Hello folks, I am trying to benchmark the sequence...

# stdlib

[George]

Hello folks, I am trying to benchmark the sequence API vs the stream API vs simple transformations. I am seeing some weird results that the stream API is always the slowest no matter the scenario (even for short circuit).
Here is the repo: https://github.com/GeorgePap-719/StreamAllocationBenchmark

[George]

Is this expected? Maybe I am not using the jmh properly or smth

[Filipp Zhinkin]

Hey!
The dataSet list is always empty, because buildList { for (number in 1..size) { .. } uses list builder's size, not the benchmark parameter.

The teardown method is called after each iteration, but the setup one is called only once per benchmark execution. As a result, the size parameter won't affect results anyway, as the list will be empty after the first iteration.

Also, I'd suggest increasing the number of warmup iterations and using a more up-to-date version of JMH (1.37).

[George]

Hey Filipp, thanks for the help!

[George]

Run the benchmark again with your pointers, but still stream API is the slowest at short-circuit operations. Is this expected? From my understanding that is the strong point of the stream/sequence API where if we use take(k) the time com will be k and not n . 🤔

[Huib Donkers]

Did you look into this:

The teardown method is called after each iteration, but the setup one is called only once per benchmark execution. As a result, the size parameter won't affect results anyway, as the list will be empty after the first iteration.

Try instead @Setup(Level.Iteration)

[George]

damn, my bad. My brain just skipped this. Thanks!