Transactional Memory

I just read "Composable memory transactions" (Harris, Marlow, Peyton-Jones, Herlihy) on Friday. It's really very similar to what Henry Baker was talking about and what I had in mind for the transactionality of mutable state in ThiNG, but it's different enough to make me consider changing ThiNG to be more like the language they discuss in the paper. The advantage is that you can build communication primitives out of transactional shared memory - see their discussion of MVars, which are essentially identical to the cells I've been considering for ThiNG - whereas it's more difficult to do things the other way around.