Memory Leak - GC scope

Question

Hi,
I have a pervasive memory leak in my server application. I have used JProfiler to track it down and indeed after 10 or so hours of execution I can see clearly the Java classes which are increasing in count. For my application they are (in order of total memory size used):
byte[]
<class>[]
String
QueueLink(custom class)
Knowing where my system leaks memory is nice, but on trawling the code I cannot really see why it really leaks and therefore am unable to fix it. One thing I do not quite get about Java is the garbage collector scope. If I have a loop, like so below:
while (true)
Object o = new Object();
...do something with 'o'
then for every loop iteration 'o' is initialize to a new Object. But the old reference of o (from the previous iteration) is for all intents a purposes gone. BUT, does this get collected by the GC or does it only get collected when the scope leaves the while loop? If the latter then UNTIL the loop terminates I will "leak" memory, which would go part way to explaining my problem.
Any ideas, any suggestions on how I can fix my leak(s)?
Regards,
NC

Answer 1

... then for every loop iteration 'o' is initializeto a
new Object. But the old reference of o (from the
previous iteration) is for all intents a purposes
gone. BUT, does this get collected by the GC ordoes
it only get collected when the scope leaves thewhile
loop?It has nothing to do with variable scope in this
case. The object that 'o' used to reference is no
longer reachable (unless something else in that loop
made another reference to it, such as adding it to a
collection which is still reachable), so it is
eligible for GC.The question is, will it be GCed.
I remember hearing a few years back that the VM--or certain VMs--might not collect any eligible local variables until after the method completes, or something to that effect. If so, then an infinite while(true) { new Object(); } loop will exhaust the heap rather quickly. Even if that was true back then, I doubt it's true of Sun's VM today.