Is Reactor.doLaunch too opinionated about dispatchers?

[square]

Is Reactor.doLaunch too opinionated about dispatchers?

[zach-klippenstein]

Specifically looking for feedback from @rjrjr @loganj @vRallev @afollestad.

[rjrjr]

Of the people you pinged I'm probably the least qualified to respond, so I'll ignore that and answer anyway.

I definitely gravitate to option 3, since it involves the least departure from standard co-routine practice.

Between your suboptions 3.i (default to EmptyCoroutineContext like usual, and don't prematurely optimize doLaunch), and 3.ii (default to Unconfined, since that's been working okay so far, and is likely the 99% case), 3.ii sounds like the right thing. I'm less confident there, but it seems to strike a balance: it does the not-quite-normal-but-right-for-reactors thing that we're already doing, while deviating from standard practice in as minimal a way as possible.

One question: is this difficulty a sign that forcing doLaunch and onReact to share a single dispatcher is a mistake?

[zach-klippenstein]

is this difficulty a sign that forcing doLaunch and onReact to share a single dispatcher is a mistake?

I mean, that's basically the heart of the question I'm trying to answer.

Since the doLaunch machinery doesn't care, I don't think it makes sense to put the effort in to avoid dispatch unless we can prove there's actually a benefit to doing so.

[zach-klippenstein]

Being addressed in the rewrite.