In the last AoC challenge you have to put marbles ...

# announcements

[kef]

In the last AoC challenge you have to put marbles in a circle (linked list), sometimes moving back 7 places from the current position.
Stripped version of the my code for that is below.
I thought that the straight forward current.prev.prev.prev.prev.prev.prev.prev should be the fastest way possible, but out of curiosity I also did it as
generateSequence(current) { it.prev }.take(8).last() and to my surprise, when I measured times of execution I got this :
- .prev chain -> 2389ms
- generateSequence -> 460ms

So why is generateSequence 5 times faster?

On my machine I get results like that when looping up to 10mln, after that generateSequence suddenly takes 3s to finish.

-.kt

[todd.ginsberg]

If you are doing AoC, there are quite a few of us in #C87V9MQFK!

[kef]

Thanks! Will link my question there 🙂

[karelpeeters]

It appears to hit some worst-case scenario in the interpreter, but after a couple of iterations I get 60ms for the sequence and 40ms for the .prev.prev thing as expected.

[kef]

I know that after JIT kicks in times get better. But how can I verify it is the case and what exactly makes it hit the worse case scenario? Also, is there any way to improve it for a single run (like running it once as .kts)?

[karelpeeters]

Maybe running it with native makes it better?