[PATCH 01/12] d2d1: Store cubic bezier control points.

[Connor McAdams]

Ah, okay. That makes sense. In my original patch, I had just stored
everything as a cubic in the same way everything was previously stored
as a quadratic.

If both are to co-exist, should the functions for detecting
intersections between beziers all be cubic, or should they be
temporarily promoted to a cubic in that instance? Same goes with
bezier-line intersections. One of the nice things about storing them
all with both cubic and quadratic coordinates was that it meant that
things that used one or the other could do so interchangeably, but I
guess it can be done that way too with them as separate types with a
little more work.

Also, when they're eventually sent to the pixel shader, would you like
there to be separate paths for cubics and quadratics? Or can I promote
a quadratic to a cubic in that case? It kind of gets a little messy
supporting both, but if that's what's best, I can do that.

On Tue, Feb 25, 2020 at 9:45 AM Henri Verbeet <hverbeet at gmail.com> wrote:
>
> Hi Connor,
>
> It will probably take me a while to go through the entire series, but
> here are some comments/questions on the first two patches.
>
> On Tue, 25 Feb 2020 at 06:03, Connor McAdams <conmanx360 at gmail.com> wrote:
> > +struct d2d_bezier_controls
> > +{
> > +    D2D1_POINT_2F c0, c1;
> > +    D2D1_POINT_2F cq0;
> > +};
> > +
> Why do you need this structure? It seems more convenient to rename
> D2D_VERTEX_TYPE_BEZIER to D2D_VERTEX_TYPE_QUADRATIC_BEZIER, and
> introduce D2D_VERTEX_TYPE_CUBIC_BEZIER, while keeping the control
> point arrays the way they are. Not in the least because
> d2d_path_geometry_Stream() will need to know the original segment
> types.
>
> > @@ -1907,8 +1928,13 @@ static BOOL d2d_geometry_apply_intersections(struct d2d_geometry *geometry,
> >          d2d_point_lerp(&q[0], p[0], p[1], t);
> >          d2d_point_lerp(&q[1], p[1], p[2], t);
> >
> > -        figure->bezier_controls[inter->control_idx + control_offset] = q[0];
> > -        if (!(d2d_figure_insert_bezier_control(figure, inter->control_idx + control_offset + 1, &q[1])))
> > +        d2d_bezier_quad_to_cubic(p[0], &q[0], &inter->p, &c[0], &c[1]);
> > +        figure->bezier_controls[inter->control_idx + control_offset].cq0 = q[0];
> > +        figure->bezier_controls[inter->control_idx + control_offset].c0 = c[0];
> > +        figure->bezier_controls[inter->control_idx + control_offset].c1 = c[1];
> > +
> > +        d2d_bezier_quad_to_cubic(&inter->p, &q[1], p[2], &c[0], &c[1]);
> > +        if (!(d2d_figure_insert_bezier_control(figure, inter->control_idx + control_offset + 1, &c[0], &c[1], &q[1])))
> >              return FALSE;
> >          ++control_offset;
> >
> ...
> > @@ -2769,8 +2795,15 @@ static BOOL d2d_geometry_split_bezier(struct d2d_geometry *geometry, const struc
> >      d2d_point_lerp(&q[1], p[1], p[2], 0.5f);
> >      d2d_point_lerp(&q[2], &q[0], &q[1], 0.5f);
> >
> > -    figure->bezier_controls[idx->control_idx] = q[0];
> > -    if (!(d2d_figure_insert_bezier_control(figure, idx->control_idx + 1, &q[1])))
> > +    d2d_bezier_quad_to_cubic(p[0], &q[0], &q[2], &c[0], &c[1]);
> > +
> > +    figure->bezier_controls[idx->control_idx].cq0 = q[0];
> > +    figure->bezier_controls[idx->control_idx].c0 = c[0];
> > +    figure->bezier_controls[idx->control_idx].c1 = c[1];
> > +
> > +    d2d_bezier_quad_to_cubic(&q[0], &q[1], p[2], &c[0], &c[1]);
> > +
> > +    if (!(d2d_figure_insert_bezier_control(figure, idx->control_idx + 1, &c[0], &c[1], &q[1])))
> >          return FALSE;
> >      if (!(d2d_figure_insert_vertex(figure, idx->vertex_idx + 1, q[2])))
> >          return FALSE;
> Likewise, you wouldn't need these awkward bits of code.