Bug 129743

Created attachment 58126 [details]
Screenshot

fixed

Verified in NetBeans IDE Dev (Build 200803140008)

I can reproduce this bug in 200803200007 build

Unpredictable behavior of PropertySheet.setNodes(Node[] nodes);
Method sometimes set null content.
Method is invoked in EventDispatchThread.
This is fine trace of right method behavior:
-------------------------
FINE [org.openide.explorer.propertysheet.PropertySheet]: SetNodes
[org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@13858b1[Name=Sheet, displayName=Sheet]]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Detaching listeners from
org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@afac1d[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]:
SetCurrentNode:org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@afac1d[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Now listening for changes on
org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@afac1d[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Delayed updater setting nodes to
[org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@13858b1[Name=Sheet, displayName=Sheet]]
FINE [org.openide.explorer.propertysheet.PropertySheet]:
SetCurrentNode:org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@13858b1[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Detaching listeners from
org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@afac1d[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Now listening for changes on
org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@13858b1[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Detaching listeners from
org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@13858b1[Name=Sheet, displayName=Sheet]
-------------------------
This is fine trace of wrong method behavior:
-------------------------
FINE [org.openide.explorer.propertysheet.PropertySheet]: SetNodes
[org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@4a8e91eb[Name=Sheet, displayName=Sheet]]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Detaching listeners from
org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@674e5fac[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]:
SetCurrentNode:org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@674e5fac[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Now listening for changes on
org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@674e5fac[Name=Sheet, displayName=Sheet]
SEVERE [global]
java.lang.NullPointerException
       at java.util.Arrays$ArrayList.<init>(Arrays.java:3357)
       at java.util.Arrays.asList(Arrays.java:3343)
       at org.openide.explorer.propertysheet.PropertySheet$2$1.run(PropertySheet.java:591)
       at java.awt.event.InvocationEvent.dispatch(InvocationEvent.java:209)
       at java.awt.EventQueue.dispatchEvent(EventQueue.java:597)
       at org.netbeans.core.TimableEventQueue.dispatchEvent(TimableEventQueue.java:104)
[catch] at java.awt.EventDispatchThread.pumpOneEventForFilters(EventDispatchThread.java:273)
       at java.awt.EventDispatchThread.pumpEventsForFilter(EventDispatchThread.java:183)
       at java.awt.EventDispatchThread.pumpEventsForHierarchy(EventDispatchThread.java:173)
       at java.awt.EventDispatchThread.pumpEvents(EventDispatchThread.java:168)
       at java.awt.EventDispatchThread.pumpEvents(EventDispatchThread.java:160)
       at java.awt.EventDispatchThread.run(EventDispatchThread.java:121)
-------------------------

it seems a result of fixing of #126818 Mar 19, 2008
because property sheet worked right before

fixed

376a484b0204

You fixed NPE that do not have any impact on user because it reproduced if fine log switched on.
Issue is: Changes in Mar 20, 2008 make a behavior of the method PropertySheet.setNodes(Node[] nodes) unpredictable.
In some computers second method call bring EMPTY property sheet content.
For example:
On Solaris method works right.
On Linix and Windows method works wrong.

I think that problem is in a changeset:
http://hg.netbeans.org/main/rev/bbc5d9e69e9d
in line:
- final Node[] nodes = helperNodes;
The changeset was done for IZ#126818

NPE in fine log was fixed.
This is a fine log of two showing of property sheet in options dialog.
First invocation of options dialog sets right property sheet content.
The second invocation of options dialog sets NULL property sheet content.

First invocation:
FINE [org.openide.explorer.propertysheet.PropertySheet]: SetNodes
[org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@fc4825[Name=Sheet, displayName=Sheet]]
FINE [org.openide.explorer.propertysheet.PropertySheet]:  Scheduling delayed update of selected nodes.
FINE [org.openide.explorer.propertysheet.PropertySheet]: Delayed updater setting nodes to
[org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@fc4825[Name=Sheet, displayName=Sheet]]
FINE [org.openide.explorer.propertysheet.PropertySheet]:
SetCurrentNode:org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@fc4825[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Now listening for changes on
org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@fc4825[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Delayed updater setting nodes to
[org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@fc4825[Name=Sheet, displayName=Sheet]]
FINE [org.openide.explorer.propertysheet.PropertySheet]:
SetCurrentNode:org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@fc4825[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Detaching listeners from
org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@fc4825[Name=Sheet, displayName=Sheet]

