[OpenJDK 2D-Dev] <AWT Dev> [9] Review request for 8130136: Swing window sometimes fails to repaint partially when it becomes exposed

[Alexey Ivanov]

Hello Alex,

On 14.10.2015 17:11, Alexander Potochkin wrote:
> Hello Alexey
>
>>>>> When does the GetHWnd() method return zero?
>>>>> Do we need to check it in RemoveDC?
>>>> GetHWnd() cannot return NULL in WndProc. No NULL-check is necessary.
>>>
>>> As far as I can see AwtComponent::GetHWnd can return zero.
>>
>> Well, it could... but if it had done so, something bad would already 
>> have happened. The code in RemoveDC() won't cause an error if hWnd is 
>> NULL, however, an hDC could be leaked.
>>
>> Anyway, if GetHWnd() returns NULL during message processing, AWT is 
>> already screwed up, it would not be able to work correctly.
>
> I looked through the rest of the code, looks like your assumption is 
> correct.
>
> The fix looks good to me.

Thank you for your review.


Regards,
Alexey

>
> Thanks
> alexp
>
>>
>> Regards,
>> Alexey
>>
>>>
>>> Thanks
>>> alexp
>>>
>>>>
>>>>>
>>>>> I guess you didn't add the test for a good reason,
>>>>> could you please comment on that?
>>>> It's impossible to write a regression test because the failure 
>>>> could be caught only from the native AWT code, no user code could 
>>>> detect it. Additionally, the time it takes to reproduce the issue 
>>>> is indefinite.
>>>>
>>>> Regards,
>>>> Alexey
>>>>>
>>>>> Thanks
>>>>> alexp
>>>>>
>>>>> On 10/12/2015 12:11, Alexey Ivanov wrote:
>>>>>> Hello,
>>>>>>
>>>>>> Please review the fix for jdk9:
>>>>>>     bug: https://bugs.openjdk.java.net/browse/JDK-8130136
>>>>>>     webrev: 
>>>>>> http://cr.openjdk.java.net/~aivanov/8130136/jdk9/webrev.00/
>>>>>>
>>>>>>
>>>>>> Problem description:
>>>>>> Sometimes a window fails to repaint when it becomes exposed.
>>>>>>
>>>>>> When repaint failure occurs, SetDIBitsToDevice() in 
>>>>>> GDIBlitLoops_nativeBlit() returns 0 with error 87 ("Parameter 
>>>>>> Error"). It happens because invalid, already released, hDC is 
>>>>>> passed to SetDIBitsToDevice().
>>>>>>
>>>>>>
>>>>>> Root cause:
>>>>>> When a window is disposed, the DCs associated with it are 
>>>>>> released. The handles can be reused for another window.
>>>>>>
>>>>>> A thread disposes the window and releases hDC. At this time, DC 
>>>>>> handle is reused for another window. When Disposer thread cleans 
>>>>>> up for a died thread, it tries to release the cached hDC. But 
>>>>>> that handle is already associated with another window. Thus a 
>>>>>> valid pair hWnd/hDC becomes invalid: painting to released DC fails.
>>>>>>
>>>>>>
>>>>>> The fix:
>>>>>> Save both hDC and hWnd in the thread local storage, and pair hDC 
>>>>>> with hWnd when removing it to avoid accidentally releasing hDC 
>>>>>> which is already reused for another window.
>>>>>>
>>>>>>
>>>>>> Regards,
>>>>>> Alexey
>>>>>
>>>>
>>>
>>
>