Community
Participate
Working Groups
Created attachment 135077 [details] bnd output of org.apache.servicemix.specs.jaxb-api-2.1-1.3.0.jar Eclipse Version: 3.4.2 Build id: M20090211-1700 java version "1.6.0_0" IcedTea6 1.4 (fedora-15.b14.fc10-x86_64) Runtime Environment (build 1.6.0_0-b14) OpenJDK 64-Bit Server VM (build 14.0-b08, mixed mode) I have a Target Runtime Definition with Equinox 3.4.2 plus > 160 other bundles installed. When I add the following bundles: org.apache.servicemix.specs.jaxb-api-2.1-1.3.0.jar (javax.xml.bind.*) org.apache.servicemix.specs.jaxp-api-1.4-1.3.0.jar (javax.xml.*) org.apache.servicemix.specs.jaxws-api-2.1-1.3.0.jar (javax.xml.ws.*) and click the 'Set as Target Platform' link in the Target Platform Definition editor PDE hangs during the 'Resetting Target Platform' phase. These bundles can be found in the Apache ServiceMix 4 download. I can attach if requested to do so. Steps to reproduce: 1. Download Equinox 3.4.2 SDK and unzip in <equinox_home> 2. Download Apache ServiceMix 4 and copy <smix4_base>/system/**/*.jar to <equinox_home>/plugins 3. Move the 3 mentioned files out of <equinox_home>/plugins 4. Define a new Target Platform with <equinox_home> as the 'Location' 5. Click 'Set as Target Platform' 6. The Target Platform is successfully reset 7. Move the 3 mentioned files in to <equinox_home>/plugins 8. Click 'Set as Target Platform' 9. CPU is 100%, UI is non-responsive There is no output in the console log. I have tried setting the Execution Environment to JavaSE-1.6 and JavaSE-1.5 but the behavior is the same for both. I have attached the bnd print output for the 3 bundles.
Created attachment 135078 [details] bnd output of org.apache.servicemix.specs.jaxp-api-1.4-1.3.0.jar
Created attachment 135079 [details] bnd output of org.apache.servicemix.specs.jaxws-api-2.1-1.3.0.jar
I have since tried several combinations of Target Platform bundles: 1. WORKING: Equinox 3.4.2 SDK + org.apache.servicemix.specs.jaxb-api-2.1-1.3.0.jar, org.apache.servicemix.specs.jaxp-api-1.4-1.3.0.jar, org.apache.servicemix.specs.jaxws-api-2.1-1.3.0.jar It appears that the hang only occurs if there is something that uses the packages exported from these bundles.
Moving to PDE/UI
I created a unit test to see what is going on during execution. The method call that sends PDE in to a seemingly infinite loop is: org.eclipse.osgi.internal.module.ResolverImpl.resolveBundles(ResolverBundle[], Dictionary[], ArrayList) I suspect there is an infinite recursion while checking 'uses' (you can see this from the stack trace). The interesting thing is that I can get all of these bundles to work outside of PDE and run Apache ServiceMix using Equinox. Here is a stack trace when I set a breakpoint in ResolverImpl. It seems that resolveBundles0 is being recursively called and never terminates: Thread [main] (Suspended (breakpoint at line 622 in ResolverImpl)) ResolverImpl.findBestCombination(ResolverBundle[], ResolverConstraint[][], int[], ArrayList) line: 622 ResolverImpl.findBestCombination(ResolverBundle[]) line: 594 ResolverImpl.checkUsesConstraints(ResolverBundle[], Dictionary[], ArrayList) line: 539 ResolverImpl.resolveBundles0(ResolverBundle[], Dictionary[], ArrayList) line: 535 ResolverImpl.checkUsesConstraints(ResolverBundle[], Dictionary[], ArrayList) line: 573 ResolverImpl.resolveBundles0(ResolverBundle[], Dictionary[], ArrayList) line: 535 ResolverImpl.checkUsesConstraints(ResolverBundle[], Dictionary[], ArrayList) line: 573 ResolverImpl.resolveBundles0(ResolverBundle[], Dictionary[], ArrayList) line: 535 ResolverImpl.checkUsesConstraints(ResolverBundle[], Dictionary[], ArrayList) line: 