JTablet Doesn't Close Properly

[Jason]

JTablet does not seem to close correctly. If the **last** JTablet-
enabled applet is closed, and then some time later a new JTablet-
enabled applet started **in the same JVM**, drawing does not work in
the newly-started applet. While this bug sounds hard to trigger, its
really quite easy. Since most browsers keep Java running for at least
a few seconds the last applet has closed, the reproduction steps are
simply:

1. Go to http://jtablet.cellosoft.com/jtablet2-demo.html
2. Draw something
3. Refresh the page
4. Try to draw something again

If you wait for Java to exit before reloading the applet (for example,
replace step 3 with "leave the page, wait a minute or so, go back to
the page") then there should be no problem. The easiest way to see
when Java has exited is to do the steps listed under method 1 at
http://www.java.com/en/download/help/javaconsole.xml -- as soon as
Java starts a console will open, and the console will naturally
disappear as soon as Java exits.

I'm not 100% sure what is causing this bug, though I have a theory
after some debugging. I think that the problem occurs because of an
interaction of two things. First is that the static instance of
TabletManagerImpl held by TabletManager is **not** garbage collected
(it is a static variable, afterall). Second is that when the applet is
closed, NativeTabletManager threads are stopped by **Java** not the
**applet**. I believe the two interact in the following way:

1. JVM starts up
2. DemoApplet calls TabletManager.getDefaultManager()
3. A static instance of TabletManagerImpl is created (which creates a
NativeTabletManager to do the dirty work)
4. DemoApplet calls addTabletListener() and addScreenTabletListener()
5. Each of these methods in turn call startIfNeeded() in
ScreenTabletManager to start the NativeTabletManager threads
6. NativeTabletManager threads begin relaying events

7. Time passes... User draws something... User exits applet

8. Applet gets a "close" singal
9. Java stops running threads
10. Objects are garbage collected

11. Time passes... User opens a new applet

12. Applet calls TabletManager.getDefaultManager()
13. The static instance of TabletManagerImpl is created (referencing
the same NativeTabletManager it always has)
14. DemoApplet calls addTabletListener() and addScreenTabletListener()
15. Each of these methods in turn call startIfNeeded(), but since
'started' is still 'true' (stopIfNeeded() was never called!) no thread
is started
16. No events are ever relayed.

Provided I have this timeline correct (and that it is indeed the
source of this bug) we either need to ensure that 'started' is reset
to 'false' when the last applet using JTablet is closed, or to ensure
our NativeTabletManager is garbage collected. Implementing either seem
problematic: the former requires we know when the last JTablet applet
has ended, while the latter seems to be incompatible with the
singleton approach used in TabletManagerImpl.

[Marcello Bastéa-Forte]

To clarify, which OS(s) are you seeing this on?

Marcello

--
You received this message because you are subscribed to the Google Groups "jtablet-dev" group.
To post to this group, send email to jtabl...@googlegroups.com.
To unsubscribe from this group, send email to jtablet-dev...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/jtablet-dev?hl=en.

[Jason]

Right now I'm only seeing it on Linux under Chromium and Epiphany
since that's all I have access to. However, it seems like this should
affect all platforms (provided I debugged correctly). I'll see if I
can find a Windows box to plug my tablet into for testing.

Jason

On Jun 25, 4:32 pm, Marcello Bastéa-Forte <marce...@cellosoft.com>
wrote:

> To clarify, which OS(s) are you seeing this on?
>
> Marcello
>
>
>
> On Fri, Jun 25, 2010 at 4:02 PM, Jason <killert...@gmail.com> wrote:
> > JTablet does not seem to close correctly. If the **last** JTablet-
> > enabled applet is closed, and then some time later a new JTablet-
> > enabled applet started **in the same JVM**, drawing does not work in
> > the newly-started applet. While this bug sounds hard to trigger, its
> > really quite easy. Since most browsers keep Java running for at least
> > a few seconds the last applet has closed, the reproduction steps are
> > simply:
>

> > 1. Go tohttp://jtablet.cellosoft.com/jtablet2-demo.html

> > 2. Draw something
> > 3. Refresh the page
> > 4. Try to draw something again
>
> > If you wait for Java to exit before reloading the applet (for example,
> > replace step 3 with "leave the page, wait a minute or so, go back to
> > the page") then there should be no problem. The easiest way to see
> > when Java has exited is to do the steps listed under method 1 at

> >http://www.java.com/en/download/help/javaconsole.xml-- as soon as

> > jtablet-dev...@googlegroups.com<jtablet-dev%2Bunsubscribe@googlegr oups.com>
> > .

[Jason]

Well now, this is weird. Although I can't replicate this bug on
Windows, I can't find any reason why not.

Even though the "JTablet-WinTab" thread doesn't restart (as
predicted), the DemoApplet still somehow receives input from WinTab.
*scratches head*

Time for more spelunking through the code I guess...

Jason

> > >http://www.java.com/en/download/help/javaconsole.xml--as soon as

[Marcello Bastéa-Forte]

I haven't had a chance to try out your specific problem on windows/linux. In theory, as soon as the components that addTabletManager were called on are garbage collected, an implicit call to removeTabletManager occurs. 

This is done by hiding the pointer to TabletListener inside the Component itself via a HierarchyListener, and using weak references everywhere else. Unfortunately there's no solution for screen listeners, since there's no way of knowing when the application/applet that added the screenlistener is garbage collected.

Your point about the wintab thread stopping is interesting, though, and I should double check that. What happens if you override the addTabletListener and addScreenTabletListener methods (or make startIfNeeded protected and override it) to double check the thread is still alive? Would that help?

Marcello

To unsubscribe from this group, send email to jtablet-dev...@googlegroups.com.