summaryrefslogtreecommitdiffstats
path: root/src/3rdparty/2geom/src/2geom/elliptical-arc.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/3rdparty/2geom/src/2geom/elliptical-arc.cpp')
-rw-r--r--src/3rdparty/2geom/src/2geom/elliptical-arc.cpp962
1 files changed, 962 insertions, 0 deletions
diff --git a/src/3rdparty/2geom/src/2geom/elliptical-arc.cpp b/src/3rdparty/2geom/src/2geom/elliptical-arc.cpp
new file mode 100644
index 0000000..96837bc
--- /dev/null
+++ b/src/3rdparty/2geom/src/2geom/elliptical-arc.cpp
@@ -0,0 +1,962 @@
+/*
+ * SVG Elliptical Arc Class
+ *
+ * Authors:
+ * Marco Cecchetti <mrcekets at gmail.com>
+ * Krzysztof KosiƄski <tweenk.pl@gmail.com>
+ * Copyright 2008-2009 Authors
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it either under the terms of the GNU Lesser General Public
+ * License version 2.1 as published by the Free Software Foundation
+ * (the "LGPL") or, at your option, under the terms of the Mozilla
+ * Public License Version 1.1 (the "MPL"). If you do not alter this
+ * notice, a recipient may use your version of this file under either
+ * the MPL or the LGPL.
+ *
+ * You should have received a copy of the LGPL along with this library
+ * in the file COPYING-LGPL-2.1; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ * You should have received a copy of the MPL along with this library
+ * in the file COPYING-MPL-1.1
+ *
+ * The contents of this file are subject to the Mozilla Public License
+ * Version 1.1 (the "License"); you may not use this file except in
+ * compliance with the License. You may obtain a copy of the License at
+ * http://www.mozilla.org/MPL/
+ *
+ * This software is distributed on an "AS IS" basis, WITHOUT WARRANTY
+ * OF ANY KIND, either express or implied. See the LGPL or the MPL for
+ * the specific language governing rights and limitations.
+ */
+
+#include <cfloat>
+#include <limits>
+#include <memory>
+
+#include <2geom/bezier-curve.h>
+#include <2geom/ellipse.h>
+#include <2geom/elliptical-arc.h>
+#include <2geom/path-sink.h>
+#include <2geom/sbasis-geometric.h>
+#include <2geom/transforms.h>
+#include <2geom/utils.h>
+
+#include <2geom/numeric/vector.h>
+#include <2geom/numeric/fitting-tool.h>
+#include <2geom/numeric/fitting-model.h>
+
+namespace Geom
+{
+
+/**
+ * @class EllipticalArc
+ * @brief Elliptical arc curve
+ *
+ * Elliptical arc is a curve taking the shape of a section of an ellipse.
+ *
+ * The arc function has two forms: the regular one, mapping the unit interval to points
+ * on 2D plane (the linear domain), and a second form that maps some interval
+ * \f$A \subseteq [0,2\pi)\f$ to the same points (the angular domain). The interval \f$A\f$
+ * determines which part of the ellipse forms the arc. The arc is said to contain an angle
+ * if its angular domain includes that angle (and therefore it is defined for that angle).
+ *
+ * The angular domain considers each ellipse to be
+ * a rotated, scaled and translated unit circle: 0 corresponds to \f$(1,0)\f$ on the unit circle,
+ * \f$\pi/2\f$ corresponds to \f$(0,1)\f$, \f$\pi\f$ to \f$(-1,0)\f$ and \f$3\pi/2\f$
+ * to \f$(0,-1)\f$. After the angle is mapped to a point from a unit circle, the point is
+ * transformed using a matrix of this form
+ * \f[ M = \left[ \begin{array}{ccc}
+ r_X \cos(\theta) & -r_Y \sin(\theta) & 0 \\
+ r_X \sin(\theta) & r_Y \cos(\theta) & 0 \\
+ c_X & c_Y & 1 \end{array} \right] \f]
+ * where \f$r_X, r_Y\f$ are the X and Y rays of the ellipse, \f$\theta\f$ is its angle of rotation,
+ * and \f$c_X, c_Y\f$ the coordinates of the ellipse's center - thus mapping the angle
+ * to some point on the ellipse. Note that for example the point at angluar coordinate 0,
+ * the center and the point at angular coordinate \f$\pi/4\f$ do not necessarily
+ * create an angle of \f$\pi/4\f$ radians; it is only the case if both axes of the ellipse
+ * are of the same length (i.e. it is a circle).
