[foreign-memaccess+abi] RFR: 8263018: Improve API for lifecycle of native resources [v2]

[Jorn Vernee]

On Wed, 10 Mar 2021 16:05:38 GMT, Maurizio Cimadamore <mcimadamore at openjdk.org> wrote:

>> src/jdk.incubator.foreign/share/classes/jdk/internal/foreign/ResourceList.java line 122:
>> 
>>> 120: 
>>> 121:         void cleanup() {
>>> 122:             if (fst != ResourceCleanup.DUMMY_CLEANUP) {
>> 
>> Shouldn't this be a volatile/acquire read?
>
> Probably - but, to be clear - at this point (and I need to add doc on this), we are only interested about add vs. close races - not close vs. close. So, the only "bad" thing that could happen is that some other thread adds to this list while we're closing it. So I guess in this case I omitted the read acquire because the only case where you could read a DUMMY_CLEANUP is if somebody has already stored that in there (otherwise you'll see something else, and move on). But no harm in making it more robust.
> 
> Another way we could go would be to manage races in the MemoryScope impl. That is, in a shared memory scope we could have some intermediate state (e.g. ADDING) which we go through and which is mutually exclusive w.r.t. CLOSE. I didn't want to go down that path since then in ResourceList I'd have to deal with add vs. add races anyway.

Ok, that makes sense. I don't think an ADDING state is needed, but would be nice if you could comment this part.

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

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