more comments on cairo's and our own woes

(bzr r2455)

2 files changed, 17 insertions, 2 deletions

diff --git a/src/display/canvas-arena.cpp b/src/display/canvas-arena.cpp
index 3212d0bd2..a7615b3d7 100644
--- a/src/display/canvas-arena.cpp
+++ b/src/display/canvas-arena.cpp
@@ -200,6 +200,16 @@ sp_canvas_arena_render (SPCanvasItem *item, SPCanvasBuf *buf)
 	bh = buf->rect.y1 - buf->rect.y0;
 	if ((bw < 1) || (bh < 1)) return;
 
+	// FIXME: currently this function is a huge waste. It receives a buffer but creates a new one and loops 
+	// within the large one, doing arena painting in several blocks. This just makes no sense because the 
+	// buf that we are given is already only a strip of the screen, created by one iteration of a loop in
+	// sp_canvas_paint_rect_internal. With the current numbers, this function's buffer is always 1/4 
+	// smaller than the one we get, because they both are the same number of bytes but
+	// buf uses 3 bytes per pixel (24bpp, packed) while the pixblock created here uses 4 bytes (32bpp).
+	// Eventually I want to switch buf to using 4 bytes (see comment in canvas.cpp) and then remove 
+	// from here the sw/sh calculation, the loop, and creating the intermediate buffer, allowing arena 
+	// just render into buf in one go.  
+
 	if (arena->arena->rendermode != RENDERMODE_OUTLINE) { // use 256K as a compromise to not slow down gradients
 		/* 256K is the cached buffer and we need 4 channels */
 		if (bw * bh < 65536) { // 256K/4
diff --git a/src/display/sp-canvas.cpp b/src/display/sp-canvas.cpp
index f716d383a..75ffeed84 100644
--- a/src/display/sp-canvas.cpp
+++ b/src/display/sp-canvas.cpp
@@ -1560,7 +1560,13 @@ sp_canvas_paint_single_buffer (SPCanvas *canvas, int x0, int y0, int x1, int y1,
                             x1 - x0, y1 - y0);
     } else {
 /*
- CAIRO FIXME: after SPCanvasBuf is made 32bpp throughout, this rgb_draw below can be replaced with:
+// CAIRO FIXME: after SPCanvasBuf is made 32bpp throughout, this rgb_draw below can be replaced with the below.
+// Why this must not be done currently:
+// - all canvas items (handles, nodes etc) paint themselves assuming 24bpp
+// - cairo assumes bgra, but we have rgba, so r and b get swapped (until we paint all with cairo too)  
+// - it does not seem to be any faster; in fact since with 32bpp, buf contains less pixels, 
+// we need more bufs to paint a given area and as a result it's even a bit slower
+
     cairo_surface_t* cst = cairo_image_surface_create_for_data (
         buf.buf,
         CAIRO_FORMAT_RGB24,  // unpacked, i.e. 32 bits! one byte is unused
@@ -1570,7 +1576,6 @@ sp_canvas_paint_single_buffer (SPCanvas *canvas, int x0, int y0, int x1, int y1,
         cairo_t *ct = gdk_cairo_create(SP_CANVAS_WINDOW (canvas));
         cairo_set_source_surface (ct, cst, x0 - canvas->x0, y0 - canvas->y0);
         cairo_paint (ct);
-
     cairo_destroy (ct);
     cairo_surface_finish (cst);
     cairo_surface_destroy (cst);