On Thu, 18 Jun 2026 14:11:51 GMT, Chen Liang <[email protected]> wrote:
>> src/java.base/share/classes/java/lang/invoke/X-VarHandle.java.template line
>> 619:
>>
>>> 617: @ForceInline
>>> 618: static boolean compareAndSet(VarHandle ob, $type$ expected,
>>> $type$ value) {
>>> 619: FieldStaticReadWrite handle = (FieldStaticReadWrite)
>>> ob.onStaticFieldAccess(true, true);
>>
>> CompareAndSet do both a read and write, so here you pass `true, true`. This
>> will end up calling `notifyStrictStaticAccess` which then clears the 'unset'
>> bit of the field for writes. However, a CAS is not guaranteed to do a write.
>> Isn't that problematic? The VM would consider the field initialized after a
>> failed CAS.
>>
>> Would be nice if you could add a test for that.
>
> I can degrade the calls to be reporting only read for cas: after the first
> write, we don't really care about future writes.
Right, but shouldn't we be able to read after a _successful_ CAS? Reporting the
write seems needed as well.
I just had a look with a debugger, and it seems the ordering of the read check
before the write check in `initialize` saves us. The read check will throw an
exception, so the write check (which clears the bit) never actually happens.
I think it would still be good to add a test case for CAS, and maybe a comment
that the read check should be ordered before the write check in `initialize`.
-------------
PR Review Comment:
https://git.openjdk.org/valhalla/pull/2550#discussion_r3436492893