Second invocation:
FINE [org.openide.explorer.propertysheet.PropertySheet]:
SetCurrentNode:org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@fc4825[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Now listening for changes on
org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@fc4825[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]: SetNodes
[org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@17a03d[Name=Sheet, displayName=Sheet]]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Detaching listeners from
org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@fc4825[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]:
SetCurrentNode:org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@fc4825[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Now listening for changes on
org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@fc4825[Name=Sheet, displayName=Sheet]
FINE [org.openide.explorer.propertysheet.PropertySheet]: Delayed updater setting nodes to null
FINE [org.openide.explorer.propertysheet.PropertySheet]: Detaching listeners from
org.netbeans.modules.cnd.editor.options.EditorPropertySheet$DummyNode@fc4825[Name=Sheet, displayName=Sheet]

are there any step to reproduce? it works fine for me (win xp, jdk1.5)

Scenario:
- Push 'Tools|Options' menu item
- Click on 'C/C++' button
- Select 'Formatting Style' tab
Property sheet is right.
- Close dialog by 'OK'.
- Push 'Tools|Options' menu item
Property sheet on 'Formatting Style' tab empty.
Repeated on Windows XP an LUNIX

I thinhk that bug reproducibility depends on CPU speed.

I see the empty property sheet everytime I bring up the panel. I'm on a MacBook Pro and running Mac OS X.

It seem old core problem.
See also IZ#105525

after some digging i've found out that this is actually a problem in options dialog: when the options window is created
it also creates its category panels and thats when setNodes() is called on the property sheet. however when invoked for
the second time, category panels are created again (including setNodes()) but then the options window activates the
panel that was selected the last time. but the panel already is active so it is unnecessarily removed and added to
component hierarchy. it means that removeNotify() is called on the propertysheet which correctly clears any context
created by previous setNodes(). 
reassigning to options dialog, below is a simple patch that fixes the problem

diff -r aa2c78f9446e options.api/src/org/netbeans/modules/options/OptionsPanel.java
--- a/options.api/src/org/netbeans/modules/options/OptionsPanel.java    Wed Mar 26 09:43:20 2008 +0100
+++ b/options.api/src/org/netbeans/modules/options/OptionsPanel.java    Wed Mar 26 11:04:50 2008 +0100
@@ -166,8 +166,10 @@
         JComponent component = category.getComponent();
         category.update(controllerListener, false);
         final Dimension size = component.getSize();
-        pOptions.add(component, category.getCategoryName());
-        cLayout.show(pOptions, category.getCategoryName());
+        if( component.getParent() == null || !pOptions.equals(component.getParent()) ) {
+            pOptions.add(component, category.getCategoryName());
+            cLayout.show(pOptions, category.getCategoryName());
+        }
         checkSize (size);
         /*if (CategoryModel.getInstance().getCurrent() != null) {
             ((CategoryButton) buttons.get (CategoryModel.getInstance().getCurrentCategoryID())).requestFocus();

stando, go ahead and fix it. I don't expect radek to work on it anymore

fixed - however i was hoping somebody would do a sanity check. who's taking care of spi.options now anyway?

58019eb6fd4c

I do not see changeset 58019eb6fd4c in main.
options.api/src/org/netbeans/modules/options/OptionsPanel.java also is not changed.
Do you forget to push changes?

I tested the proposed fix on my Windows XP. It works.

Thomas (on a MacBook Pro and running Mac OS X):
I brought over his change and it does seem to work.... 

I see fix in main now:
http://hg.netbeans.org/main/rev/58019eb6fd4c
It seems to work.

works on my machine also (verified in 20080327062852 build)

"Debug Executable" dialog in sunstudio would sometimes get <No Properties>

bits that can reproduce problem :
/net/balui/kits/compilers/pibs/sstrunk_masters/dbx/dbxtool/inst/$VARIANT1.inst//opt/SUNWspro//bin/sunstudio

- bring up dialog from Run | Debug Executable
- enter some valid executable in "Executable" textfield
- PropertySheet would change to <No Properties>

- or switching among Project drop down would reproduce problem too.

