tail-call optimization seems to not be implemented for suspend lambdas on JVM. Is that expected? I'l...

# compiler

[Youssef Shoaib [MOD]]

tail-call optimization seems to not be implemented for suspend lambdas on JVM. Is that expected? I'll provide sample later

[ephemient]

do you mean tailrec suspend?

[Youssef Shoaib [MOD]]

No no, tail call optimization, as in if the last call is a suspend function, the parent continuation is passed to it instead of the current one. This causes a significant difference in the bytecode between e.g. { foo() } and ::foo. The first creates a pointless state machine, the second doesn't

[ephemient]

ah that's not what I think of at all when it comes to TCO

[ephemient]

good question. that's how it should be working in regular functions, missing it in lambdas seems like an oversight

[ephemient]

I expect that a callable object

object : suspend () -> Unit {
    override suspend fun invoke() {
        foo()
    }
}

would do the right thing?

[Youssef Shoaib [MOD]]

It does, but they're not allowed on JS. A workaround I've done is:

suspend fun myFoo() = foo(arg1)
bar(::myFoo)

Which does the right thing

[ephemient]

hmm. I'd ask if

bar(suspend fun() = foo(arg1))

would be a workaround but it seems the compiler doesn't work with those either (https://youtrack.jetbrains.com/issue/KT-62021)