RFR (M) JDK-8151178: Move the collection set out of the G1 collector policy

[Mikael Gerdin]

Hi Thomas,

On 2016-03-07 15:25, Thomas Schatzl wrote:
> Hi Mikael,
>
> On Mon, 2016-03-07 at 10:46 +0100, Mikael Gerdin wrote:
>> Hi all,
>>
>> Here's a new version of the webrev based on feedback from Jesper and
>> Tom.
>>
>> http://cr.openjdk.java.net/~mgerdin/8151178/version2/
>>
>> The incremental layout is similar to version1, with the addition of
>> webrev_4 which contains the review changes.
>>
>> In this update I also took the liberty to change G1CollectionSet to
>> become a value member in G1CollectedHeap instead of a separately C
>> -heap
>> allocated object. I see no value in it being allocated separately.
>
>    some minor issues:
>
> - maybe the include changes in g1CollectorState.hpp should be a
> separate CR, but then again, it's very small.

I had to change g1CollectorState.hpp, I could not include it otherwise 
(unless including allocation.hpp before). I took the liberty to sort it 
while I was editing its include list.

>
> - instead of G1CollectionSet::set_g1p() I would have chosen
>   G1CollectionSet::set_policy() for a name.

Hm, yeah I was torn on naming this. On one hand we usually name 
accessors after the member variable name. On the other hand I agree that 
set_g1p is not a nice name. I also thought about "initialize" or 
something, but that makes it sound like it does a lot more.

Maybe I should just rename it _policy an have a set_policy?

>
> - as a future cleanup suggestion, maybe the CollectionSetChooser could
> be one of the first targets, and probably hidden from direct access by
> the G1CollectorPolicy. Not sure.

It might be possible, but it's not on my refactor list at this time, 
unfortunately.

>
> - g1CollectorPolicy.cpp:977 - superfluous newline ;)

Will fix.

Thanks for the review.

/Mikael

>
> Feel free to implement any of these suggestions, no need for re-review.
>
> Thanks,
>    Thomas
>