I meant you can do dlopen("libatk-bridage,so", RTLD_NOW); and
intentionally leak the handle.
Ginn
On Aug 21, 2009, at 11:33 AM, Ke Wang wrote:
> It seems that I can not even detect when JNI environment is detached.
> Maybe invoking dl_shutdown() when the last child of top level app is
> removed will be a solution.
> But still I'm not sure this will always happen before exit func,
> because they execute in different thread.
> So, we should wait for the removeal of g_atexit() invocation in ATK
> bridge.
>
> Thanks,
>
> Ke
>
> Ginn Chen wrote:
>> So JAW would not be detached when the exit function is called?
>> Can you hold a ref to the ATK bridge library to make sure it won't
>> be detached earlier than JAW?
>>
>> Ginn
>>
>> On Aug 20, 2009, at 10:40 AM, Ke Wang wrote:
>>
>>> The exit function will be called in a context that we can not
>>> control in Java.
>>> At the time the exit function registered by ATK bridge is called,
>>> the JNI environment may have been detached from JVM.
>>> So any callback from registryd would either crash or hang the Java
>>> app at that time.
>>>
>>> This is just a workaround.
>>> After the modification of ATK bridge's exit handler being done by
>>> Li, I'll do corresponding change in JAW.
>>>
>>> Thanks,
>>>
>>> Ke
>>>
>>> On 08/19/09 18:02, Ginn Chen wrote:
>>>> It looks like you're covering the ATK bridge bug with the same
>>>> usage of g_atexit().
>>>> What about just not do dl_shutdown()?
>>>> Would it crash?
>>>>
>>>> Ginn
>>>>
>>>> On Aug 19, 2009, at 5:40 PM, Ke Wang wrote:
>>>>
>>>>> Hi all,
>>>>>
>>>>> This patch forces Java app to quit before the invocation of exit
>>>>> func registered by ATK bridge.
>>>>>
>>>>> Regards,
>>>>>
>>>>> Ke
>>>>> <AtkWrapper.diff>
>>>>
>>>
>>
>
