RFR: Refactoring GC phase and heap allocation tracking out of policy

[Zhengyu Gu]

On 09/23/2017 06:50 AM, Aleksey Shipilev wrote:
> On 09/23/2017 02:59 AM, Zhengyu Gu wrote:
>> Refactoring GC phase timing tracking and heap allocation tracking out of ShenandoahCollectorPolicy.
>>
>>
>> Webrev: http://cr.openjdk.java.net/~zgu/shenandoah/prof_refactor/webrev.00/index.html
> 
> I like it!
> 
> Would be interesting to try and conditionalize timing gathering on gc+stats. It seems wasteful to
> poll timers when we don't consume the data. Does not have to be in this patch though, but maybe the
> change should have that work in mind.

Our heuristics need timing feed also.

> 
> Few nits:
> 
>   *) ShenandoahGCPhaseTimings -> ShenandoahPhaseTimings?
> 
>   *) ShenandoahHeap::gc_phase_timings() -> ShenandoahHeap::phase_timings()?
> 
>   *) ShenandoahGCPhaseTimings::TimingPhase -> ShenandoahPhaseTimings::Phase?
> 
Okay.

>   *) Do we need the forward declaration in shenandoahCollectorPolicy.hpp? outputStream was used
> before in this class...
> 
>    41 class outputStream;

Well, I think we should avoid indirect dependencies ...

> 
>   *) outputStream must have cr() methods for this?
> 
>    1293     ls.print_cr("");
>    1294     ls.print_cr("");
> 
Fixed.

http://cr.openjdk.java.net/~zgu/shenandoah/prof_refactor/webrev.01/

Reran all tests.

Thanks,

-Zhengyu


> 
> Thanks,
> -Aleksey
>