Hello! I am using Ktor for some time now and I could say that I became pretty familiar, but there is...

# ktor

[Florentin Simion]

Hello! I am using Ktor for some time now and I could say that I became pretty familiar, but there is still a thing that puzzles me 😅 What problem addresses / solves SuspendFunctionGun in respect to the naive implementation of DebugPipelineExecutor? @e5l maybe you can help me understand, as you are one of the main committers, Thank you!

[e5l]

Hey @Florentin Simion, the only purpose of SuspendFunctionGun is reduce amount of allocations during Pipeline execution.

[e5l]

DebugPipelineExecutor preserves stacks of all interceptors so it's possible to observe full coroutine stack frames with all plugins during the debug process

[Florentin Simion]

Thank you for the quick response. So, just to be sure I got it, SuspendFunctionGun only keeps the stack trace of the current running interceptor?

[Florentin Simion]

I am asking this, because what you said, made sense for me too, in the old implementation when you were starting a new coroutine for every interceptor which didn't aggregate stack traces. But now, I saw that the implementation was changed. which no longer does that. So, I was thinking if they are running in the same coroutine, aren't now the stack traces aggregated?

[e5l]

It depends on suspension points you have actually

[Florentin Simion]

I think I may got it this time 😄 This is due to the Continuation implementation in SuspendFunctionGun, which has override fun getStackTraceElement(): StackTraceElement? = null and this is how you keep the stack trace only for the current interceptor? And the coroutine start in the old implementation had nothing to do with it, it was just an implementation detail.

[Florentin Simion]

Awesome, thanks a lot!

[e5l]

Yep, and the continuation is reused