+ *
+ * @image html ellipse-angular-coordinates.png "An illustration of the angular domain"
+ *
+ * Each arc is defined by five variables: The initial and final point, the ellipse's rays,
+ * and the ellipse's rotation. Each set of those parameters corresponds to four different arcs,
+ * with two of them larger than half an ellipse and two of them turning clockwise while traveling
+ * from initial to final point. The two flags disambiguate between them: "large arc flag" selects
+ * the bigger arc, while the "sweep flag" selects the arc going in the direction of positive
+ * angles. Angles always increase when going from the +X axis in the direction of the +Y axis,
+ * so if Y grows downwards, this means clockwise.
+ *
+ * @image html elliptical-arc-flags.png "Meaning of arc flags (Y grows downwards)"
+ *
+ * @ingroup Curves
+ */
+
+
+/** @brief Compute bounds of an elliptical arc.
+ * The bounds computation works as follows. The extreme X and Y points
+ * are either the endpoints or local minima / maxima of the ellipse.
+ * We already have endpoints, and we can find the local extremes
+ * by computing a partial derivative with respect to the angle
+ * and equating that to zero:
+ * \f{align*}{
+ x &= r_x \cos \varphi \cos \theta - r_y \sin \varphi \sin \theta + c_x \\
+ \frac{\partial x}{\partial \theta} &= -r_x \cos \varphi \sin \theta - r_y \sin \varphi \cos \theta = 0 \\
+ \frac{\sin \theta}{\cos \theta} &= \tan\theta = -\frac{r_y \sin \varphi}{r_x \cos \varphi} \\
+ \theta &= \tan^{-1} \frac{-r_y \sin \varphi}{r_x \cos \varphi}
+ \f}
+ * The local extremes correspond to two angles separated by \f$\pi\f$.
+ * Once we compute these angles, we check whether they belong to the arc,
+ * and if they do, we evaluate the ellipse at these angles.
+ * The bounding box of the arc is equal to the bounding box of the endpoints
+ * and the local extrema that belong to the arc.
+ */
+Rect EllipticalArc::boundsExact() const
+{
+ if (isChord()) {
+ return chord().boundsExact();
+ }
+
+ Coord coord[2][4] = {
+ { _initial_point[X], _final_point[X], 0, 0 },
+ { _initial_point[Y], _final_point[Y], 0, 0 }
+ };
+ int ncoord[2] = { 2, 2 };
+
+ Angle extremes[2][2];
+ double sinrot, cosrot;
+ sincos(rotationAngle(), sinrot, cosrot);
+
+ extremes[X][0] = std::atan2( -ray(Y) * sinrot, ray(X) * cosrot );
+ extremes[X][1] = extremes[X][0] + M_PI;
+ extremes[Y][0] = std::atan2( ray(Y) * cosrot, ray(X) * sinrot );
+ extremes[Y][1] = extremes[Y][0] + M_PI;
+
+ for (unsigned d = 0; d < 2; ++d) {
+ for (unsigned i = 0; i < 2; ++i) {
+ if (containsAngle(extremes[d][i])) {
+ coord[d][ncoord[d]++] = valueAtAngle(extremes[d][i], d ? Y : X);
+ }
+ }
+ }
+
+ Interval xival = Interval::from_range(coord[X], coord[X] + ncoord[X]);
+ Interval yival = Interval::from_range(coord[Y], coord[Y] + ncoord[Y]);
