[foreign-memaccess+abi] RFR: 8263018: Improve API for lifecycle of native resources [v10]

[Paul Sandoz]

On Fri, 12 Mar 2021 11:19:08 GMT, Maurizio Cimadamore <mcimadamore at openjdk.org> wrote:

>> I believe here there is some effort towards making everything thread-safe - in the sense that, based on the scope provided by the client (confined or shared) a different kind of allocator impl will be returned to the user. A shared bounded arena has a synchronized allocation method - while a shared unbound segment uses multiple underlying thread-local allocators.
>> 
>> Perhaps you refer to the fact that we don't check (in confined cases) that we are in the same thread as the scope's thread _before_ responding to an allocation request?
>
> I've added confinement check in latest iteration

Ok, i realized that i missed the `synchronized` on `BoundedSharedArenaAllocator.allocate`.

Perhaps we need to say in the `SegmentAllocator` something like:

If the allocator's resource scope is shared, then allocation is thread-safe and may be performed concurrently. If the resource scope is confined then allocation is confined to the owning thread of the allocator's resource scope.

And for the case of `SegmentAllocator.of(MemorySegment segment)` the allocator's resource scope is the scope of the given segment. 

Re-reading the specification i am confused by "The segment returned by this method is associated with a segment which cannot be closed".

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

PR: https://git.openjdk.java.net/panama-foreign/pull/466