[aarch64-port-dev ] RFR: Bulk integration of Shenandoah 2018-05-15

[Roman Kennke]

Am 17.05.2018 um 19:51 schrieb Aleksey Shipilev:
> On 05/17/2018 07:41 PM, Andrew Haley wrote:
>> That looks reasonable.
> 
> Have you reviewed the entire patch already? Oh wow, that was fast! I suspected it would take a full
> day to read through them.
> 
>> Is there a well-defined process for handling changes as the GC
>> interface progresses in upstream trunk?  Is it a matter of
>> back-porting the GC interface to JDK 8, or do you still do things the
>> "old" way?
> 
> We are trying our best not to touch the shared parts in backports, preferring to be slightly less
> performance efficient with the benefits for our own development sanity. Backporting the JDK 10+ GC
> interface to JDK 8 would be an insane endeavor.
> 
> So, sh/jdk8, sh/jdk9, sh/jdk10, sh/jdk all touch the runtime in slightly different ways, but not
> frustratingly different. Most of the barrier logic is encapsulated in ShenandoahBarrierSet and
> friends. The actual calls around the runtime are one-liner differences.
> 

We also have a bunch of compat-wrappers e.g. for new hotspot logging
which is completely absent in jdk8 last time I looked.

Roman