Description Sandip Chitale 2013-09-27 17:17:18 EDT

We have CLI (non-gui) Eclipse application. In the implementation of start() method we added a call to....

try {
    IJobManager jobManager = ...;
    jobManager.join(null);

} catch (...) {
}

to make sure that we wait for termination of all jobs and shutdown gracefully. When we did not have this code we use to see this message on the STDERR:

"Job found still running after platform shutdown.  Jobs should be canceled by the plugin that scheduled them during shutdown: "

which is emitted by this code:


org.eclipse.core.internal.jobs.JobManager:
:
:
:
      private void doShutdown() {
:
:
                        String msg = "Job found still running after platform shutdown.  Jobs should be canceled by the plugin that scheduled them during shutdown: " + jobName; //$NON-NLS-1$

:
:

                        // TODO the RuntimeLog.log in its current implementation won't produce a log 
                        // during this stage of shutdown. For now add a standard error output.
                        // One the logging story is improved, the System.err output below can be removed:
                        System.err.println(msg);

This message on STDERR use to fail our CLI application callers.

Anyhow it appears that the code in JobManager.join(null) manages a local HashSet of running/active jobs and only terminates when there is no job left in that set. Our suspicion is that there may be a bug in that logic.

----
                     jobs = Collections.synchronizedSet(new HashSet(select(family, states))); <------- All jobs are returned in a HashSet – please note that the select method build the list using a local variable.
                     jobCount = jobs.size();
                     if (jobCount > 0) {
                           //if there is only one blocking job, use it in the blockage callback below <----------- Code starting here to below
                           if (jobCount == 1)
                                  blocking = (Job) jobs.iterator().next();
                           listener = new JobChangeAdapter() {
                                  public void done(IJobChangeEvent event) {
                                         //don't remove from list if job is being rescheduled
                                         if (!((JobChangeEvent) event).reschedule)
                                                jobs.remove(event.getJob());
                                  }

                                  //update the list of jobs if new ones are added during the join
                                  public void scheduled(IJobChangeEvent event) {
                                         //don't add to list if job is being rescheduled
                                         if (((JobChangeEvent) event).reschedule)
                                                return;
                                         Job job = event.getJob();
                                         if (job.belongsTo(family))
                                                jobs.add(job);
                                  }
                           };
                           addJobChangeListener(listener); <------------ to here tries to then manage the addition/deletion and rescheduling of jobs 
-----

We suspect that because  I set a breakpoint at line:

:
:
:
                           Thread.sleep(100);
:
:


and evaluated the expression:

int states = suspended ? Job.RUNNING : Job.RUNNING | Job.WAITING | Job.SLEEPING;
select(family, states)

But it returned an empty list.