573 ResolverImpl.resolveBundles0(ResolverBundle[], Dictionary[], ArrayList) line: 535 ResolverImpl.resolveBundles(ResolverBundle[], Dictionary[], ArrayList) line: 501 ResolverImpl.resolve(BundleDescription[], Dictionary[]) line: 388 UserState(StateImpl).resolve(boolean, BundleDescription[]) line: 428 UserState(StateImpl).resolve(boolean) line: 488 PDEState(MinimalState).internalResolveState(boolean) line: 206 PDEState(MinimalState).resolveState(boolean) line: 201 PDEState.readTargetState(URL[], IProgressMonitor) line: 121 PDEState.<init>(URL[], URL[], boolean, IProgressMonitor) line: 98 PDEState.<init>(URL[], boolean, IProgressMonitor) line: 90 LoadTargetOperation.handleReload(String, List, Preferences, IProgressMonitor) line: 234 LoadTargetOperation.loadPlugins(Preferences, IProgressMonitor) line: 147 LoadTargetOperation.run(IProgressMonitor) line: 53 LoadTargetDefinitionTest.testLoadTarget() line: 36 NativeMethodAccessorImpl.invoke0(Method, Object, Object[]) line: not available [native method] NativeMethodAccessorImpl.invoke(Object, Object[]) line: 57 DelegatingMethodAccessorImpl.invoke(Object, Object[]) line: 43 Method.invoke(Object, Object...) line: 616 LoadTargetDefinitionTest(TestCase).runTest() line: 164 LoadTargetDefinitionTest(TestCase).runBare() line: 130 TestResult$1.protect() line: 106 TestResult.runProtected(Test, Protectable) line: 124 TestResult.run(TestCase) line: 109 LoadTargetDefinitionTest(TestCase).run(TestResult) line: 120 TestSuite.runTest(Test, TestResult) line: 230 TestSuite.run(TestResult) line: 225 JUnit3TestReference.run(TestExecution) line: 130 TestExecution.run(ITestReference[]) line: 38 RemotePluginTestRunner(RemoteTestRunner).runTests(String[], String, TestExecution) line: 460 RemotePluginTestRunner(RemoteTestRunner).runTests(TestExecution) line: 673 RemotePluginTestRunner(RemoteTestRunner).run() line: 386 RemotePluginTestRunner.main(String[]) line: 62 UITestApplication$1.run() line: 114 RunnableLock.run() line: 35 UISynchronizer(Synchronizer).runAsyncMessages(boolean) line: 133 Display.runAsyncMessages(boolean) line: 3378 Display.readAndDispatch() line: 3036 Workbench.runEventLoop(Window$IExceptionHandler, Display) line: 2384 Workbench.runUI() line: 2348 Workbench.access$4(Workbench) line: 2200 Workbench$5.run() line: 495 Realm.runWithDefault(Realm, Runnable) line: 288 Workbench.createAndRunWorkbench(Display, WorkbenchAdvisor) line: 490 PlatformUI.createAndRunWorkbench(Display, WorkbenchAdvisor) line: 149 IDEApplication.start(IApplicationContext) line: 113 UITestApplication.start(IApplicationContext) line: 46 EclipseAppHandle.run(Object) line: 193 EclipseAppLauncher.runApplication(Object) line: 110 EclipseAppLauncher.start(Object) line: 79 EclipseStarter.run(Object) line: 386 EclipseStarter.run(String[], Runnable) line: 179 NativeMethodAccessorImpl.invoke0(Method, Object, Object[]) line: not available [native method] NativeMethodAccessorImpl.invoke(Object, Object[]) line: 57 DelegatingMethodAccessorImpl.invoke(Object, Object[]) line: 43 Method.invoke(Object, Object...) line: 616 Main.invokeFramework(String[], URL[]) line: 549 Main.basicRun(String[]) line: 504 Main.run(String[]) line: 1236 Main.main(String[]) line: 1212
Move to Equinox/Framework
Ok, last update. I left the unit test running (I can provide it if necessary all it does it call LoadTargetOperation.run()) and it did succeed. The test ran for 1,032.159 seconds which is about 17 minutes. That is far too long for a test/dev cycle. Is there a way for me to narrow down which combination of bundles is causing this search to take so long?
