If I have a suspend function that does nothing but...

# coroutines

[Allan Wang]

If I have a suspend function that does nothing but change contexts, how do I ensure that I catch any failures and ensure that my logic continues? My parent scope is already a supervisor job, and I try catch my suspended calls. However, I notice that I need to do so multiple times to ensure that I actually catch errors

After some tests, I can use a supervisorScope and try catch it. However, it still doesn’t stop the cancellation of every other job, despite the main context already being a supervisor

[Dico]

Try adding NonCancellable to your context

[Allan Wang]

Does NonCancellable still allow job.cancel()?

My use case:

launch {
   // a
   suspendFunB()
   // c
}
launch {
  // d
}

Suspend B may crash, but I always want a and c to execute. I also want d to always execute without being affected by the previous launch. If the parent scope is cancelled, I’d like everything to cancel.

Should NonCancellable be around suspendFunB or the first launch?

[Dico]

As far as I'm concerned, NonCancellable means the coroutine ignores cancellation from job.cancel(). So I would do:

launch(NonCancellable) {
   // a
   try {
     suspendFunB()
   } finally {
     // c
   }
}

[Allan Wang]

the problem is that even if c runs, if it does something like offer a value to a channel, which receives the events in d, d ends up getting cancelled and nothing happens. I just want it so that no matter what happens in suspendFunB, it will not cancel anything else or pass an exception elsewhere

[Allan Wang]

And I wouldn’t want to wrap d in noncancellable because I do want all parts to cancel if the job.cancel() is called.