On Tue, 3 Nov 2026 19:49:40 GMT, Phil Race <[email protected]> wrote:
>> The native code on macOS which initializes cached JNI refs is not thread
>> safe.
>> The fix is to use the macOS dispatch_once mechanism as a lock.
>> The fix is essentially all in JNIUtilities.h. There's some refactoring of
>> the macros in there to make things a bit cleaner
>> and DECLARE_*_ONLY macros are added which the non-local clients now use to
>> declare variables.
>>
>> The problem can be demonstrated with multi-threaded initialisation, and
>> using -Xcheck:jni
>> A case which reproduces "fairly often" has been added as a test.
>>
>> ---------
>> - [x] I confirm that I make this contribution in accordance with the
>> [OpenJDK Interim AI Policy](https://openjdk.org/legal/ai).
>
> Phil Race has updated the pull request incrementally with one additional
> commit since the last revision:
>
> 8393035
test/jdk/java/awt/FontMetrics/MultiThreadFontMetrics.java line 105:
> 103:
> 104: public void run() {
> 105: Graphics2D graphics = image.createGraphics();
main thread joins its worker/child threads but never captures their uncaught
exceptions. If createGraphics() or getFontMetrics() fails or throws exception,
the worker terminates, but join() will succeed,
and the child JVM can still exit with status zero as the parent checks only
that exit status,
so shouldn't we capture worker failures and rethrow them on the main thread?
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/33481#discussion_r4490499890