After tracing through the code I found: osgi.resolver.usesMode which I set to ignore in my PDE JUnit test launch configuration. This fixes the problem in that the unit test takes 4.39 seconds but I'm not sure what other issues this property is going to introduce. I also don't know how to set this so that Eclipse PDE uses it when 'Set as Target Platform' is clicked or the Target Platform preferences panel is updated.
Tom, didn't you fix an issue around 'uses' recently? Wondering if this is a dupe.
(In reply to comment #9) > Tom, didn't you fix an issue around 'uses' recently? Wondering if this is a > dupe. > No, not recently. Will have to investigate, but we are kinda stuck until we can really focus on a true "uses" clause solution. This wont happen in 3.5.
I'm a bit perplexed by 'uses' Does any framework have a good solution for this problem? If not, we should have a general recommendation of people to not have 'uses' clauses in their manifests.
I tried this on 3.5. I'm not seeing anything near 17 minutes but it could be some ordering thing. I am also surprised by 17 minutes since we should timeout way before that (after 90 seconds). That timeout is implemented in 3.4.0.
(In reply to comment #12) > I tried this on 3.5. I'm not seeing anything near 17 minutes but it could be > some ordering thing. I am also surprised by 17 minutes since we should timeout > way before that (after 90 seconds). That timeout is implemented in 3.4.0. > I just duplicated this on my Mac with Eclipse 3.4.2: 1. Download eclipse-equinox-SDK-3.4.2.zip 2. Download apache-servicemix-4.0.0.tar.gz 3. Unpack both 4. cd eclipse/plugins 5. cp `find <path to servicemix>/system -name "*.jar" | xargs` . 6. Window -> Preferences -> Target Platform 7. Browse to the unpacked eclipse/plugins 8. Click Ok 9. 'Reading Target Platform...' has been going for 8 minutes now.
Created attachment 135174 [details] List of all plugins being loaded in the target platform
(In reply to comment #11) > I'm a bit perplexed by 'uses' > > Does any framework have a good solution for this problem? If not, we should > have a general recommendation of people to not have 'uses' clauses in their > manifests. > Could this be solved by having a PDE preference that sets osgi.resolver.usesMode ? I don't think it is going to be feasible to have people avoid 'uses' clauses; especially in light of bnd generating them by default.
(In reply to comment #12) > I tried this on 3.5. I'm not seeing anything near 17 minutes but it could be > some ordering thing. I am also surprised by 17 minutes since we should timeout > way before that (after 90 seconds). That timeout is implemented in 3.4.0. > I just tried this on a vanilla 3.5M7 and 3.4.2 SR1 Mac OS X: 1. 3.5M7 works perfectly; the new target platform management system is nice (new, empty workspace). 2. Vanilla 3.4.2 (eclipse-rcp-ganymede-SR1-macosx-carbon.tar.gz) still has the issue (new, empty workspace) 3. 3.4.2 + all my plugins but with a new, empty workspace still has the issue.
> Tom, didn't you fix an issue around 'uses' recently? Wondering if this is a > dupe. > This could be related to bug 272561(In reply to comment #9) Do you see anything new errors in your log file when you reproduce on 3.4.x?
(In reply to comment #17) > > Tom, didn't you fix an issue around 'uses' recently? Wondering if this is a > > dupe. > > > > This could be related to bug 272561(In reply to comment #9) > > Do you see anything new errors in your log file when you reproduce on 3.4.x? > I am in the habit of running Eclipse from the commandline via ./eclipse -consolelog .. I didn't see the StackOverflow there or in the .log file of my workspace.
*** This bug has been marked as a duplicate of bug 261849 ***