+ Rect result(xival, yival);
+ return result;
+}
+
+
+Point EllipticalArc::pointAtAngle(Coord t) const
+{
+ Point ret = _ellipse.pointAt(t);
+ return ret;
+}
+
+Coord EllipticalArc::valueAtAngle(Coord t, Dim2 d) const
+{
+ return _ellipse.valueAt(t, d);
+}
+
+std::vector<Coord> EllipticalArc::roots(Coord v, Dim2 d) const
+{
+ std::vector<Coord> sol;
+
+ if (isChord()) {
+ sol = chord().roots(v, d);
+ return sol;
+ }
+
+ Interval unit_interval(0, 1);
+
+ double rotx, roty;
+ if (d == X) {
+ sincos(rotationAngle(), roty, rotx);
+ roty = -roty;
+ } else {
+ sincos(rotationAngle(), rotx, roty);
+ }
+
+ double rxrotx = ray(X) * rotx;
+ double c_v = center(d) - v;
+
+ double a = -rxrotx + c_v;
+ double b = ray(Y) * roty;
+ double c = rxrotx + c_v;
+ //std::cerr << "a = " << a << std::endl;
+ //std::cerr << "b = " << b << std::endl;
+ //std::cerr << "c = " << c << std::endl;
+
+ if (a == 0)
+ {
+ sol.push_back(M_PI);
+ if (b != 0)
+ {
+ double s = 2 * std::atan(-c/(2*b));
+ if ( s < 0 ) s += 2*M_PI;
+ sol.push_back(s);
+ }
+ }
+ else
+ {
+ double delta = b * b - a * c;
+ //std::cerr << "delta = " << delta << std::endl;
+ if (delta == 0) {
+ double s = 2 * std::atan(-b/a);
+ if ( s < 0 ) s += 2*M_PI;
+ sol.push_back(s);
+ }
+ else if ( delta > 0 )
+ {
+ double sq = std::sqrt(delta);
+ double s = 2 * std::atan( (-b - sq) / a );
+ if ( s < 0 ) s += 2*M_PI;
+ sol.push_back(s);
+ s = 2 * std::atan( (-b + sq) / a );
+ if ( s < 0 ) s += 2*M_PI;
+ sol.push_back(s);
+ }
+ }
+
+ std::vector<double> arc_sol;
+ for (double & i : sol) {
+ //std::cerr << "s = " << deg_from_rad(sol[i]);
+ i = timeAtAngle(i);
+ //std::cerr << " -> t: " << sol[i] << std::endl;
+ if (unit_interval.contains(i)) {
+ arc_sol.push_back(i);
+ }
+ }
+ return arc_sol;
+}
+
+
+// D(E(t,C),t) = E(t+PI/2,O), where C is the ellipse center
+// the derivative doesn't rotate the ellipse but there is a translation
+// of the parameter t by an angle of PI/2 so the ellipse points are shifted
+// of such an angle in the cw direction
+Curve *EllipticalArc::derivative() const
+{
+ if (isChord()) {
+ return chord().derivative();
+ }
+
+ EllipticalArc *result = static_cast<EllipticalArc*>(duplicate());
+ result->_ellipse.setCenter(0, 0);
+ result->_angles.setInitial(result->_angles.initialAngle() + M_PI/2);
+ result->_angles.setFinal(result->_angles.finalAngle() + M_PI/2);
+ result->_initial_point = result->pointAtAngle( result->initialAngle() );
+ result->_final_point = result->pointAtAngle( result->finalAngle() );
+ return result;
+}
+
+
+std::vector<Point>
+EllipticalArc::pointAndDerivatives(Coord t, unsigned int n) const
+{
+ if (isChord()) {
+ return chord().pointAndDerivatives(t, n);
+ }
+
+ unsigned int nn = n+1; // nn represents the size of the result vector.
