I don't agree that whether you're using TM, shared-memory monitors, or actors with message queues is an implementation detail or that there is a better programming model that hides the differences between them. You can implement any of them on top of any of the others, but you're still programming to whatever model you put on top.

▲

vacuity 20 minutes ago | parent [-]

In the implementation that runs, yes, you have to choose something. However, I think the fundamental design is independent of those options, and probably should be developed independently.

	▲	kragen a minute ago \| parent [-]
		There is surely some sense in which that's true; the choice of concurrency primitives, even among such radically different choices, won't change literally every design decision in your system. But it is very pervasive.