[cairo] patch [1/1] invert image in _cairo_gl_surrface_map_to_image

[Andrea Canciani]

On Wed, Jan 4, 2012 at 1:56 AM, Henry (Yu) Song - SISA
<hsong at sisa.samsung.com> wrote:
> Date:   Fri Dec 30 11:24:54 2011 -0800
>
>    gl: Add support to invert non texture gl surface image that does not have
>    GL_MESA_pack_invert when map from gl surface to image surface
>
> diff --git a/src/cairo-gl-surface.c b/src/cairo-gl-surface.c
> index 8f69033..8b8f196 100644
>
> --- a/src/cairo-gl-surface.c
> +++ b/src/cairo-gl-surface.c
> @@ -1001,6 +1001,10 @@ _cairo_gl_surface_map_to_image (void      *abstract_surface,
>     unsigned int cpp;
>     cairo_bool_t invert;
>     cairo_status_t status;
> +    cairo_t *inv_cr = NULL;
> +    cairo_image_surface_t *inv_image = NULL;
> +    cairo_pattern_t *inv_pattern = NULL;
> +    cairo_matrix_t inv_matrix;
>
>     /* Want to use a switch statement here but the compiler gets whiny. */
>     if (surface->base.content == CAIRO_CONTENT_COLOR_ALPHA) {
> @@ -1093,6 +1097,50 @@ _cairo_gl_surface_map_to_image (void      *abstract_surface,
>        image = (cairo_image_surface_t *)
>            _cairo_surface_create_in_error (status);
>     }
> +
> +    /*
> +     * FIXME: we must invert the image if it is non texture surface,
> +     * and it does not have GL_MESA_pack_invert.
> +     * Is there more efficient way to invert image?

Using negative stride might be more efficient, but some backends might
not (yet) support them correctly.

Side note:
It is already possible to create negative stride images.
Should we add a test trying to use them and support them?
Unless I get feedback, I'll try to write the test this weekend.

> +     */
> +
> +    if (! _cairo_gl_surface_is_texture (surface) && ! invert) {
> +        inv_image = (cairo_image_surface_t*)
> +       _cairo_image_surface_create_with_pixman_format (NULL,
> +                                                       pixman_format,
> +                                                       extents->width,
> +                                                       extents->height,
> +                                                       -1);
> +       if (unlikely (inv_image->base.status)) {
> +           goto CLEAR_IMAGE;
> +       }
> +
> +       image->base.is_clear = FALSE;
> +       inv_pattern = cairo_pattern_create_for_surface (&image->base);
> +       if (unlikely (inv_pattern->status)) {
> +           goto CLEAR_PATTERN;
> +       }
> +       cairo_matrix_init_scale (&inv_matrix, 1.0, -1.0);
> +       cairo_matrix_translate (&inv_matrix, 0, -(extents->height));
> +       cairo_pattern_set_matrix (inv_pattern, &inv_matrix);
> +
> +       inv_cr = cairo_create (&inv_image->base);
> +       if (unlikely (inv_cr->status)) {
> +           goto CLEAR_CAIRO;
> +       }
> +       cairo_set_source (inv_cr, inv_pattern);
> +       cairo_set_operator (inv_cr, CAIRO_OPERATOR_SOURCE);
> +       cairo_paint (inv_cr);

You're using the same surface as source and as destination.
This results in undefined behavior.
If you want to flip the image like this, you should use two independent surfaces


> +       cairo_surface_destroy (&image->base);
> +       image = (cairo_image_surface_t *)cairo_surface_reference (&inv_image->base);
> +    }
> +
> +CLEAR_CAIRO:
> +    cairo_destroy (inv_cr);
> +CLEAR_PATTERN:
> +    cairo_pattern_destroy (inv_pattern);
> +CLEAR_IMAGE:
> +    cairo_surface_destroy (&inv_image->base);
>
>     return &image->base;
>  }
> --
> cairo mailing list
> cairo at cairographics.org
> http://lists.cairographics.org/mailman/listinfo/cairo