Re: [jdt-dev] Why does JDT subscribe to JVM class unload events?

[Andrey Loskutov <loskutov@xxxxxx>]

see commented out code in addCachedMirror(ReferenceTypeImpl). This also could explain the classes leak we observe if debugger is attached. 

Am 17. September 2019 20:13:00 MESZ schrieb Simeon Andreev <simeon.danailov.andreev@xxxxxxxxx>:
>Hi all,
>
>why does JDT subscribe to JVM class unload events after attaching to a
>JVM
>to debug it? It subscribes with the following code:
>
>org.eclipse.jdi.internal.VirtualMachineImpl.packetSendManager()
>
>// TBD: This call should be moved to addKnownRefType() when it can
>// be made specific
>// for a reference type.
>eventRequestManagerImpl().enableInternalClasUnloadEvent();
>
>/**
>* Creates ClassUnloadRequest for maintaining class information for
>within
>JDI.
>* Needed to known when to flush the cache.
>*/
>public void enableInternalClasUnloadEvent(/* TBD: ReferenceTypeImpl
>refType*/) {
>   // Note that these requests are not stored in the set of outstanding
>requests because
>     // they must be invisible from outside.
>     ClassUnloadRequestImpl reqUnload = new
>ClassUnloadRequestImpl(virtualMachineImpl());
>     reqUnload.setGeneratedInside();
>     // TBD: It is now yet possible to only ask for unload events for
>     // classes that we know of due to a limitation in the J9 VM.
>     // reqUnload.addClassFilter(refType);
>     reqUnload.setSuspendPolicy(EventRequest.SUSPEND_NONE);
>     reqUnload.enable();
>}
>
>Unfortunately I'm not seeing which cache is being flushed (see Java doc
>above) from just the code above. I'll keep looking but if someone knows
>it
>would help speed-up the process a lot.
>
>*Background:*
>
>We recently observed that
>https://bugs.openjdk.java.net/browse/JDK-8227269
>hinders performance in our product.
>
>This has to do with the JVM keeping track of loaded classes, in order
>to
>notify the attached debugger about unloaded classes. We then observed
>that
>JDT does care about class unload events, notifications are requested as
>described above.
>
>The code is present from the very first git commit:
>http://git.eclipse.org/c/jdt/eclipse.jdt.debug.git/diff/org.eclipse.jdt.debug/jdi/org/eclipse/jdi/internal/VirtualMachineImpl.java?id=3c7939022de48bfa72324a4d2451d33cfb56a6d0
>
>Best regards and thanks,
>Simeon

--
Kind regards,
Andrey Loskutov

https://www.eclipse.org/user/aloskutov
Спасение утопающих - дело рук самих утопающих