RFR: 8303467: Serial: Refactor reference processor [v2]

[Albert Mingkun Yang]

On Fri, 3 Mar 2023 15:06:25 GMT, Thomas Schatzl <tschatzl at openjdk.org> wrote:

>> src/hotspot/share/gc/serial/markSweep.cpp line 246:
>> 
>>> 244:   MarkSweep::_string_dedup_requests = new StringDedup::Requests();
>>> 245: 
>>> 246:   static AlwaysTrueClosure always_true;
>> 
>> Please put this static into the class as static class member; it's very hard to track if statics are put into random places.
>> Why we can use `AlwaysTrueClosure` here should probably be documented.
>> The reference processor for `MarkSweep` can also be a static, as well as `MarkAndPushClosure` initialized statically with that reference processor.
>
>> Please put this static into the class as static class member; it's very hard to track if statics are put into random places. 
> 
> This refers to the `AlwaysTrueClosure`.
> 
>>Why we can use `AlwaysTrueClosure` here should probably be documented.
> 
> Something like: "Mark&Sweep collection covers the whole heap so all objects should be subject to discovery"
> 
>>The reference processor for `MarkSweep` can also be a static, as well as `MarkAndPushClosure` initialized statically with that reference processor.
> 
> Here I ommitted the important part: `_ref_processor` does not need to be a reference/pointer any more and we can get rid of the new/delete (I think).

Directly embedding `ReferenceProcessor` as a field doesn't work, because its constructor uses static variables (e.g. `_default_soft_ref_policy`) that are lazily initialized.

-------------

PR: https://git.openjdk.org/jdk/pull/12809