+ std::vector<Point> result;
+ result.reserve(nn);
+ double angle = angleAt(t);
+ std::unique_ptr<EllipticalArc> ea( static_cast<EllipticalArc*>(duplicate()) );
+ ea->_ellipse.setCenter(0, 0);
+ unsigned int m = std::min(nn, 4u);
+ for ( unsigned int i = 0; i < m; ++i )
+ {
+ result.push_back( ea->pointAtAngle(angle) );
+ angle += (sweep() ? M_PI/2 : -M_PI/2);
+ if ( !(angle < 2*M_PI) ) angle -= 2*M_PI;
+ }
+ m = nn / 4;
+ for ( unsigned int i = 1; i < m; ++i )
+ {
+ for ( unsigned int j = 0; j < 4; ++j )
+ result.push_back( result[j] );
+ }
+ m = nn - 4 * m;
+ for ( unsigned int i = 0; i < m; ++i )
+ {
+ result.push_back( result[i] );
+ }
+ if ( !result.empty() ) // nn != 0
+ result[0] = pointAtAngle(angle);
+ return result;
+}
+
+Point EllipticalArc::pointAt(Coord t) const
+{
+ if (isChord()) return chord().pointAt(t);
+ return _ellipse.pointAt(angleAt(t));
+}
+
+Coord EllipticalArc::valueAt(Coord t, Dim2 d) const
+{
+ if (isChord()) return chord().valueAt(t, d);
+ return valueAtAngle(angleAt(t), d);
+}
+
+Curve* EllipticalArc::portion(double f, double t) const
+{
+ // fix input arguments
+ if (f < 0) f = 0;
+ if (f > 1) f = 1;
+ if (t < 0) t = 0;
+ if (t > 1) t = 1;
+
+ if (f == t) {
+ EllipticalArc *arc = new EllipticalArc();
+ arc->_initial_point = arc->_final_point = pointAt(f);
+ return arc;
+ }
+
+ EllipticalArc *arc = static_cast<EllipticalArc*>(duplicate());
+ arc->_initial_point = pointAt(f);
+ arc->_final_point = pointAt(t);
+ arc->_angles.setAngles(angleAt(f), angleAt(t));
+ if (f > t) arc->_angles.setSweep(!sweep());
+ if ( _large_arc && fabs(angularExtent() * (t-f)) <= M_PI) {
+ arc->_large_arc = false;
+ }
+ return arc;
+}
+
+// the arc is the same but traversed in the opposite direction
+Curve *EllipticalArc::reverse() const
+{
+ using std::swap;
+ EllipticalArc *rarc = static_cast<EllipticalArc*>(duplicate());
+ rarc->_angles.reverse();
+ swap(rarc->_initial_point, rarc->_final_point);
+ return rarc;
+}
+
+#ifdef HAVE_GSL // GSL is required for function "solve_reals"
+std::vector<double> EllipticalArc::allNearestTimes( Point const& p, double from, double to ) const
+{
+ std::vector<double> result;
+
+ if ( from > to ) std::swap(from, to);
+ if ( from < 0 || to > 1 )
+ {
+ THROW_RANGEERROR("[from,to] interval out of range");
+ }
+
+ if ( ( are_near(ray(X), 0) && are_near(ray(Y), 0) ) || are_near(from, to) )
+ {
+ result.push_back(from);
+ return result;
+ }
+ else if ( are_near(ray(X), 0) || are_near(ray(Y), 0) )
+ {
+ LineSegment seg(pointAt(from), pointAt(to));
+ Point np = seg.pointAt( seg.nearestTime(p) );
+ if ( are_near(ray(Y), 0) )
