Hi Thomas,

> For now this is not enabled by default. I propose to thouroughly test
> > this and make it default when things turn out to work.
> 
> This may make sense for portableNativeSync -- have you talked to the JikesRVM 
> developers about this change?  I don't think making it the default is a good 
> idea though, for two reasons:
> 
> 1) It eliminates the possibility of using the GTK AWT peers in the same 
> process
>     as another GTK-using toolkit.  We've already hit this problem because SWT
>     does something similar to what you're proposing:
> 
>     http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25375
>     https://bugs.eclipse.org/bugs/show_bug.cgi?id=139468
> 
> 2) Most GDK-using applications use the default non-reentrant lock, which means
>     that reentrant code paths through GDK won't have been as well-tested.  If 
> all
>     AWT apps start using these code paths then this opens up the possibility 
> of
>     more subtle race conditions being exposed (we've had to track down many 
> over
>     the years, in our GTK peers and in components of GNOME).
> 
> If we really want a reentrant GDK lock then I think we should propose the 
> change 
> to GDK itself.  SWT would likely adopt the result (resolving those bug 
> reports) 
> and the GDK maintainers could help audit the current GDK/GTK code to 
> eliminate 
> reentrancy problems.  Once that work has landed in GDK we'll be able to 
> simplify 
> our peer code.

I agree with that. I've filed a bugreport for GDK. Let's see how it
works out.

http://bugzilla.gnome.org/show_bug.cgi?id=425995

/Roman
-- 
http://kennke.org/blog/


Reply via email to