Hi everyone, I wonder if someone could clarify something for me. When I first learned about coroutines, I remember being taught that one should prefer to use

launch

from your existing context and then move it elsewhere using

withContext

instead of calling

launch

and specifying dispatcher.

scope.launch { withContext(<http://Dispatchers.IO|Dispatchers.IO>) { work() } }

instead of

scope.launch(<http://Dispatchers.IO|Dispatchers.IO>) { work() }

. I remember this being described as preferential since there was some performance penalty when launching in a different context that did not occur when moving a running coroutine using

withContext

. I’m trying to find guidance on this again and I can’t find anything about it. Did I misunderstand something when I was first learning?

I’ve never heard that ever, and it doesn’t make any sense to me. Just pass the context to launch.

Yeah, it does seem strange. I just wish I could figure out where I got this idea. It’s definitely not something I would have made up, I must have misunderstood something really early on. 🤔

you are probably thinking of

Default

and

IO

which share a common thread pool and switching between those may not do an actual thread switch

What you remember might be more like a convention. It goes like this. A suspend function should never block the calling thread. This results in two invariants: a) you never need to switch context before calling a suspend function. b) a suspend function that blocks (e.g. for blocking io) needs to switch context before doing so. So in most cases you just launch and leave the context switching to the suspend functions you call

@uli This sounds right! I did more reading last night and found recommendations like this. I must have conflated the reasoning for it in my head a while back and didn’t question it until now. This does make sense. Thank you all for weighing in!