JavaLeakTests failing on Java 23

[akurtakov]

Failures are like 'Can not get final java.lang.Class field java.lang.Class.componentType on java.lang.Class'

See https://download.eclipse.org/eclipse/downloads/drops4/I20240919-1230/testresults/html/org.eclipse.jdt.ui.tests_ep434I-unit-cen64-gtk3-java23_linux.gtk.x86_64_23.html for details.

[akurtakov]

@noopur2507 You seem to be the one involved with this failing on Java 17 so would you please look into it?

[noopur2507]

I am not sure if these were tests enabled based on the discussion in #23.

[noopur2507]

Also, I don't see any test failures in the last Y-build on gtk3-java23.
https://download.eclipse.org/eclipse/downloads/drops4/Y20240918-1000/testResults.php

@MohananRahul @stephan-herrmann any idea what changed?

[jukzi]

Stacktrace for the record:

Can not get final java.lang.Class field java.lang.Class.componentType on java.lang.Class

java.lang.IllegalArgumentException: Can not get final java.lang.Class field java.lang.Class.componentType on java.lang.Class
at java.base/jdk.internal.reflect.MethodHandleFieldAccessorImpl.newGetIllegalArgumentException(MethodHandleFieldAccessorImpl.java:86)
at java.base/jdk.internal.reflect.MethodHandleObjectFieldAccessorImpl.get(MethodHandleObjectFieldAccessorImpl.java:61)
at java.base/java.lang.reflect.Field.get(Field.java:444)
at org.eclipse.jdt.ui.leaktest.reftracker.ReferenceTracker.followFieldReference(ReferenceTracker.java:80)
at org.eclipse.jdt.ui.leaktest.reftracker.ReferenceTracker.visit(ReferenceTracker.java:144)
at org.eclipse.jdt.ui.leaktest.reftracker.ReferenceTracker.start(ReferenceTracker.java:168)
at org.eclipse.jdt.ui.leaktest.LeakTestCase.collect(LeakTestCase.java:58)
at org.eclipse.jdt.ui.leaktest.LeakTestCase.assertInstanceCount(LeakTestCase.java:123)
at org.eclipse.jdt.ui.tests.leaks.JavaLeakTest.internalTestEditorOpen(JavaLeakTest.java:378)
at org.eclipse.jdt.ui.tests.leaks.JavaLeakTest.testJavaEditorActionDelegate(JavaLeakTest.java:465)
at java.base/java.lang.reflect.Method.invoke(Method.java:580)

the field is rather old:
https://github.com/openjdk/jdk/blame/e087edeb256a9743d1fdb6c295cb5add78d4552e/src/java.base/share/classes/java/lang/Class.java#L1411

The IllegalArgumentException hides a ClassCastException which would be nice to have::
https://github.com/openjdk/jdk/blob/e087edeb256a9743d1fdb6c295cb5add78d4552e/src/java.base/share/classes/jdk/internal/reflect/MethodHandleObjectFieldAccessorImpl.java#L60

looks like a error in JDK

java.vm.version=23+37-2369

the Y build instead uses
java.vm.version=23-ea+27-2275

[stephan-herrmann]

Also, I don't see any test failures in the last Y-build on gtk3-java23. https://download.eclipse.org/eclipse/downloads/drops4/Y20240918-1000/testResults.php

@MohananRahul @stephan-herrmann any idea what changed?

I can locally reproduce on jdk-23+37.

The original ClassCastException reads: Cannot cast [I to java.lang.Class

And indeed the debugger shows Class.componentType to hold as its value an empty array.

Am I having a deja vu? See #2 (comment) :)

[jukzi]

I can locally reproduce on jdk-23+37.

do you agree its a JDK bug? Mind to file a error report to JDK how to reproduce?

[stephan-herrmann]

I can locally reproduce on jdk-23+37.

do you agree its a JDK bug?

That's the feeling I already had in February ...

Mind to file a error report to JDK how to reproduce?

I don't think Oracle / openjdk wants to run JavaLeakTests, and atm I have no idea how to reduce this to a small reproducer.

Reminds me, I still need to file the recent JIT bug - let's see how they'll react to that one.

[stephan-herrmann]

in #2 I mentioned that the bug occurs with JDK 18+. This isn't wrong, but interestingly, from those JDKs 18+ I have on disk, 20, 21 and 22 do not show the bug, but 23 does again.

I should add, however, that I didn't repeat each of the test runs to check if the bug happens / does not happen reliably. I gave up partly because since today (?) equinox startup takes more than a minute on my machine with several cpu cores at 100%. Anybody else seeing this?

[jukzi]

created https://bugs.openjdk.org/browse/JDK-8341343 for Can not get final java.lang.Class field java.lang.Class.componentType on java.lang.Class

[stephan-herrmann]

created https://bugs.openjdk.org/browse/JDK-8341343 for Can not get final java.lang.Class field java.lang.Class.componentType on java.lang.Class

cool, thanks.
The linked bugs JDK-8301663 (fixed in 20) and JDK-8337622 (fixed in 24) correspond to my observations about versions where it was fixed, then broken again.

The good news, once we build on JDK 24 leak tests should be functional again.

[jukzi]

Asked to updated JDK: https://gitlab.eclipse.org/eclipsefdn/helpdesk/-/issues/5179

[noopur2507]

These tests are now failing on win32-java17 also since I20241024-1800.

https://download.eclipse.org/eclipse/downloads/drops4/I20241029-1800/testResults.php

[iloveeclipse]

These tests are now failing on win32-java17 also since I20241024-1800.

https://download.eclipse.org/eclipse/downloads/drops4/I20241029-1800/testResults.php

I believe this could be Windows specific UI issue, not Java bug like on Java 23 build

[jukzi]

the fails on windows are unrelated to this Java 23 issue. I created a new ticket for it: #1752

[jukzi]

#1789 worked around this JDK error. now jdt.ui tests have 0 fails in https://download.eclipse.org/eclipse/downloads/drops4/I20241111-1800/testresults/html/org.eclipse.jdt.ui.tests_ep434I-unit-linux-x86_64-java23_linux.gtk.x86_64_23.html