-----------------------------------------------------------------------------------------------------------------------
Here is the stack trace when things don't work, note, "nodes" is null

[t@30 l@5]: print nodes
org.openide.nodes.Node[] nodes = null
[t@30 l@5]: where 
=>[1] org.openide.explorer.propertysheet.PropertySheet.doSetNodes(this = <OBJECT>, nodes = <ARRAY>) 
"PropertySheet.java":437
  [2] org.openide.explorer.propertysheet.PropertySheet.access$100(x0 = <OBJECT>, x1 = <ARRAY>)  "PropertySheet.java":117
  [3] org.openide.explorer.propertysheet.PropertySheet.2.1.run(this = <OBJECT>)  "PropertySheet.java":597
  [4] java.awt.event.InvocationEvent.dispatch(--- Not compiled -g ---)  "InvocationEvent.java":209
  [5] java.awt.EventQueue.dispatchEvent(--- Not compiled -g ---)  "EventQueue.java":461
  [6] org.netbeans.core.TimableEventQueue.dispatchEvent(--- Not compiled -g ---)  "TimableEventQueue.java":9
  [7] java.awt.EventDispatchThread.pumpOneEventForHierarchy(--- Not compiled -g ---)  "EventDispatchThread.java":242
  [8] java.awt.EventDispatchThread.pumpEventsForHierarchy(--- Not compiled -g ---)  "EventDispatchThread.java":163
  [9] java.awt.EventDispatchThread.pumpEventsForHierarchy(--- Not compiled -g ---)  "EventDispatchThread.java":153
  [10] java.awt.Dialog.1.run(--- Not compiled -g ---)  "Dialog.java":517
  [11] java.awt.Dialog.2.run(--- Not compiled -g ---)  "Dialog.java":545
  [12] Java_java_security_AccessController_doPrivileged__Ljava_security_PrivilegedAction_2(0x9145f20, 0xd01ff328,
0xd01ff330, 0xd01ff304, 0x0, 0xd01ff330, 0xeca833c0, 0x0, 0xeca82e30, 0xd01ff330, 0xd01ff350, 0xf94029a4, 0xeca83368,
0xf9407053, 0xd9a36088, 0xd01ff334, 0xecfa2c6f, 0xd01ff378, 0xee1cc858, 0x0), at 0xfe7388e2
  [13] java.awt.Dialog.show(--- Not compiled -g ---)  "Dialog.java":543
  [14] java.awt.Component.show(--- Not compiled -g ---)  "Component.java":1300
  [15] java.awt.Component.setVisible(--- Not compiled -g ---)  "Component.java":1253
  [16] com.sun.tools.debugger.dbxgui.debugger.actions.DebugExecutableNodeAction.perform(this = <OBJECT>, executablePath
= <null>)  "DebugExecutableNodeAction.java":109
  [17] com.sun.tools.debugger.dbxgui.debugger.actions.DebugExecutableNodeAction.performAction(this = <OBJECT>,
activatedNodes = <ARRAY>)  "DebugExecutableNodeAction.java":80

Some more insight into the actual problem.
Chihin and I stepped through our reproducible testcase and here is what we found:

PropertySheet.setNodes(node) will assign it's parameter, nodes, to
class member helperNodes and schedule a task with a delay. When
the test runs it calls doSetNodes(helperNodes). 

The reason various dialogs end up with an empty property sheet
is that by the time doSetNodes(helperNodes) is c alled helperNodes
is null.

Why is that? A quick search shows that helperNodes gets set to null in 
removeNotify(). Among other times removeNotify() gets called when 
the sheet (or one of it's containers) get removed from their containers.

Here's one scenario:
Panel panel2 = new Panel();
Panel panel1 = new Panel();

Sheet sheet = new Sheet()
sheet.setNodes()
panel1.add(sheet)
panel2.add(panel1)

# after a while ... like when the model underlying the sheet changes

sheet = new Sheet()
(1) sheet.setNodes()
panel1.add(sheet)
(2) panel2.add(panel1)

At time (1) a task gets sceduled to eventually call doSetNodes(helperNodes)
At time (2) a panel1 which already exists in panel2 is being re-added
to panel2. It gets removed and then re-added. As it gets removed helperNodes goes
to null. Because it's the same object being re-added the delayed doSetNodes()
gets the null helperNodes.

Putting an invokeLater around (1) helps because it postpones the setting of
helperNodes til after them getting cleared at (2).
In general inserting delays in client code should not fix this problem.
A delay after (1) just makes the problem worse. A delay after (2) is too
late.
Making sure that one doesn't re-add a panel is probably the most
reliable work around. That was the "fix" for this bug and that is what Chihin
is verifying in dbxgui. But IMO this is still a workaround.

At the moment I would state this bug as follows: PropertySheet does
not behave correctly when it or one of it's containers is redundantly 
added to a container. 

