RFR: Rework ClassUnloading* flags handling

[Aleksey Shipilev]

On 05/18/2018 04:40 PM, Roman Kennke wrote:
> Am 18.05.2018 um 16:26 schrieb Aleksey Shipilev:
>> http://cr.openjdk.java.net/~shade/shenandoah/class-unload-rework/webrev.01/
>>
>> We have lost some of this during upstream merges, and we have not noticed because no tests verified
>> how we handle ClassUnloading* flags. This patch makes sure ClassUnloading* flags are treated well by
>> Shenandoah in every mode. New tests capture the intended behavior, including OOME when class
>> unloading is disabled and classes leak.
>>
>> Testing: hotspot_gc_shenandoah
>>
>> Thanks,
>> -Aleksey
>>
> 
> One question: it does not disable concurrent class unloading by default,
> right? I'm asking because if I understood correctly, you meant to do that?

Yes, separately, because it has performance implications that I still study. After this
infrastructure fix, making class unloading opt-in is a two-liner fix.

-Aleksey