Reimplemented Workers using side effects (GUWT).

[zach-klippenstein]

Issues:

• Fixes Implement GUWT Workers #82, which is the Kotlin half of Implement keyed side effects in place of / as underpinnings to Worker (Grand Unified Workflow Theory) workflow#1021.
• Fixes Add a renderTester method for expecting typed workers #92.
• Fixes Kotlin workers are started even if scope is cancelled workflow#1197.

[zach-klippenstein]

Workers are a lot slower now, unfortunately. Fortunately, any optimizations we do for rendering Workflows will now also speed up workers, theoretically.

[zach-klippenstein]

This isn't the API I'd ideally like to have, but since we don't have access to the full reflection library in this module, it's the best we can do. If we did, we could create the full worker type from the worker's class and the worker's outputType property, or maybe using the reified T type parameter. But since we don't, all we have is typeOf(), which means we need to reify the entire worker type.

However, while less flexible, it has the nice property that the type that is used to compare the worker will always be exactly the type for the worker that the IDE is showing you when you pass it to runningWorker. There's no mystery or hidden implementation details anymore. You might need to pass a few more keys in, but that's probably a good idea anyway.

[rjrjr]

I'm unclear why these changes to true are okay.

[zach-klippenstein]

Good point. The thinking was that the old default, comparing classes, would always be true now since the runtime itself is ensuring that worker types match. However, we ended up only comparing on the declared type of the worker, which means two workers that the runtime thinks are equivalent could still be different classes, so it would be safer to leave the class checks in here.

That said, by defaulting to relying on concrete types again, we lose the nice simplicity for testing. Not sure what the best thing to do is.

[zach-klippenstein]

Reverted this default implementation, and added a bunch of tests around it. This implementation makes sense, because if the concrete type of a worker changes (e.g. was running a parent type, then runs a subtype with the same declared type), then the worker should restart. I need to document this somewhere…

[zach-klippenstein]

Additional documentation added to runningWorker and doesSameWorkAs.

[zach-klippenstein]

Just found a pretty bad bug that blocks the release, but I want to fix separately: #120