<AWT Dev> [14] Review Request: 8234522 [macos] Crash with use of native file dialog

[Dmitry Markov]

Hi Sergey,

I agree, it would be better to replace NSApplicationDefined with NSEventTypeApplicationDefined under separate task.

The fix still looks good to me.

Thanks,
Dmitry

PS
Do not forget to add a summary to the regression test

> On 10 Dec 2019, at 18:20, Sergey Bylokhov <Sergey.Bylokhov at oracle.com> wrote:
> 
> Hi, Dmitry.
> 
> We probably can replace NSApplicationDefined by NSEventTypeApplicationDefined, but then we will need
> to replace NSApplicationDefinedMask by NSEventMaskApplicationDefined, and probably something else.
> So I leave it for some future deprecation-cleanup.
> 
> On 12/10/19 4:03 am, Dmitry Markov wrote:
>> Hi Sergey,
>> The changes look OK, but I have a few remarks. I would recommend using NSEventTypeApplicationDefined due to deprecation of NSApplicationDefined. Also it would be good to add some kind of summary to the regression test. I do not need a new webrev with suggested changes.
>> Thanks,
>> Dmitry
>>> On 6 Dec 2019, at 23:05, Sergey Bylokhov <sergey.bylokhov at oracle.com> wrote:
>>> 
>>> Hello.
>>> Please review the fix for JDK 14.
>>> 
>>> Bug: https://bugs.openjdk.java.net/browse/JDK-8234522
>>> Fix: http://cr.openjdk.java.net/~serb/8234522/webrev.00
>>> 
>>> This bug is tricky, we have a special mechanism to postpone deallocation
>>> of the objects on the Appkit thread.
>>> It solves the next problem: the native window pointer is disposed in the
>>> inner AppKit loop, while it is still referenced on the stack in the native
>>> Cocoa method which caused the mentioned inner loop. When the inner loop is
>>> exited Cocoa crashes while dereferencing this window pointer.
>>> 
>>> This mechanism is implemented via posting NSEvent using NSApplicationDefined
>>> type and filter out this event in the nested event loop.
>>> 
>>> But for some reason, macOS send us exactly the same event(type+subtype), which
>>> we try to use for deallocation and of course, this attempt will crashes.
>>> (probably because all constants of NSEvent we use were deprecated?)
>>> 
>>> In the fix I have added additional marks(via data fields) to the event to make
>>> it easy to check it was created by us, also I have bump the initial value of
>>> subtype to minimize possible collisions.
>>> 
>>> 
>>> -- 
>>> Best regards, Sergey.
> 
> 
> -- 
> Best regards, Sergey.