Dan Sugalski wrote:
Whups, misunderstanding there. I realize that we need to throw an exception (or a junction of exception and not exception) if evaluating one of the junction members. The question is whether we should evaluate them all regardless and then figure it out at the end, and what to do with currently running junction evaluations if we've spawned off multiple threads to evaluate them in parallel. I expect I'm getting a bit too Quantum here, though.
Not at all. It's an important question, especially if the other threads have side effects. I suspect, however, that once one state of a junction throws an exception, we should just kill off the other states immediately.
I'm thinking that we shouldn't parallelize junction evaluation by default. Dealing with threads has too many issues that must be dealt with to spring it on unsuspecting programs.
Perhaps. But we need to think through the issues so that we can eventually move to threaded implementations without changing the semantics. Damian
