Date: Fri, 27 Sep 1996 14:17:04 -0500
From: Max Hailperin <max@gac.edu>
To: f-math-computer-science-78@gac.edu
Subject: shutdown with pending puts

I mentioned in lab today that until one lab group raised it, I hadn't
considered the question of what should happen if a bounded buffer is
shutdown while one or more puts are waiting for space.  I said that
the "right" solution was to wait for the puts to complete before
actually doing the shutdown, but that it was OK with me if you just
ignored this issue.

Since then I've thought about this a bit more, and there are at least
four options (other than ignoring the issue) that are arguably
"right":

 (1) As suggested above, wait for the puts to complete before doing
     the actual shutdown.  Moreover, the call to the shutdown
     procedure shouldn't return until after the actual shutdown has
     occured (so it blocks until the pending puts are complete).

 (2) As above, wait for the puts to complete before doing the actual
     shutdown.  However, have the shutdown procedure return
     immediately anyway, even though the buffer won't really be
     shutdown until the last pending put is done.

 (3) Signal an error from the shutdown procedure if any puts are
     pending, analogous to the error in put if the buffer is already
     shut.

 (4) Instead, have shutdown wake up the pending puts and have them
     then signal the error (exactly as though the shutdown had been
     done before they were called).  If error signalling terminates
     the whole program (as currently) this is no different than 3, but
     in other circumstances it might matter whether the error was
     signaled from shutdown or from put.

I said that implementing the "right" solution would require a change
in the data structure (i.e., in the boundedbuf.h file).  As it turns
out, one of the above, namely option 4, doesn't require such a
change. It does, however, require a change to the code of the put
procedure I gave you.  All the others also require changes to the code
of put, and -- as far as I can see -- also changes to the data
structure.

As I said before, I will accept a solution of just ignoring this issue
-- so it would be a good thing to postpone.  However, I will also
accept any of the four options above, rather than just a single
"right" approach.  I would encourage you to think about the relative
merits of the four approaches, even if you don't implement any of
them.

 -max