RFR: 8241160: Concurrent class unloading reports GCTraceTime events as JFR pause sub-phase events

[Kim Barrett]

> On Mar 26, 2020, at 5:34 AM, Stefan Karlsson <stefan.karlsson at oracle.com> wrote:
> 
> On 2020-03-25 23:37, Kim Barrett wrote:
>> src/hotspot/share/gc/shared/gcTraceSend.cpp
>>  306     assert(phase->level() < 2, "There is only two levels for ConcurrentPhase");
>> 
>> Seems like there ought to be a named constant for that "2", similar to
>> PhasesStack::PHASE_LEVELS used in a similar place in visit_pause. But
>> then, there's a mismatch between PHASE_LEVELS (6) and the number of
>> cases in visit_pause (4). That's "interesting"...
> 
> Yeah. visit_pause checks that we don't push too many levels of phases, but we already check that when we push phases:
> 
> […]
> I think it would make sense to remove that assert, to get rid of that confusion.
> 
> I'm not so sure about removing 'assert(phase->level() < 2' since it's there to to catch when we start to use deeper nesting of the concurrent phases. Maybe if we also add EventGCPhaseConcurrentLevel2-4 (analogous to visit_pause) and then get rid of this assert?
> 
> If we want to go ahead and do any of these changes (find a name for the 2 or adding events) I'll create a separate RFR.

Thanks for the walk-through.  Your suggestion seems reasonable.