RFR: Setup process references and class unloading once before the cycle

[Roman Kennke]

Do I read this correctly that it moves the setup from VMThread to
SHConcThread ? Is this a good idea? Is it safe?

Roman

On Fri, Mar 2, 2018 at 12:49 PM, Aleksey Shipilev <shade at redhat.com> wrote:
> http://cr.openjdk.java.net/~shade/shenandoah/heuristics-pr-cu/webrev.01/
>
> We used to rely on CollectorPolicy process_references/unload_classes calls to be idempotent. But
> they are not required to be! For example, "aggressive" answers it randomly, and other heuristics are
> relying on inputs that might change abruptly.
>
> The cleaner solution would be to ask the policy once, before starting the cycle, and store that
> decision somewhere. It seems ShHeap itself is the best way to share this data. The methods in
> CollectorPolicy are renamed to capture accidental use. Getters in ShConcMark delegate back to heap
> for simplicity.
>
> Additionally, this allows us to print "Init Mark" with attributes, like we do with "Final Mark" today.
>
> Testing: hotspot_gc_shenandoah
>
> Thanks,
> -Aleksey
>