Pixbuf to Cairo and vice versa

[Mario Kemper <mario kemper googlemail com>]

Hi,

I was benchmarking one of my applications today and one of the slowest operations I detected is related to 
the Cairo > Pixbuf conversion.

This is what the code is currently doing:
1) Get the current screen content as a pixbuf

my $clean_pixbuf = Gtk2::Gdk::Pixbuf->get_from_drawable(
        $self->{_root}, undef, 0, 0, 0, 0,
        $self->{_root}->{w},
        $self->{_root}->{h}
);

2) I want to draw something on the pixbuf with Cairo

my $surface = Cairo::ImageSurface->create( 'argb32', $self->{_root}->{w}, $self->{_root}->{h} );
my $cr = Cairo::Context->create($surface);

Gtk2::Gdk::Cairo::Context::set_source_pixbuf( $cr, $clean_pixbuf, 0, 0 );
$cr->paint;

<<< more painting goes here >>>

3) Show the final image in a widget, i.e. convert the cairo surface back to a pixbuf

#write surface to pixbuf
my $loader = Gtk2::Gdk::PixbufLoader->new;
$surface->write_to_png_stream(
        sub {
                my ( $closure, $data ) = @_;
                $loader->write($data);
        }
);
$loader->close;

<<< use $loader->get_pixbuf somewhere >>>



Especially the last operation is pretty slow (the root window is big, dual monitor). My question is: Is this 
the correct workflow to do such things? Is there any way to avoid the conversions between pixbufs and cairo?


Thanks in advance.

Bye
Mario