Auto merge of #7679 - bjwbell:reftest-twitter-fail-whale, r=pcwalton

Add css twitter fail whale html test & update ellipse_to_bezier comment

cgaebel had a TODO for a css twitter fail whale reftest, which depended on elliptical border-radius support.
I didn't see any feasible way other than a reference image for border_twitter_fail_whale_b.html.

r? @pcwalton

<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/servo/servo/7679)
<!-- Reviewable:end -->
This commit is contained in:
bors-servo 2015-09-18 22:54:44 -06:00
commit c91c0188f2
2 changed files with 1430 additions and 81 deletions

View file

@ -484,7 +484,81 @@ impl<'a> PaintContext<'a> {
radius.width <= 0. || radius.height <= 0.
}
// Adapted from gecko:gfx/2d/PathHelpers.h:EllipseToBezier
// The following comment is wonderful, and stolen from
// gecko:gfx/thebes/gfxContext.cpp:RoundedRectangle for reference.
// ---------------------------------------------------------------
//
// For CW drawing, this looks like:
//
// ...******0** 1 C
// ****
// *** 2
// **
// *
// *
// 3
// *
// *
//
// Where 0, 1, 2, 3 are the control points of the Bezier curve for
// the corner, and C is the actual corner point.
//
// At the start of the loop, the current point is assumed to be
// the point adjacent to the top left corner on the top
// horizontal. Note that corner indices start at the top left and
// continue clockwise, whereas in our loop i = 0 refers to the top
// right corner.
//
// When going CCW, the control points are swapped, and the first
// corner that's drawn is the top left (along with the top segment).
//
// There is considerable latitude in how one chooses the four
// control points for a Bezier curve approximation to an ellipse.
// For the overall path to be continuous and show no corner at the
// endpoints of the arc, points 0 and 3 must be at the ends of the
// straight segments of the rectangle; points 0, 1, and C must be
// collinear; and points 3, 2, and C must also be collinear. This
// leaves only two free parameters: the ratio of the line segments
// 01 and 0C, and the ratio of the line segments 32 and 3C. See
// the following papers for extensive discussion of how to choose
// these ratios:
//
// Dokken, Tor, et al. "Good approximation of circles by
// curvature-continuous Bezier curves." Computer-Aided
// Geometric Design 7(1990) 33--41.
// Goldapp, Michael. "Approximation of circular arcs by cubic
// polynomials." Computer-Aided Geometric Design 8(1991) 227--238.
// Maisonobe, Luc. "Drawing an elliptical arc using polylines,
// quadratic, or cubic Bezier curves."
// http://www.spaceroots.org/documents/ellipse/elliptical-arc.pdf
//
// We follow the approach in section 2 of Goldapp (least-error,
// Hermite-type approximation) and make both ratios equal to
//
// 2 2 + n - sqrt(2n + 28)
// alpha = - * ---------------------
// 3 n - 4
//
// where n = 3( cbrt(sqrt(2)+1) - cbrt(sqrt(2)-1) ).
//
// This is the result of Goldapp's equation (10b) when the angle
// swept out by the arc is pi/2, and the parameter "a-bar" is the
// expression given immediately below equation (21).
//
// Using this value, the maximum radial error for a circle, as a
// fraction of the radius, is on the order of 0.2 x 10^-3.
// Neither Dokken nor Goldapp discusses error for a general
// ellipse; Maisonobe does, but his choice of control points
// follows different constraints, and Goldapp's expression for
// 'alpha' gives much smaller radial error, even for very flat
// ellipses, than Maisonobe's equivalent.
//
// For the various corners and for each axis, the sign of this
// constant changes, or it might be 0 -- it's multiplied by the
// appropriate multiplier from the list before using.
// ---------------------------------------------------------------
//
// Code adapted from gecko:gfx/2d/PathHelpers.h:EllipseToBezier
fn ellipse_to_bezier(path_builder: &mut PathBuilder,
origin: Point2D<AzFloat>,
radius: Size2D<AzFloat>,
@ -629,86 +703,6 @@ impl<'a> PaintContext<'a> {
}
}
// The following comment is wonderful, and stolen from
// gecko:gfx/thebes/gfxContext.cpp:RoundedRectangle for reference.
//
// It does not currently apply to the code, but will be extremely useful in
// the future when the below TODO is addressed.
//
// TODO(cgaebel): Switch from arcs to beziers for drawing the corners.
// Then, add http://www.subcide.com/experiments/fail-whale/
// to the reftest suite.
//
// ---------------------------------------------------------------
//
// For CW drawing, this looks like:
//
// ...******0** 1 C
// ****
// *** 2
// **
// *
// *
// 3
// *
// *
//
// Where 0, 1, 2, 3 are the control points of the Bezier curve for
// the corner, and C is the actual corner point.
//
// At the start of the loop, the current point is assumed to be
// the point adjacent to the top left corner on the top
// horizontal. Note that corner indices start at the top left and
// continue clockwise, whereas in our loop i = 0 refers to the top
// right corner.
//
// When going CCW, the control points are swapped, and the first
// corner that's drawn is the top left (along with the top segment).
//
// There is considerable latitude in how one chooses the four
// control points for a Bezier curve approximation to an ellipse.
// For the overall path to be continuous and show no corner at the
// endpoints of the arc, points 0 and 3 must be at the ends of the
// straight segments of the rectangle; points 0, 1, and C must be
// collinear; and points 3, 2, and C must also be collinear. This
// leaves only two free parameters: the ratio of the line segments
// 01 and 0C, and the ratio of the line segments 32 and 3C. See
// the following papers for extensive discussion of how to choose
// these ratios:
//
// Dokken, Tor, et al. "Good approximation of circles by
// curvature-continuous Bezier curves." Computer-Aided
// Geometric Design 7(1990) 33--41.
// Goldapp, Michael. "Approximation of circular arcs by cubic
// polynomials." Computer-Aided Geometric Design 8(1991) 227--238.
// Maisonobe, Luc. "Drawing an elliptical arc using polylines,
// quadratic, or cubic Bezier curves."
// http://www.spaceroots.org/documents/ellipse/elliptical-arc.pdf
//
// We follow the approach in section 2 of Goldapp (least-error,
// Hermite-type approximation) and make both ratios equal to
//
// 2 2 + n - sqrt(2n + 28)
// alpha = - * ---------------------
// 3 n - 4
//
// where n = 3( cbrt(sqrt(2)+1) - cbrt(sqrt(2)-1) ).
//
// This is the result of Goldapp's equation (10b) when the angle
// swept out by the arc is pi/2, and the parameter "a-bar" is the
// expression given immediately below equation (21).
//
// Using this value, the maximum radial error for a circle, as a
// fraction of the radius, is on the order of 0.2 x 10^-3.
// Neither Dokken nor Goldapp discusses error for a general
// ellipse; Maisonobe does, but his choice of control points
// follows different constraints, and Goldapp's expression for
// 'alpha' gives much smaller radial error, even for very flat
// ellipses, than Maisonobe's equivalent.
//
// For the various corners and for each axis, the sign of this
// constant changes, or it might be 0 -- it's multiplied by the
// appropriate multiplier from the list before using.
#[allow(non_snake_case)]
fn create_border_path_segment(&self,
path_builder: &mut PathBuilder,

File diff suppressed because it is too large Load diff