RFR: Make weakref processing always use Executor

Aleksey Shipilev shade at redhat.com
Mon Aug 13 10:58:47 UTC 2018


On 08/13/2018 12:45 PM, Roman Kennke wrote:
> You think? I think it's cleaner this way.
> If all GCs adopted this approach, it could simplify the code paths in RP a *lot*. It could be taken even further:
> - move 1 threaded execution as VMThread into WorkGang to begin with (thus avoiding this special casing in each GC)
> - fold upcoming MT-WeakProcessor into executor too, thus getting rid of the extra clutter to drive that.

Well, maybe. But current form looks like bending the RP mechanics. Especially when we "fake"
single-threaded execution by pretending it is the MT one, only to enter the MT codepath, which we
then hijack. It is far from being conceptually clean...

-Aleksey



More information about the shenandoah-dev mailing list