Invoking a coroutine with a custom continuation

[Simn]

I have a case in #119 where I'm in a coroutine and want to call another coroutine, but with a different continuation. So something like this:

import haxe.coro.CoroIntrinsics;
import hxcoro.Coro.*;
import hxcoro.CoroRun;
import hxcoro.concurrent.CoroLatch;
import hxcoro.continuations.FunctionContinuation;

function main() {

	@:coroutine function g() {
		trace("g()");
		yield();
	}

	@:coroutine function f() {
		trace("f()");

		final latch = new CoroLatch(1);
		final myCont = new FunctionContinuation(CoroIntrinsics.getContext(), (_, _) -> {
			trace("Called!");
			latch.arrive(1);
		});

		// workaround
		(() -> g(myCont))();

		latch.wait();
	}
	CoroRun.run(f);
}

This works, but I wonder if there should be a more obvious way of doing this.

[Aidan63]

I was going to suggest something like this

	@:coroutine function f() {
		trace("f()");

		final latch = new CoroLatch(1);

        suspend(cont -> {
            final myCont = new FunctionContinuation(cont.context, (r, e) -> {
                trace("Called!");
                latch.arrive(1);

                cont.resume(r, e);
		    });

            g(myCont);
        });

		latch.wait();
	}

It printed out f(), g(), and then Called! on cpp, but then I noticed the exit code from the program was not zero... So I tried the JVM target, and on that I see f(), g(), and then it hangs.
Not looked into it yet but my immediate suspicion is something around the return type from g(myCont), I don't think we're technically allowed to return a value from suspend, but I guess it sort of works for some targets?

[Simn]

I think you found some legit multi-threading problem here because this only fails when using more than one thread. I'll try to find out what actually happens, it seems very odd that the control flow in g falls through after the yield() but doesn't resume the continuation.