Community
Participate
Working Groups
Created attachment 274774 [details] Ecore model used as test data I have noticed a NullPointerException when NavigationHelper.getInverseReferences is called: java.lang.NullPointerException at org.eclipse.viatra.query.runtime.base.core.EMFBaseIndexInstanceStore.getFeatureKeysPointingTo(EMFBaseIndexInstanceStore.java:303) at org.eclipse.viatra.query.runtime.base.core.NavigationHelperImpl.getSettingsForTarget(NavigationHelperImpl.java:403) at org.eclipse.viatra.query.runtime.base.core.NavigationHelperImpl.getInverseReferences(NavigationHelperImpl.java:398) I am calling this method to compute all objects referencing the given object. The bug appears calling the method on an object which is not referenced by another object. If called with an object which is referenced, the method works fine. Test setup consists of two resources containing one object each where one references the other (see attached cmcore.ecore for dummy test model). Configuration is plain default (Retebackend, no special flags set at all) but shouldn't influence the bug. After taking a look at the code, i assume the bug is in the following method in class org.eclipse.viatra.query.runtime.base.core.EMFBaseIndexInstanceStore#getFeatureKeysPointingTo. Patching the method public Set<Object> getFeatureKeysPointingTo(Object target) { return getValueToFeatureMap().get(target).distinctValues(); } with public Set< Object > getFeatureKeysPointingTo(final Object target) { final IMultiset< Object > multiset = getValueToFeatureMap().get(target); if (multiset == null) { return Collections.emptySet(); } else { return multiset.distinctValues(); } } seems to fix the problem. Also this kind of null check seems to have been added in similar methods like org.eclipse.viatra.query.runtime.base.core.EMFBaseIndexInstanceStore#getFeatureKeysPointingTo.FeatureData#getDistinctHoldersOfValue. I assume the bug has been introduced in Viatra 1.7 with the switch to EclipseCollections, since the code is working with Viatra 1.6 and the collections used there return empty collection instead of null per default. I flagged the bug as major, because this forces me to switch back to EcoreUtils#CrossReferencer or not use Viatra 2.0 (and therefor Eclipse Photon) but Eclipse Oxygen with Viatra 1.6.
Thanks for the report, and sorry for the inconvenience caused by the update. I tried to reproduce the issue, but when I tried to open the given Ecore model, I received the following two error messages: * `Resource '/de.tesis.dynaware.common.model/model/Base.ecore' does not exist.` * `java.net.MalformedURLException: Unsupported "platform:" protocol variation "de.tesis.dynaware.common.model".` Furthermore (assuming these errors can be safely ignored), I tried to find out what to do with the model, but the report was a bit unclear to me. It is possible that every required information is present and I have simply missed it, but I did not understand entirely what I have to do with the given example to reproduce the NullPointerException. Could you provide some more details, specifically: * How to open the test data correctly? * Exactly what operations are done on the model? * Exactly which model element should be queried/how? It would help a lot to understand what happens here if some executable version could be shared, or at least you could provide a code snippet showing precisely the code you are trying to execute here. Thanks in advance.
Regardless of the reproducability, I can confirm that there really is a missing null check there, and the solution would be similar to what Markus is suggesting.
Created attachment 274779 [details] Test project I have added a test project for Eclipse Photon. The test NavigationHelperTest should reproduce the bug in a standalone setup, just run it as Junit test. My environment was a standard Eclipse Photon vs. Running Platform but since the initial project was copied from the Viatra repository there shouldn't be any non-Viatra dependencies. The model is taken from the Viatra examples, master branch, plugin org.eclipse.viatra.examples.cps.model. It was changed in that way, that the EOpposite entry from EReference applications has been removed (otherwise it works fine). The test fails for me in line 68 where i request the references for the object hostInstance.
New Gerrit change created: https://git.eclipse.org/r/125496
@Markus: Thanks for the project, I have managed to reproduce the original issue this way and used that a motivation for a test case as well for this case. As Gábor said, your proposed fix also seems to be going into the right direction, and I have verified that the same map is used correctly in the file. The Gerrit change mentioned in the previous commit contains this fix, and it will be backported to a bugfix release for 2.0. We don't have any concrete plans when willthis bugfix version (2.0.1) available, but my guess is we wil do it a week or two. @Gábor: I have opened bug 536680 to solve this kind of issue in a more general way instead of providing this kind of code in many different places.
Perfect, thanks a lot for the fix.
Gerrit change https://git.eclipse.org/r/125496 was merged to [master]. Commit: http://git.eclipse.org/c/viatra/org.eclipse.viatra.git/commit/?id=8e8ac8802107a46a1ed43957dbcb981ee4c39e0a
New Gerrit change created: https://git.eclipse.org/r/125522
Gerrit change https://git.eclipse.org/r/125522 was merged to [2.0-maintenance]. Commit: http://git.eclipse.org/c/viatra/org.eclipse.viatra.git/commit/?id=ac3691822aa9d018035d4aa275dcefdb22f7d33e
Fixed in master and backported to 2.0-maintenance branch.
