I noticed that ```CoroutineScope.launch{ }``` takes 40 ms. Isn’t this a bit much? Part of the point ...

# getting-started

[Ido Flax]

I noticed that

CoroutineScope.launch{ }

takes 40 ms. Isn’t this a bit much? Part of the point of launching is that you don’t need to wait for it, but apparently you do need to wait 40 ms as least

val scope = CoroutineScope(Dispatchers.Default)
    println(
        measureTimeMillis {
            scope.launch { }
        }
    )
// 40

This is done on a macbook pro 2021 m1 pro

[gildor]

Only on first launch? Maybe it’s just class loader penalty on first access

[gildor]

Yep, looks like classloader/bytecode optimization penalty on first lauch, I have 42/5/0/0/1/0/etc results per launch

[Ido Flax]

Ok, interesting, thanks

[Joffrey]

Also maybe the first use of Dispatchers.Default actually creates the threads in that pool

[Joffrey]

In any case, don't trust your microbenchmark if you don't use JMH :)

[gildor]

Yep, good point, Default/IO also will creates all thread pools on first access

[Ido Flax]

val scope = CoroutineScope(newFixedThreadPoolContext(1, "test"))
println(
    measureTimeMillis {
        scope.launch { }
    }
)
println(
    measureTimeMillis {
        scope.launch { }
    }
)
println(
    measureTimeMillis {
        scope.launch { }
    }
)

44
22
0

[Ido Flax]

anyway, i guess all good then 🙂 thanks

[gildor]

But anyway, hard to say what takes this time without profiling, it worth to do if it’s a problem (for example there are many Dispatcher.Main for Android improvements, which optimizes first access)