On Mon, 27 Jan 2020 at 16:40, Giovanni Mascellani gio@debian.org wrote:
@@ -2296,11 +2290,21 @@ static BOOL d2d_geometry_add_figure_outline(struct d2d_geometry *geometry, else next = &figure->vertices[i + 1];
if ((figure_end == D2D1_FIGURE_END_CLOSED || (i && i < figure->vertex_count - 1))&& !d2d_geometry_outline_add_join(geometry, prev, p0, next))
if (figure_end == D2D1_FIGURE_END_CLOSED || (i && i < figure->vertex_count - 1)) {
ERR("Failed to add join.\n");return FALSE;
D2D1_POINT_2F q_next, q_prev;d2d_point_subtract(&q_prev, prev, p0);d2d_point_subtract(&q_next, next, p0);d2d_point_normalise(&q_prev);d2d_point_normalise(&q_next);Why are these normalization necessary? Vectors prev and next will be normalized again in the shader (and they have to, because the transformation matrix is not necessarily orthogonal), so normalization could be skipped here.
They may not be necessary. They're mostly there because d2d_geometry_outline_add_join() originally had them. I think that in turn was mostly a leftover from when d2d_geometry_outline_add_join() (incorrectly) used d2d_point_dot() instead of d2d_point_ccw(). Note that the "prev" and "next" vectors are also used for vertex positions in the "ccw == 0.0f" case though. In any case, it would probably be best to do a potential removal in a separate follow-up patch.