Additionally, GraalJIT consistently yields higher throughput than the stock JDK, with a similar warmup time

\\Special\acknowledgement\goes\to\Windows\path handling\

Such things rarely can happen individually, often they cannot even happen within a single team or a product.

I am eternally grateful to have @stukalov.bsky.social
on my side. Congrats! That's your achievement as much as it's Kotlin's 💜

We poured the real effort, broke really thick walls, had heated debates and pitches, stretched IntelliJ limits (sometimes as far as using test infra to make it work in the LSP), and damn we did it! And there is still more to come :)

Not only was that a hell of a team effort, pushing the IJ codebase to its limits, but it is also a great step forward for Kotlin and a huge shift within JetBrains that we made happen.

gist.github.com/qwwdfsad/eb3...

Something like that can do the trick (for less trivial chains still, similar to streams).

By the way, beware of benchmarks not consuming (returning) the result from the @Benchmark function, maybe it's worth remeasuring the original bench (especially the small one)

Thinking about it more, we can potentially provide a push-based sequence almost seamlessly, but it will require quite some tweaks.
And for trivial use cases (small collections, trivial transformations) unlikely to be *much* faster

Iterator-based protocol (which sequence is) is really unfortunate, the constant field-spilling and hasNext-next pair (which is strict with its contract) absorbs all the CPU :(

Also, a nice showcase of why we picked the push-based Flow model!

totally deserved!

Is it going to deadlock because of the clinit "deadlock" where lambda body (private static fun) cannot be invoked from another thread ("parallel") unless clinit is completed which in turn awaits parallel operator to complete?

Not all notes are that useful though

...and today is one of these days. Thank you, Seva from the past, I guess

By the way, @volebamor.bsky.social suggested a really beautiful solution -- count all the junctions (i.e. ▢▢ has two junctions: ⊥ and ⊤) and subtract the number from the perimeter. I am still mad that it actually works!

Damn these modern business practices! Who on earth counts discounts LIKE THAT?!

you can probably guess what the section is about :)