+ {
+ if ( are_near(rotationAngle(), M_PI/2)
+ || are_near(rotationAngle(), 3*M_PI/2) )
+ {
+ result = roots(np[Y], Y);
+ }
+ else
+ {
+ result = roots(np[X], X);
+ }
+ }
+ else
+ {
+ if ( are_near(rotationAngle(), M_PI/2)
+ || are_near(rotationAngle(), 3*M_PI/2) )
+ {
+ result = roots(np[X], X);
+ }
+ else
+ {
+ result = roots(np[Y], Y);
+ }
+ }
+ return result;
+ }
+ else if ( are_near(ray(X), ray(Y)) )
+ {
+ Point r = p - center();
+ if ( are_near(r, Point(0,0)) )
+ {
+ THROW_INFINITESOLUTIONS(0);
+ }
+ // TODO: implement case r != 0
+// Point np = ray(X) * unit_vector(r);
+// std::vector<double> solX = roots(np[X],X);
+// std::vector<double> solY = roots(np[Y],Y);
+// double t;
+// if ( are_near(solX[0], solY[0]) || are_near(solX[0], solY[1]))
+// {
+// t = solX[0];
+// }
+// else
+// {
+// t = solX[1];
+// }
+// if ( !(t < from || t > to) )
+// {
+// result.push_back(t);
+// }
+// else
+// {
+//
+// }
+ }
+
+ // solve the equation <D(E(t),t)|E(t)-p> == 0
+ // that provides min and max distance points
+ // on the ellipse E wrt the point p
+ // after the substitutions:
+ // cos(t) = (1 - s^2) / (1 + s^2)
+ // sin(t) = 2t / (1 + s^2)
+ // where s = tan(t/2)
+ // we get a 4th degree equation in s
+ /*
+ * ry s^4 ((-cy + py) Cos[Phi] + (cx - px) Sin[Phi]) +
+ * ry ((cy - py) Cos[Phi] + (-cx + px) Sin[Phi]) +
+ * 2 s^3 (rx^2 - ry^2 + (-cx + px) rx Cos[Phi] + (-cy + py) rx Sin[Phi]) +
+ * 2 s (-rx^2 + ry^2 + (-cx + px) rx Cos[Phi] + (-cy + py) rx Sin[Phi])
+ */
+
+ Point p_c = p - center();
+ double rx2_ry2 = (ray(X) - ray(Y)) * (ray(X) + ray(Y));
+ double sinrot, cosrot;
+ sincos(rotationAngle(), sinrot, cosrot);
+ double expr1 = ray(X) * (p_c[X] * cosrot + p_c[Y] * sinrot);
+ Poly coeff;
+ coeff.resize(5);
+ coeff[4] = ray(Y) * ( p_c[Y] * cosrot - p_c[X] * sinrot );
+ coeff[3] = 2 * ( rx2_ry2 + expr1 );
+ coeff[2] = 0;
+ coeff[1] = 2 * ( -rx2_ry2 + expr1 );
+ coeff[0] = -coeff[4];
+
+// for ( unsigned int i = 0; i < 5; ++i )
+// std::cerr << "c[" << i << "] = " << coeff[i] << std::endl;
+
+ std::vector<double> real_sol;
+ // gsl_poly_complex_solve raises an error
+ // if the leading coefficient is zero
+ if ( are_near(coeff[4], 0) )
+ {
+ real_sol.push_back(0);
+ if ( !are_near(coeff[3], 0) )
+ {
+ double sq = -coeff[1] / coeff[3];
+ if ( sq > 0 )
+ {
+ double s = std::sqrt(sq);
+ real_sol.push_back(s);
+ real_sol.push_back(-s);
+ }
+ }
+ }
+ else
+ {
+ real_sol = solve_reals(coeff);
+ }
+
+ for (double & i : real_sol)
+ {
+ i = 2 * std::atan(i);
+ if ( i < 0 ) i += 2*M_PI;
+ }
+ // when s -> Infinity then <D(E)|E-p> -> 0 iff coeff[4] == 0