JavaDoc for Container.addImpl() says:
    Note that if the component already exists in this container or a child
    of this container, it is removed from that container before being added to
    this container.
So,
a) redundant adds are anticipated and allowed.
b) The removal behaviour of add is not capricious.
Ergo OptionDialog and Chihins Dialog are not doing anything wrong.
It is ProperytSheet which should work correctly.

fixed, setNodes() now survives multiple addNotify()/removeNotify() calls, weak ref is used to prevent memory leaks

01115076c999

I installed 200804250004 build and I see empty table in 'Formatting Style' tab again

Scenario:
- Run NetBeans with C/C++ pack with fresh user dir
- Push 'Tools|Options' menu item
- Click on 'C/C++' button
- Select 'Formatting Style' tab
==> Table is empty

well, it's working fine for me in custom build from today's sources nb080425, win xp, jdk1.5.0_13
i also checked the source files and the fix is there...

I verified on Windows (java 1.5.0_09), Linux and Solaris (1.6.0_04). And on all platforms I see empty table

I upgraded this bug to P1, because I can not launch a lot of automatic tests on nightly builds (although on NetBeans 6.1
release all test worked correctly).

First scenario:
- Launch NetBeans with C/C++ pack
- Push 'Tools|Options' menu item
- Click on 'C/C++' button
- Select 'Formatting Style' tab
==> Table is empty

Other scenario:
- Push 'Tools|Options' menu item
- Click on 'C/C++' button
- Select 'Formatting Style' tab
- Switch formatting style on 'ANSI' or 'Apache' for example (repeat this step if table is not emty)
==> I see empty table in ~50% cases

can QA try reproducing this issue pls? everything looks fine on my machine...
(doesn't seem like a P1 to me after reading priority guidelines though:)

I have tried to reproduce on Ubuntu, Mac OS X and Win XP but I cannot reproduce. Hopefully Petr Ch. will be more
successful in reproducing this.

reproduced in:

Product Version: NetBeans IDE Dev (Build 200805050002)
Java: 1.6.0_05; Java HotSpot(TM) Client VM 10.0-b19
System: Linux version 2.6.24-17-generic running on i386; UTF-8; en_US (nb)
Userdir: /home/petr/.netbeans/dev-0505

It's Xubuntu 8.04 on Core2Duo, dual monitor.
We were also able to reproduced it on Kubuntu 8.04 on different machine.

also reproduced on Solaris:

Product Version: NetBeans IDE Dev (Build 200805051203)
Java: 1.6.0_04; Java HotSpot(TM) Client VM 10.0-b19
System: SunOS version 5.10 running on x86; UTF-8; cs_CZ (nb)
Userdir: /home/petr/.netbeans/dev

Standa, 
please could you look at this again ? There is a request from Sun Studion to have it fixed in patch 2. Petr reproduced
the bug... Thanks in advance.

fixed (hopefully) - ed78f533aed4

QA, pls also test there's no regression to #125057

Integrated into 'main-golden', available in NB_Trunk_Production #237 build
Changeset: http://hg.netbeans.org/main/rev/6df061db4f5c
User: Alexander Simon <alexvsimon@netbeans.org>
Log: remove work around after fixing IZ#129743

Sun Studio team, does the fix work for you?

Yes, C/C++ Formating styles work fine.

I'm not able to verify in sunstudio/dbxgui yet, for I need to wait for Thomas to upgrade netbeans bit to this version
for sunstudio. I'll do the verification once he upgrade NB.

It may take a while because SS Express probably won't be based on Patch 2 after all. Cut-off date is now June 12 and Patch 2 won't be ready by then.

I will suggest you download a trunk build of NB and stick it into your own SS build and test it this way. You can find daily trunk builds here:
/net/phoenix.czech/space/builds/netbeans/trunk/daily 

i've got a trunk build ... rebuilding ... gimme a sec.

We're not in a position to verify it with dbxgui bits.
The symptoms don't arise in the regular IDE, only in dbxtool
which uses a subset of NB modules. As best as I can tell
the latest trunk modules have a rearranged dependency so our
dbxtool module-pick-and-chooser ends up not picking enough
modules.
But the main sunstudio IDE (debugger area) is not impacted.
We'll be able to verify again when we get the new binary bits 
and have some time to spend on the module dependency issue.

C/C++ Formating styles work fine.

All problems with table which I saw in NetBeans are fixed.

The fix has been ported into the release61_fixes branch:

http://hg.netbeans.org/release61_fixes/8eaa5c793f54