<Sound Dev> [9] Review request for 8168751: Two "Direct Clip" threads are created to play the same "AudioClip" object, what makes clip sound corrupted

[Anton Litvinov]

Hello Sergey,

Thank you for review of this fix. Yes, I agree that 4 fields of 
"DirectAudioDevice.DirectClip" class which you specified 
("loopStartFrame", "loopEndFrame", "newFramePosition", 
"clipBytePosition") are also not thread-safe and should be marked with 
"volatile" modifier, though they are not directly related to this bug.

The second version of the fix, which addresses your remark, was created. 
In the second version of the fix "volatile" modifier was added to the 
next 8 fields of "DirectClip" class, all of which are accessed both from 
"DirectClip.run()" method and other methods of this class: "audioData", 
"frameSize", "m_lengthInFrames", "loopCount", "clipBytePosition", 
"newFramePosition", "loopStartFrame", "loopEndFrame". Could you please 
review the second version of the fix.

Webrev: http://cr.openjdk.java.net/~alitvinov/8168751/jdk9/webrev.01

Thank you,
Anton

On 1/19/2017 12:34 AM, Sergey Bylokhov wrote:
> Hi, Anton.
> Probably some other fields in the DirectClip should be marked as volatile? For example loopStartFrame/loopEndFrame/ are used in the run() but assigned in the setLoopPoints
> () w/o any synchronization?(the same setFramePosition +newFramePosition/clipBytePosition)
>
>> Hello,
>>
>> Could you please review the following fix for the bug.
>>
>> Bug: https://bugs.openjdk.java.net/browse/JDK-8168751
>> Webrev: http://cr.openjdk.java.net/~alitvinov/8168751/jdk9/webrev.00
>>
>> The bug consists in the fact that after many repetitive sequential calls to the methods "loop()", "stop()" on the same instance of "java.applet.AudioClip" at some moment two threads with the name "Direct Clip", which play this "AudioClip" instance, are created and start existing in parallel, what makes the being played clip sound as corrupted.
>>
>> "Direct Clip" thread is represented by the instance variable "com.sun.media.sound.DirectAudioDevice.DirectClip.thread", is created in the method "com.sun.media.sound.DirectAudioDevice.DirectClip.open()" and is stopped by assigning "null" value to "DirectClip.thread" instance variable in the method "com.sun.media.sound.DirectAudioDevice.DirectClip.implClose()".
>>
>> THE DEFINED ROOT CAUSES OF THE BUG:
>>
>> 1. The fact that "DirectClip.thread" instance variable is not thread-safe. There is a possibility that null value assigned to it in one thread through the former mentioned "DirectClip.implClose()" method does not become visible for the running "Direct Clip" thread in the method "DirectClip.run()".
>>
>> 2. In the method "DirectClip.run()" not taking into account the fact that a new "Direct Clip" thread could have already been started and assigned to "DirectClip.thread" instance variable after the call to "DirectClip.implClose()", by the moment when "DirectClip.thread" value is checked for null in "DirectClip.run()" method, or by the moment when the previous "Direct Clip" thread is awakened and continues execution after the following code block in "DirectClip.run()" from the file "jdk/src/java.desktop/share/classes/com/sun/media/sound/DirectAudioDevice.java".
>>
>> 1354                 try {
>> 1355                     lock.wait();
>>
>> 1356                 } catch(InterruptedException ie) {}
>>
>> THE SOLUTION:
>> The solution is based on correction of the listed above 2 root causes of the bug.
>>
>> Thank you,
>> Anton