+ // so we add M_PI to the solutions being lim arctan(s) = PI when s->Infinity
+ if ( (real_sol.size() % 2) != 0 )
+ {
+ real_sol.push_back(M_PI);
+ }
+
+ double mindistsq1 = std::numeric_limits<double>::max();
+ double mindistsq2 = std::numeric_limits<double>::max();
+ double dsq = 0;
+ unsigned int mi1 = 0, mi2 = 0;
+ for ( unsigned int i = 0; i < real_sol.size(); ++i )
+ {
+ dsq = distanceSq(p, pointAtAngle(real_sol[i]));
+ if ( mindistsq1 > dsq )
+ {
+ mindistsq2 = mindistsq1;
+ mi2 = mi1;
+ mindistsq1 = dsq;
+ mi1 = i;
+ }
+ else if ( mindistsq2 > dsq )
+ {
+ mindistsq2 = dsq;
+ mi2 = i;
+ }
+ }
+
+ double t = timeAtAngle(real_sol[mi1]);
+ if ( !(t < from || t > to) )
+ {
+ result.push_back(t);
+ }
+
+ bool second_sol = false;
+ t = timeAtAngle(real_sol[mi2]);
+ if ( real_sol.size() == 4 && !(t < from || t > to) )
+ {
+ if ( result.empty() || are_near(mindistsq1, mindistsq2) )
+ {
+ result.push_back(t);
+ second_sol = true;
+ }
+ }
+
+ // we need to test extreme points too
+ double dsq1 = distanceSq(p, pointAt(from));
+ double dsq2 = distanceSq(p, pointAt(to));
+ if ( second_sol )
+ {
+ if ( mindistsq2 > dsq1 )
+ {
+ result.clear();
+ result.push_back(from);
+ mindistsq2 = dsq1;
+ }
+ else if ( are_near(mindistsq2, dsq) )
+ {
+ result.push_back(from);
+ }
+ if ( mindistsq2 > dsq2 )
+ {
+ result.clear();
+ result.push_back(to);
+ }
+ else if ( are_near(mindistsq2, dsq2) )
+ {
+ result.push_back(to);
+ }
+
+ }
+ else
+ {
+ if ( result.empty() )
+ {
+ if ( are_near(dsq1, dsq2) )
+ {
+ result.push_back(from);
+ result.push_back(to);
+ }
+ else if ( dsq2 > dsq1 )
+ {
+ result.push_back(from);
+ }
+ else
+ {
+ result.push_back(to);
+ }
+ }
+ }
+
+ return result;
+}
+#endif
+
+/** @brief Convert the passed intersections to curve time parametrization
+ * and filter out any invalid intersections.
+ */
+std::vector<ShapeIntersection> EllipticalArc::_filterIntersections(std::vector<ShapeIntersection> &&xs,
+ bool is_first) const
+{
+ std::vector<ShapeIntersection> result;
+ result.reserve(xs.size());
+ for (auto &xing : xs) {
+ if (_validateIntersection(xing, is_first)) {
+ result.emplace_back(std::move(xing));
+ }
+ }
+ return result;
+}
+
+/** @brief Convert the passed intersection to curve time and check whether the intersection
+ * is numerically sane.
+ *
+ * @param xing The intersection to convert to curve time and to validate.
+ * @param is_first If true, this arc is the first of the intersected curves; if false, it's second.
+ * @return Whether the intersection is valid.
+ *
+ * Note that the intersection is guaranteed to be converted only if the return value is true.
