I've already told one person that for that problem they should do it using a lock and a CV. So out of fairness I think that pronouncement should stand. It's only very slightly more complicated that way and it's a good exercise. -- - David A. Holland | VINO project home page: dholland(a)eecs.harvard.edu | http://www.eecs.harvard.edu/vino

so now we have: david: no. amos: yes. would another tf like to cast the deciding vote? -gwa- On Thu, 22 Feb 2001, Amos Blackman wrote:

we're supposed to simulate tf's answering questions with thread_yield. but we can only allow one person to change the contents of the message box at any given time. Should the tf who's answering a question release control of the mailbox while she's answering the question? this becomes a question of whether we want to be holding a lock while we thread_yield. it doesn't make much sense to do so, because no one else can do anything without the lock. cy

I guess my question was whether you considered the act of taking the quesiton out of the box as part of "answering the question". If you release the lock before you call thread_yield, it seems like you're taking the question out of the box, and then separately answering the question. you get situations like: teacher1 acquires the lock teacher1 takes question1 out of the box and releases the lock teacher2 acquires the lock teacher2 takes question2 out of the box and releases the lock student1 acquires the lock student1 submits question3 and releases the lock teacher3 acquires the lock teacher3 takes question3 out of the box and releases the lock teacher3 answers question3 (thread_yield) teacher1 answers question1 (thread_yield) teacher2 answers question2 (thread_yield) so the order that we see messages saying that a teacher answered a question has absolutely nothing to do with when it was taken out of the box. It seems like all the teachers could wait until the end of the program before answering any of the questions. cy On Thu, 22 Feb 2001, gwa wrote:

This is like the question in whalemating of whether matings can take place in parallel or not. The answer is the same: it's desirable to not hold locks on scarce or highly contended resources while performing slow operations. Mating is a fairly slow operation. So is answering a question, conceptually at least. On the other hand, in both cases, allowing the slow operations to occur in parallel makes the solution substantially more complicated. You don't have to implement the more elaborate solutions, but they're certainly not prohibited. If it bothers you, print an additional message when the teacher takes a message out of the box. -- - David A. Holland | VINO project home page: dholland(a)eecs.harvard.edu | http://www.eecs.harvard.edu/vino