Re: Thread safe instance variables

["David Morris" <David.Morris@xxxxxxxxxxxxx>]

Larry,

It sounds like we are referring to similar techniques. What
you call a "shadow object" is what I was referring to with
a "writing to a non-synchronized clone". I don't see why
anyone would think this is a bad idea. I copied what I
saw was being done by Tomcat itself.

David Morris

>>> lwloen@us.ibm.com 04/17/02 12:25PM >>>

>I think I made the assumption that Daniel is talking about
>Servlets, where it makes sense to avoid synchronization and
>not Swing.

Well, in servlets, I am of the opinion that synchronization can be
kept
largely or entirely optional.  The way I would work around the
"multi-threading" of the servlet object architecture is that I create
a
"shadow" object (one-to-one with each servlet I need).  At servlet
entry,
I construct the shadow object normally or perhaps fetch (using
synchronize)
from some stash of recently used shadow objects.  I then initialize
the
shadow's instance variables with whatever I need.  It would typically
include stuff passed in, like the output and input objects, and so on.
This is how I think the servlet architecture should have been defined
to
start with.  I then do something like shadowobject.run() to do the
real
work of the servlet.

Since the object that does the real work "sits" on the stack, it is
inherently single-threaded (my code is started up in a single thread,
no
matter who else is logically capable of being launched in the
underlying
original Servlet object itself).  It is different in detail, but it is
functionally identical to implementing a run() method for a Runnable
object.

Perhaps someone can show that this is a bad idea, but it has worked
for
what I have done.  This restores normal object-oriented rules at any
rate
-- I can treat instance variables in the shadow object normally and
I'm
back to normal considerations between the styles 1, 2, or 3 of my
original
note if I need more than the original thread to get work done.

Now, if those "shadow" objects have to deal with state-ful beans of
some
kind, beans that live over the course of multiple servlet invocations,
then
I'll probably be synchronizing to reach those.  Ditto for anything that
for
some reason has to sit in the servlet object.  For the rest, I think I
can
be synchronize free and yet not have hard-to-think-about thread safety
considerations.

All this, as long as one can confine most or all of the work to the
shadow
object.  What am I missing, here, that this would not usually work?


Larry W. Loen  -   Senior Linux, Java, and iSeries Performance Analyst
                          Dept HP4, Rochester MN