This Bugzilla instance is a read-only archive of historic NetBeans bug reports. To report a bug in NetBeans please follow the project's instructions for reporting issues.
Finding Participant wanted to finish the session bean wizard but the "Scanning in progress..." inline error didn't disappear even though the scanning finished. Thus the finish button was still disabled. Severity 1 participant had to cancel the wizard because of this problem. Recommendation This is a bug. The inline error should disappear when the scanning finishes.
I've just tried. Saw the mentioned error message, but it really disappeared immediatelly after scanning disappeared and Finish button was dissabled. So probably already fixed. According to CVS annotation Andrei is probably the one who fixed it.
Sorry i have to reopen this, because same issue appears in new websvc/websvc client wizards. It is not easy to reproduce (don't have exact steps) but sometimes it happens. workaround for this is to type something to any textfield/combo in the wizard - then the inline error disappears
So I suppose that the code is copy-pasted everywhere. So as soon as we find the culprit we have to fix it on all places. I'm leaving this reassinged to me with Andrei on CC as a reviewer so we are sure we do not miss anything. Any information which would take RANDOM away is apprecitated. Probably some synchronization problem....
I shortly looked at the code in ClientInfo which displays the "Scanning in progress" message and I don't see anything wrong with it. But on the other hand I see no synchronization in JMManager.waitScanFinished() and JMManager.isScanInProgress(). Moreover, the condition of waitScanFinished() is different from the condition of isScanInProgress(). What is the threading model of these methods? I thought they are thread-safe, but the implementation doesn't look so. If by any chance it happens that isScanInProgress() returns true even after waitScanFinished() has returned then indeed the "Scanning in progress" message stays displayed. As a workaround we could do isWaitingForScan = false; before firing the change event. But I'd prefer to put a assert !JavaMetamodel.getManager().isScanInProgress(); there instead. Honzo, can you please comment?
Threading model is really stupid, but should work for UI purposes. waitScanFinished iterates through all unresolved roots and waits, till they are resolved, while isScanInProgress just shows, whether progress bar is still open. We don't have any synchronization intentionaly because we want to avoid deadlocks. It is possible, that isScanInProgress shows "true" *few ms* after waitScanFinished has ended. But I believe, that *few ms* should not cause any problems in UI.
There is no guarantee in the Java memory model that lack of synchronization works for UI purposes. Java memory model makes no difference beetween the AWT thread and other threads. Please see http://www.cs.umd.edu/~pugh/java/memoryModel/ for more information, especially http://www.cs.umd.edu/~pugh/java/memoryModel/jsr-133-faq.html There are two issues in this case: 1. The two methods are incorrectly synchronized, which can cause visibility problems (a thread can set ProgressDisplayer.instance to null, but another one can still read the old, non-null value). 2. Even if there was correct synchronization, it is unexpected isScanInProgress() returns true when waitScanFinished() has finished. It can be worked around, but it is counter-intuitive. Filed issue 79173.
So workarounding as described by Andrei (thanks for evaluation). We may rollback this one once the issue 79173 is fixed. j2ee/ejbcore/src/org/netbeans/modules/j2ee/ejbcore/ejb/wizard/entity/EntityEJBWizardDescriptor.java; new revision: 1.2.2.4.2.2; previous revision: 1.2.2.4.2.1 j2ee/ejbcore/src/org/netbeans/modules/j2ee/ejbcore/ejb/wizard/session/SessionEJBWizardDescriptor.java; new revision: 1.2.2.2.2.1; previous revision: 1.2.2.2 websvc/core/src/org/netbeans/modules/websvc/core/client/wizard/ClientInfo.java; new revision: 1.20.8.3.2.32; previous revision: 1.20.8.3.2.31 websvc/dev/src/org/netbeans/modules/websvc/dev/wizard/WebServiceFromWSDLPanel.java; new revision: 1.1.2.16; previous revision: 1.1.2.15 websvc/dev/src/org/netbeans/modules/websvc/dev/wizard/WebServiceTypePanel.java; new revision: 1.1.2.14; previous revision: 1.1.2.13
1. I agree, that issue 79173 is javacore bug 2. Imlementatation of "Scanning in progress..." should not be that difficult: waitScanFinished(); SwingUtilites.invokeLater(...removeScanningInProgress..); 3. I did not say, that lack of synchronization works for UI porposes. I said, that I believe, that these two methods works for UI purposes. We use them in refactoring, editor, navigator and other modules without problems. So at least there is an easy way how to implement it correctly.
> So at least there is an easy way how to implement it correctly. Yes, once you are aware of the bug then it is easy ;) The question is whether all those implemenation are really working correctly since this bug is quite random.