FULL PRODUCT VERSION :
java 1.6.038 64bit

ADDITIONAL OS VERSION INFORMATION :
linux red hat 6.3 64bit

A DESCRIPTION OF THE PROBLEM :
In package com.sun.corba.se.impl.encoding;
the BufferManagerReadStream's underflow method
contains the following code:

            while (fragmentQueue.size() == 0) {

                if (endOfStream) {
                      throw wrapper.endOfStream() ;
                }

                boolean interrupted = false;
                try {
                    fragmentQueue.wait(FRAGMENT_TIMEOUT);
                } catch (InterruptedException e) {
                    interrupted = true;
                }

                if (!interrupted && fragmentQueue.size() == 0) {
                    throw wrapper.bufferReadManagerTimeout();
                }

                if (receivedCancel) {
                    throw new RequestCanceledException(cancelReqId);
                }
            }

If a spurious wakeup happens, the queue size may still be zero resulting
in timeout exception.
This is not handled by the statment

                if (!interrupted && fragmentQueue.size() == 0) {
                    throw wrapper.bufferReadManagerTimeout();
                }

 Make sure timeout occured before exception is given.

Possible solution:

            while (fragmentQueue.size() == 0) {

                if (endOfStream) {
 throw wrapper.endOfStream() ;
                }

                boolean interrupted = false;
                long timeWaited = 0;
                try {
                     timeWaited = System.currentTimeMillis();
                     fragmentQueue.wait(FRAGMENT_TIMEOUT);
 timeWaited = System.currentTimeMillis() - timeWaited;
                } catch (InterruptedException e) {
                    interrupted = true;
                }

                if (!interrupted && fragmentQueue.size() == 0 && timeWaited >= FRAGMENT_TIMEOUT) {
                    throw wrapper.bufferReadManagerTimeout();
                }

                if (receivedCancel) {
                    throw new RequestCanceledException(cancelReqId);
                }
            }



REGRESSION. Last worked in version 5.0

STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
Code inspection.

ACTUAL -
Corba comm_failure timout occured randomly without any communication or memory problems.
Instrumentation of the code given the exception by measure time used in execution was done: found out that 0-1 ms was used when exception was given,
i.e nothing near the timeout of 60 sec used in code.

REPRODUCIBILITY :
This bug can be reproduced rarely.