[icedtea-web] RFC: Patch to fix applet exit

[Deepak Bhole]

* Denis Lila <dlila at redhat.com> [2011-05-02 14:24]:
> > Yep, I thought about that case but the probability of that is very
> > low.
> > Between having the VM potentially shut down before an applet starts
> > (fixable by refreshing) and having a running applet freeze (which
> > could
> > include unsaved work), the former sounds like the lesser of 2
> > evils and more acceptable in the short term.
> > 
> > Omair's original comment about having a separate ThreadGroup is the
> > correct approach, but is not easy to implement atm, especially within
> > the 1.1 time frame.
> > 
> > One thing we can do is add this to 1.1 only, but I really don't think
> > the case of "VM shut down when another applet is starting" is worth
> > worrying about for now, given how unlikely it is.
> 
> I see.
> I assume this will be easier to fix after the security redesign, right?
> In any case, what you say makes sense, and it's true that all you have
> to do to make the applet work is to refresh the page, I say this is good to
> go for head and 1.1 (but maybe put a TODO in there? We should fix this
> eventually after all.).
> 

Good point, I will add a TODO before committing. Thanks!

Cheers,
Deepak

> Regards,
> Denis.
> 
> ----- Original Message -----
> > * Denis Lila <dlila at redhat.com> [2011-05-02 13:47]:
> > > > This patch addresses an issue cause by the tg.stop() call in
> > > > PluginAppletViewer.appletClose().
> > > >
> > > > With the recent classloader sharing changes, applets from the same
> > > > page
> > > > have the same threadgroup. As a result if 2 copies are opened and
> > > > one
> > > > is
> > > > closed, the other one stops too.
> > > >
> > > > With this patch, tg.stop() is no longer called and instead, the
> > > > plugin
> > > > will halt the VM after the last applet to guarantee that resources
> > > > are
> > > > freed eventually (the tg.stop was added because not all applets
> > > > exit
> > > > nicely on stop/dispose).
> > >
> > > Is it possible for an applet creation to be racing with the
> > > applet close? I'm thinking, what happens if we're handling an
> > > applet creation but we haven't called the PAV constructor yet?
> > > (so, suppose we're still in parse() and haven't called
> > > factory.createPanel yet).
> > >
> > > In that case, wouldn't we just shut down the vm, because
> > > we haven't added the PAV to appletPanels yet and so
> > > 0==countApplets()? Isn't this wrong?
> > >
> > 
> 
> > 
> > Cheers,
> > Deepak
> > 
> > > Regards,
> > > Denis.
> > >
> > > ----- Original Message -----
> > > > Hi,
> > > >
> > >
> > > > ChangeLog:
> > > > 2011-05-02 Deepak Bhole <dbhole at redhat.com>
> > > >
> > > > * plugin/icedteanp/java/sun/applet/PluginAppletViewer.java
> > > > (appletClose): Do not try to stop threads, now that the loader is
> > > > shared
> > > > and the thread group for applets on a page is identical. Call
> > > > dispose
> > > > from
> > > > invokeAndWait.
> > > > (appletSystemExit): Exit the VM when called.
> > > >
> > > > Okay for HEAD and 1.1?
> > > >
> > > > Cheers,
> > > > Deepak