 allocs/op
Benchmark_Since/time/Parallel-8 2.954 ns/op 0 B/op 0 allocs/op
Benchmark_After/ntime/Serial-8 1.898 ns/op 0 B/op 0 allocs/op
Benchmark_After/ntime/Parallel-8 0.2988 ns/op 0 B/op 0 allocs/op
Benchmark_After/time/Serial-8 1.920 ns/op 0 B/op 0 allocs/op
Benchmark_After/time/Parallel-8 0.3602 ns/op 0 B/op 0 allocs/op
Benchmark_Add/ntime/Serial-8 1.894 ns/op 0 B/op 0 allocs/op
Benchmark_Add/ntime/Parallel-8 0.3067 ns/op 0 B/op 0 allocs/op
Benchmark_Add/time/Serial-8 3.644 ns/op 0 B/op 0 allocs/op
Benchmark_Add/time/Parallel-8 0.6965 ns/op 0 B/op 0 allocs/op
Benchmark_Sub/ntime/Serial-8 1.916 ns/op 0 B/op 0 allocs/op
Benchmark_Sub/ntime/Parallel-8 0.3225 ns/op 0 B/op 0 allocs/op
Benchmark_Sub/time/Serial-8 1.967 ns/op 0 B/op 0 allocs/op
Benchmark_Sub/time/Parallel-8 0.3756 ns/op 0 B/op 0 allocs/op
PASS
ok github.com/clipperhouse/ntime