+ */
+bool EllipticalArc::_validateIntersection(ShapeIntersection &xing, bool is_first) const
+{
+ static auto const UNIT_INTERVAL = Interval(0, 1);
+ constexpr auto EPS = 1e-4;
+
+ Coord &t = is_first ? xing.first : xing.second;
+ if (!are_near_rel(_ellipse.pointAt(t), xing.point(), EPS)) {
+ return false;
+ }
+
+ t = timeAtAngle(t);
+ if (!UNIT_INTERVAL.contains(t)) {
+ return false;
+ }
+ if (!are_near_rel(pointAt(t), xing.point(), EPS)) {
+ return false;
+ }
+ return true;
+}
+
+std::vector<CurveIntersection> EllipticalArc::intersect(Curve const &other, Coord eps) const
+{
+ if (isLineSegment()) {
+ LineSegment ls(_initial_point, _final_point);
+ return ls.intersect(other, eps);
+ }
+
+ if (other.isLineSegment()) {
+ LineSegment ls(other.initialPoint(), other.finalPoint());
+ return _filterIntersections(_ellipse.intersect(ls), true);
+ }
+
+ if (auto bez = dynamic_cast<BezierCurve const *>(&other)) {
+ return _filterIntersections(_ellipse.intersect(bez->fragment()), true);
+ }
+
+ if (auto arc = dynamic_cast<EllipticalArc const *>(&other)) {
+ auto result = _filterIntersections(_ellipse.intersect(arc->_ellipse), true);
+ return arc->_filterIntersections(std::move(result), false);
+ }
+
+ // in case someone wants to make a custom curve type
+ auto result = other.intersect(*this, eps);
+ transpose_in_place(result);
+ return result;
+}
+
+
+void EllipticalArc::_updateCenterAndAngles()
+{
+ // See: http://www.w3.org/TR/SVG/implnote.html#ArcImplementationNotes
+ Point d = initialPoint() - finalPoint();
+ Point mid = middle_point(initialPoint(), finalPoint());
+
+ // if ip = sp, the arc contains no other points
+ if (initialPoint() == finalPoint()) {
+ _ellipse = Ellipse();
+ _ellipse.setCenter(initialPoint());
+ _angles = AngleInterval();
+ _large_arc = false;
+ return;
+ }
+
+ // rays should be positive
+ _ellipse.setRays(std::fabs(ray(X)), std::fabs(ray(Y)));
+
+ if (isChord()) {
+ _ellipse.setRays(L2(d) / 2, 0);
+ _ellipse.setRotationAngle(atan2(d));
+ _ellipse.setCenter(mid);
+ _angles.setAngles(0, M_PI);
+ _angles.setSweep(false);
+ _large_arc = false;
+ return;
+ }
+
+ Rotate rot(rotationAngle()); // the matrix in F.6.5.3
+ Rotate invrot = rot.inverse(); // the matrix in F.6.5.1
+
+ Point r = rays();
+ Point p = (initialPoint() - mid) * invrot; // x', y' in F.6.5.1
+ Point c(0,0); // cx', cy' in F.6.5.2
+
+ // Correct out-of-range radii
+ Coord lambda = hypot(p[X]/r[X], p[Y]/r[Y]);
+ if (lambda > 1) {
+ r *= lambda;
+ _ellipse.setRays(r);
+ _ellipse.setCenter(mid);
+ } else {
+ // evaluate F.6.5.2
+ Coord rxry = r[X]*r[X] * r[Y]*r[Y];
+ Coord pxry = p[X]*p[X] * r[Y]*r[Y];
+ Coord rxpy = r[X]*r[X] * p[Y]*p[Y];
+ Coord rad = (rxry - pxry - rxpy)/(rxpy + pxry);
+ // normally rad should never be negative, but numerical inaccuracy may cause this
+ if (rad > 0) {
+ rad = std::sqrt(rad);
+ if (sweep() == _large_arc) {
+ rad = -rad;
+ }
+ c = rad * Point(r[X]*p[Y]/r[Y], -r[Y]*p[X]/r[X]);
+ _ellipse.setCenter(c * rot + mid);
+ } else {
+ _ellipse.setCenter(mid);
+ }
+ }
+
+ // Compute start and end angles.
+ // If the ellipse was enlarged, c will be zero - this is correct.
