<Swing Dev> <SwingDev> [9] Review request: JDK-8058870 Mac: JFXPanel deadlocks in jnlp mode

[Anton V. Tarasov]

On 9/22/14 5:52 PM, Sergey Bylokhov wrote:
> Hi, Anton.
> I think after the fix,  getFlag+setFlag are not atomic anymore in the 
> revalidate().

I assumed revalidate() is called on EDT in case of pure Swing, and may 
be called from JavaFX App Thread in case of swing/javafx interop only.
However, I didn't take into account swing/awt mixing where this 
particular method is "allowed" to be called from non-EDT threads.
In this case, you're right, the getFlag+setFlag, being called on two 
concurrent threads, are not synchronized.

Ok, I'm still trying to keep the scheme lock-free. What do you think of 
the following approach?

http://cr.openjdk.java.net/~ant/JDK-8058870/webrev.1

We're eliminating the flag and using AtomicBoolean instead.

The "flags" field is private. REVALIDATE_RUNNABLE_SCHEDULED is only used 
for this single case. So, the only concern is serialization, for which I 
reinitialize the new field in readObject. At the worst case, we will 
have maximum two revalidate runnables in the queue, which shouldn't hit 
performance any noticeably (and as far as I understand, this code cares 
about performance only).

Thanks,
Anton.

>
> On 22.09.2014 17:36, Anton V. Tarasov wrote:
>> Hello,
>>
>> Please review a fix:
>>
>> https://bugs.openjdk.java.net/browse/JDK-8058870
>> http://cr.openjdk.java.net/~ant/JDK-8058870/webrev.0
>>
>> A deadlock has been revealed on MacOS X with JFXPanel involved in an 
>> app launched via JNLP. Find more details in jira, please.
>>
>> Thanks,
>> Anton.
>
>