<AWT Dev> [8] Review request for 8005465: [macosx] Evaluate if checking for the -XstartOnFirstThread is still needed in awt.m

[Sergey Bylokhov]

Hi, Anthony.
Fix looks good.

16.01.2013 19:18, Anthony Petrov wrote:
> Thanks for the review, Artem.
>
> Can I get at least one more review from AWT folks please?
>
> -- 
> best regards,
> Anthony
>
> On 1/16/2013 18:54, Artem Ananiev wrote:
>>
>> Thanks for clarification. Looks fine then.
>>
>> Artem
>>
>> On 1/16/2013 6:41 PM, Anthony Petrov wrote:
>>> Well, yes. In fact, we don't add the loop observers since 7u6 (see
>>> 7144542), so my fix implements the same behavior for JDK 8 as well.
>>>
>>> Actually, AWT doesn't need run loop observers in the headless mode. An
>>> app using the headless AWT controls when it should exit, not the
>>> AWTAutoShutdown. In some cases the native event loop may not even be
>>> running in the headless mode. At least AWT itself never runs it, so
>>> there's nothing to observe actually. The event loop is running when a
>>> headless AWT is embedded into another GUI toolkit (FX or SWT), but 
>>> again
>>> AWT itself in the headless mode doesn't care about the run loop 
>>> business.
>>>
>>> Given that the fix has been in the wild since 7u6 and no bugs have been
>>> identified, I assume this is a safe change.
>>>
>>> -- 
>>> best regards,
>>> Anthony
>>>
>>> On 1/16/2013 18:24, Artem Ananiev wrote:
>>>>
>>>> On 1/15/2013 8:27 PM, Anthony Petrov wrote:
>>>>> Hi Artem,
>>>>>
>>>>> I've split the setUpAWTAppKit() to setUpAWTAppKit() and
>>>>> setUpAppKitThreadName() (and also a separate call to
>>>>> NSSetUncaughtExceptionHandler() in the very beginning of 
>>>>> -starter:.) If
>>>>> you examine the logic in starter: carefully, you may notice that 
>>>>> all the
>>>>> required operations are performed just as before based on the current
>>>>> mode of execution.
>>>>
>>>> Before the fix, setUpAWTAppKit() is called from -starter: regardless
>>>> of "headless" value.
>>>>
>>>> After the fix, setUpAppKitThreadName() is called in headless mode, and
>>>> both setUpAWTAppKit() and setUpAppKitThreadName() in headful mode.
>>>>
>>>> The difference is than after the fix we don't add loop observers in
>>>> headless mode. Is it intentional?
>>>>
>>>> Thanks,
>>>>
>>>> Artem
>>>>
>>>>> In particular, in the headless mode we want to only set the AppKit
>>>>> thread name and only if we're not running in the "swt compatible 
>>>>> mode".
>>>>> We don't use observers in the headless mode. Also, the uncaught
>>>>> exception handler is always set unconditionally.
>>>>>
>>>>> -- 
>>>>> best regards,
>>>>> Anthony
>>>>>
>>>>> On 1/15/2013 20:11, Artem Ananiev wrote:
>>>>>> Hi, Anthony,
>>>>>>
>>>>>> after the fix, setUpAWTAppKit() is not called in headless mode, 
>>>>>> while
>>>>>> it was before. Could it be a problem?
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>> Artem
>>>>>>
>>>>>> On 1/15/2013 7:54 PM, Anthony Petrov wrote:
>>>>>>> Hello,
>>>>>>>
>>>>>>> After discussing this fix off-line, we decided to start off with a
>>>>>>> minimal version of the fix that enables FX and AWT to run in 
>>>>>>> headful
>>>>>>> mode, and be able to be initialized in any order. The new fix for
>>>>>>> http://bugs.sun.com/view_bug.do?bug_id=8005465 is at:
>>>>>>>
>>>>>>> http://cr.openjdk.java.net/~anthony/8-53-startOnFirstThreadCheck-minimal-8005465.0/ 
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> There's some cleanup of the initialization procedure of AWT that 
>>>>>>> makes
>>>>>>> it easier to follow. The most essential change is that AWT no 
>>>>>>> longer
>>>>>>> throws an exception if it's started on the main thread and the
>>>>>>> -XstartOnMainThread hasn't been specified when running the app.
>>>>>>> This is
>>>>>>> to allow one to initialize the AWT from the AppKit thread (which is
>>>>>>> the
>>>>>>> FX User Thread on the Mac).
>>>>>>>
>>>>>>> There are still a few issues remain (specifically, 8006320 and
>>>>>>> 8006326),
>>>>>>> but they may require more time to investigate, and we need to
>>>>>>> back-port
>>>>>>> this fix to a 7 update release as soon as possible.
>>>>>>>
>>>>>>> The fix has been tested with FX for the desired use-case, as 
>>>>>>> well as
>>>>>>> with SWT (thanks Petr!). No problems found.
>>>>>>>
>>>>>>> -- 
>>>>>>> best regards,
>>>>>>> Anthony
>>>>>>>


-- 
Best regards, Sergey.