Re: [PATCH] robustness for GdkEvent.xs

[Torsten Schoenfeld <kaffeetisch gmx de>]

On Sun, 2005-11-06 at 22:14 -0500, muppet wrote:

> 1. emit event-after
> 2. handler invoked
> 3. GClosure.xs tries to marshal the parameters
> 4. GdkEvent.xs tries to figure out into which package to bless the event
> 5. gdk_event_get_package() doesn't know about GDK_GRAB_BROKEN, and  
> croaks with "Illegal type %d in event->type".
> 6. we later crash in gsignal.c because it's really not good to  
> longjmp out of signals.
>
> Now, #6 isn't really what i'm worried about --- we know that  
> exceptions in signals are bad, that's why we eval handlers.  #5 is  
> the problem.  We need to be robust against this sort of thing,  
> because it's always possible for new versions of gtk+ to add event  
> types, and we shouldn't crash because of it -- we should at least  
> limp along.

Ouch, I should have been more careful when I added support for
grab-broken.  Sorry about that.  I did test compilations/runs on all
stable gtk+'s, but our test suite just doesn't cover this kind of thing.

> The attached patch turns this croak() into a warn, and falls back on  
> blessing the event into the generic Gtk2::Gdk::Event package.  It  
> also plugs a memory leak by ensuring that we have  
> Gtk2::Gdk::Event::GrabBroken::DESTROY.

The patch looks good to me.

> I think this needs to go into HEAD and 1.10x.  Should we release a  
> new 1.08x with this for distros that still have 1.08x as the current  
> stable?

If there's a reasonable possibility that distros which ship 1.08x
actually incorporate a bug fix release, then yes, I think this is worth
a 1.084.

-- 
Bye,
-Torsten