Re: RFR 8209821: Make JVMTI GetClassLoaderClasses not walk CLDG

[serguei.spit...@oracle.com]

On 8/24/18 10:25, serguei.spit...@oracle.com wrote: Hi Coleen, It looks like a nice simplification and cleanup. Thank you for taking care about it! Could you, please, explain a little bit the changes in the getClassLoaderClasses? If I understand it correctly, the iteration ClassLoaderDataGraph::dictionary_all_entries_do(&JvmtiGetLoadedClassesClosure::increment_with_loader) Sorry, wrong copy, it had to be:   ClassLoaderDataGraph::dictionary_all_entries_do(&JvmtiGetLoadedClassesClosure::add_with_loader) Thanks, Serguei is replaced with the data->dictionary()->all_entries_do(&closure), or for boot class loader, with ClassLoaderData::the_null_class_loader_data()->dictionary()->all_entries_do(&closure). It is not clear (I have just some guesses) if it would do the same and why does it support concurrent class unloading for ZGC. Is it worth to add a minimal comment explaining this? Thanks, Serguei On 8/23/18 05:37, coleen.phillim...@oracle.com wrote: Summary: And also added function with KlassClosure to remove the hacks. There are about 10 vmTestbase/nsk/jvmti tests that test various parts of this change.  Also ran mach5 tier1-7. open webrev at http://cr.openjdk.java.net/~coleenp/8209821.01/webrev bug link https://bugs.openjdk.java.net/browse/JDK-8209821 Thanks, Coleen