RFR: 8221503: vmTestbase/nsk/jdb/eval/eval001/eval001.java failes with: com.sun.jdi.InvalidTypeException: Can't assign double[][][] to double[][][]

[David Holmes]

Hi Egor,

This doesn't seem right to me sorry ...

On 2/04/2019 7:42 pm, Egor Ushakov wrote:
> Please review the fix
> 
> this test started to fail after the fix of 
> https://bugs.openjdk.java.net/browse/JDK-8146986
> previously any call to ClassLoaderReference.visibleClasses as a side 
> effect cached the results in VirtualMachineImpl.typesBySignature
> and made them available to any VirtualMachine.classesByName call.
> This is somewhat not consistent for unprepared classes, I reported that, 
> check
> Bug 26148990 : JDI - VirtualMachine.allClasses does not return loaded 
> but uninitialized class,

That issue is now a duplicate of this core-libs issue:

https://bugs.openjdk.java.net/browse/JDK-8212117

which is in the process of being fixed. Once it is fixed then this 
workaround would no longer be needed.

That said I'm not at all clear how the proposed workaround actually 
works-around the underlying issue.

> but it seems that it allowed the ArrayTypeImpl.componentType to work 
> even if the type was not prepared,
> check how ArrayTypeImpl#findComponentType uses vm.classesByName.
> It seems that ArrayTypeImpl#findComponentType should simply call 
> findType for component signature.
> Just how the unused com.sun.tools.jdi.ArrayTypeImpl#type method works, 
> not sure why it was dropped...

The package-private:

Type type() throws ClassNotLoadedException {
     return findType(elementSignature());
}

seems to have always existed but never been used as the implementation 
for the public componentType() method (previously elementType()). In 
fact the two methods appear to have different requirements with regards 
to what they should be returning. AFAICS:

- ArrayTypeImpl.type() returns the actual array type
- ArrayTypeImpl.componentType() returns the actual ultimate component type

i.e if I understand correctly, given double[][][] then
- ArrayTypeImpl.type() == double[][]
- ArrayTypeImpl.componentType() == double

so implementing them both the same way can't possibly be correct!

> bugid https://bugs.openjdk.java.net/browse/JDK-8221503
> cr http://cr.openjdk.java.net/~eushakov/8221503/webrev.00/
> 
> Thanks!
> <http://cr.openjdk.java.net/~eushakov/8221503/webrev.00/>

You'd also need to update the ProblemList so the test gets re-enabled.

diff -r 532e88de77eb test/hotspot/jtreg/ProblemList.txt
--- a/test/hotspot/jtreg/ProblemList.txt
+++ b/test/hotspot/jtreg/ProblemList.txt
@@ -161,8 +161,6 @@
 
vmTestbase/nsk/jdi/VirtualMachine/redefineClasses/redefineclasses021/TestDescription.java 
8065773 generic-all
 
vmTestbase/nsk/jdi/VirtualMachine/redefineClasses/redefineclasses023/TestDescription.java 
8065773 generic-all

-vmTestbase/nsk/jdb/eval/eval001/eval001.java 8221503 generic-all
-
  vmTestbase/metaspace/gc/firstGC_10m/TestDescription.java 8208250 
generic-all
  vmTestbase/metaspace/gc/firstGC_50m/TestDescription.java 8208250 
generic-all
  vmTestbase/metaspace/gc/firstGC_99m/TestDescription.java 8208250 
generic-all

It also appears to me that the exception message of the 
ClassNotLoadedException will be different now. Have you re-run all the 
JDI tests?

Thanks,
David
-----

> 
> -- 
> Egor Ushakov
> Software Developer
> JetBrains
> http://www.jetbrains.com
> The Drive to Develop
>