[cairo] some fixes and improvements to current cvs version of cairo

[Kristian Høgsberg]

David Reveman wrote:
>>>@@ -2006,8 +1997,12 @@
>>> 					  &device_width, &device_height);
>>> 
>>>     _cairo_pattern_init_for_surface (&pattern, surface);
>>>-    _cairo_gstate_create_pattern (gstate, &pattern);
>>> 
>>>+    if (!_cairo_matrix_is_integer_translation (&device_to_image, &xoff, &yoff))
>>>+	pattern.matrix = device_to_image;
>>>+    
>>>+    _cairo_pattern_set_alpha (&pattern, gstate->alpha);
>>>+	
>>
>>Isn't this an optimization that belongs in the backend?
> 
> I don't think any backend would mind getting an integer translation as a
> source offset rather than a transformation matrix. But if you like to
> move this into the backends instead that's fine with me. You should know
> that this optimization makes a pretty big difference on xrender and
> image backends, so it's probably worth keeping, in some form.

Right, I just assumed that xrender and pixman had a fast path for this. 
  If it is a significant improvement, maybe it should be moved into 
_cairo_pattern_get_surface so it kicks in for all source surfaces?

>>>@@ -810,16 +811,17 @@
>>> 	    }
>>> 
>>> 	    image = _cairo_surface_get_image (src);
>>>-	    
>>>-	    surface = cairo_surface_create_similar (dst,
>>>-						    CAIRO_FORMAT_ARGB32,
>>>-						    width, height);
>>>+
>>>+	    surface = cairo_surface_create_similar (dst, CAIRO_FORMAT_ARGB32,
>>>+						    image->width,
>>>+						    image->height);
>>
>>What the difference here?
> 
> It makes a difference if x or y are negative, and when using
> cairo_surface_set_image, surface dimensions must match image dimensions.
> With width and height, this might not always be true.

Ah, yes, and in that case width and height would be larger than 
image->width and image->height.

cheers,
Kristian