Community
Participate
Working Groups
Ant had a similar failure in two consecutive builds. The problem ocurrs after the test is complete and the debugger is being termianted. Looks like a possible synch problem in the Ant debug model (where a field is being nulled out as the target is shutdown, but still being used by another thread). java.lang.NullPointerException at org.eclipse.ant.internal.launching.debug.model.RemoteAntDebugBuildListener.sendRequest(RemoteAntDebugBuildListener.java:175) at org.eclipse.ant.internal.launching.debug.model.RemoteAntDebugBuildListener.resume(RemoteAntDebugBuildListener.java:219) at org.eclipse.ant.internal.launching.debug.model.AntDebugTarget.resume(AntDebugTarget.java:197) at org.eclipse.ant.internal.launching.debug.model.AntDebugTarget.terminated(AntDebugTarget.java:344) at org.eclipse.ant.internal.launching.debug.model.AntDebugTarget.terminate(AntDebugTarget.java:168) at org.eclipse.ant.tests.ui.debug.AbstractAntDebugTest.terminateAndRemove(AbstractAntDebugTest.java:526) at org.eclipse.ant.tests.ui.debug.AbstractAntDebugTest.terminateAndRemove(AbstractAntDebugTest.java:500) at org.eclipse.ant.tests.ui.debug.SteppingTests.debugStack(SteppingTests.java:127) at org.eclipse.ant.tests.ui.debug.SteppingTests.antCallStack(SteppingTests.java:111) at org.eclipse.ant.tests.ui.debug.SteppingTests.antCallStack(SteppingTests.java:106) at org.eclipse.ant.tests.ui.debug.SteppingTests.testStepBackFromAntCallSepVM(SteppingTests.java:39) at org.eclipse.ant.tests.ui.AbstractAntUIBuildTest.access$0(AbstractAntUIBuildTest.java:1) at org.eclipse.ant.tests.ui.AbstractAntUIBuildTest$1.run(AbstractAntUIBuildTest.java:44) at java.lang.Thread.run(Thread.java:595)
build: I20100513-1500
failed again on Windows in I20100907-0800
SteppingTests#testStepBackFromAntCallSepVM failed on N20110827-2000 not with the exact callstack but probably because of the same reason. Here is the call stack. ###### java.lang.NullPointerException at org.eclipse.ant.internal.launching.debug.model.AntDebugTarget.terminated(AntDebugTarget.java:345) at org.eclipse.ant.internal.launching.debug.model.AntDebugTarget.terminate(AntDebugTarget.java:168) at org.eclipse.ant.tests.ui.debug.AbstractAntDebugTest.terminateAndRemove(AbstractAntDebugTest.java:526) at org.eclipse.ant.tests.ui.debug.AbstractAntDebugTest.terminateAndRemove(AbstractAntDebugTest.java:500) at org.eclipse.ant.tests.ui.debug.SteppingTests.debugStack(SteppingTests.java:127) at org.eclipse.ant.tests.ui.debug.SteppingTests.antCallStack(SteppingTests.java:111) at org.eclipse.ant.tests.ui.debug.SteppingTests.antCallStack(SteppingTests.java:106) at org.eclipse.ant.tests.ui.debug.SteppingTests.testStepBackFromAntCallSepVM(SteppingTests.java:39) at org.eclipse.ant.tests.ui.AbstractAntUIBuildTest.access$0(AbstractAntUIBuildTest.java:1) at org.eclipse.ant.tests.ui.AbstractAntUIBuildTest$1.run(AbstractAntUIBuildTest.java:44) at java.lang.Thread.run(Thread.java:619) ######## AntDebugTarget.terminated is not synchronized and I believe that is the reason that there is another thread that would have got in. I will investigate more.
Yes, terminated gets called from two other places. ######## Daemon Thread [Ant Build Server Connection] (Suspended (breakpoint at line 341 in AntDebugTarget)) AntDebugTarget.terminated() line: 341 AntDebugTarget.terminate() line: 168 RemoteAntDebugBuildListener.shutDown() line: 191 RemoteAntBuildListener$ServerConnection.run() line: 109 ######### Thread [Worker-0] (Suspended (breakpoint at line 341 in AntDebugTarget)) AntDebugTarget.terminated() line: 341 AntDebugTarget.handleDebugEvents(DebugEvent[]) line: 454 DebugPlugin$EventNotifier.run() line: 1117 SafeRunner.run(ISafeRunnable) line: 42 DebugPlugin$EventNotifier.dispatch(DebugEvent[]) line: 1151 DebugPlugin$EventDispatchJob.run(IProgressMonitor) line: 415 Worker.run() line: 54 ######### Though the call stacks in comment 0 and comment 3 are from the tests, as they are two more stacks just from the code, we should better synchronize the terminated function.
Created attachment 202340 [details] Proposed patch Made terminated() synchronized
Mike, Can you please review the patch?
> Made terminated() synchronized I think there is more to it than that. The problem arises because we null out the fController object in terminated(), so it is possible that any one of the public methods using it could be called asynchronously and cause the NPE. For example comment #0 shows it coming from a call to resume(). A better solution might be to just not null out fController in terminated(). AntDebugTargets are not reused and to create a debug target you have to pass it a new controller.
Actually, the call stack in comment 0, shows resume from terminated(). Actually, the source had subsequentally changed, but both the calls are ineffect coming from terminated. However, you are right that there is more to this and there could be some sync problems when terminated is called at the time of other actions. One example I can think of is the target ending at the same time a suspend is called. I think just not nulling fController may just push the NPE to another point, something like the first two methods of the callstack in comment 0. So, I think RemoteAntDebugListener.sendRequest() should also check properly.
Created attachment 202733 [details] Proposed patch Patch as for my last comment.
Mike, Can you review this?
+1, the patch makes sense. I can't say definitively if it fixes the tests since they don't fail for me locally. Pushed fix to master: http://git.eclipse.org/c/platform/eclipse.platform.git/commit/?id=b88619d3d013f09e240bb10a2d0e03291ed29015