+ Point sp((p[X] - c[X]) / r[X], (p[Y] - c[Y]) / r[Y]);
+ Point ep((-p[X] - c[X]) / r[X], (-p[Y] - c[Y]) / r[Y]);
+ Point v(1, 0);
+
+ _angles.setInitial(angle_between(v, sp));
+ _angles.setFinal(angle_between(v, ep));
+
+ /*double sweep_angle = angle_between(sp, ep);
+ if (!sweep() && sweep_angle > 0) sweep_angle -= 2*M_PI;
+ if (sweep() && sweep_angle < 0) sweep_angle += 2*M_PI;*/
+}
+
+D2<SBasis> EllipticalArc::toSBasis() const
+{
+ if (isChord()) {
+ return chord().toSBasis();
+ }
+
+ D2<SBasis> arc;
+ // the interval of parametrization has to be [0,1]
+ Coord et = initialAngle().radians() + sweepAngle();
+ Linear param(initialAngle().radians(), et);
+ Coord cosrot, sinrot;
+ sincos(rotationAngle(), sinrot, cosrot);
+
+ // order = 4 seems to be enough to get a perfect looking elliptical arc
+ SBasis arc_x = ray(X) * cos(param,4);
+ SBasis arc_y = ray(Y) * sin(param,4);
+ arc[0] = arc_x * cosrot - arc_y * sinrot + Linear(center(X), center(X));
+ arc[1] = arc_x * sinrot + arc_y * cosrot + Linear(center(Y), center(Y));
+
+ // ensure that endpoints remain exact
+ for ( int d = 0 ; d < 2 ; d++ ) {
+ arc[d][0][0] = initialPoint()[d];
+ arc[d][0][1] = finalPoint()[d];
+ }
+
+ return arc;
+}
+
+// All operations that do not contain skew can be evaluated
+// without passing through the implicit form of the ellipse,
+// which preserves precision.
+
+void EllipticalArc::operator*=(Translate const &tr)
+{
+ _initial_point *= tr;
+ _final_point *= tr;
+ _ellipse *= tr;
+}
+
+void EllipticalArc::operator*=(Scale const &s)
+{
+ _initial_point *= s;
+ _final_point *= s;
+ _ellipse *= s;
+}
+
+void EllipticalArc::operator*=(Rotate const &r)
+{
+ _initial_point *= r;
+ _final_point *= r;
+ _ellipse *= r;
+}
+
+void EllipticalArc::operator*=(Zoom const &z)
+{
+ _initial_point *= z;
+ _final_point *= z;
+ _ellipse *= z;
+}
+
+void EllipticalArc::operator*=(Affine const& m)
+{
+ if (isChord()) {
+ _initial_point *= m;
+ _final_point *= m;
+ _ellipse.setCenter(middle_point(_initial_point, _final_point));
+ _ellipse.setRays(0, 0);
+ _ellipse.setRotationAngle(0);
+ return;
+ }
+
+ _initial_point *= m;
+ _final_point *= m;
+ _ellipse *= m;
+ if (m.det() < 0) {
+ _angles.setSweep(!sweep());
+ }
+
+ // ellipse transformation does not preserve its functional form,
+ // i.e. e.pointAt(0.5)*m and (e*m).pointAt(0.5) can be different.
+ // We need to recompute start / end angles.
+ _angles.setInitial(_ellipse.timeAt(_initial_point));
+ _angles.setFinal(_ellipse.timeAt(_final_point));
+}
+
+bool EllipticalArc::operator==(Curve const &c) const
+{
+ EllipticalArc const *other = dynamic_cast<EllipticalArc const *>(&c);
+ if (!other) return false;
+ if (_initial_point != other->_initial_point) return false;
+ if (_final_point != other->_final_point) return false;
+ // TODO: all arcs with ellipse rays which are too small
+ // and fall back to a line should probably be equal
+ if (rays() != other->rays()) return false;
+ if (rotationAngle() != other->rotationAngle()) return false;
+ if (_large_arc != other->_large_arc) return false;
+ if (sweep() != other->sweep()) return false;
+ return true;
+}
+
+bool EllipticalArc::isNear(Curve const &c, Coord precision) const
+{
+ EllipticalArc const *other = dynamic_cast<EllipticalArc const *>(&c);
+ if (!other) {
+ if (isChord()) {
+ return c.isNear(chord(), precision);
+ }
+ return false;
+ }
+
+ if (!are_near(_initial_point, other->_initial_point, precision)) return false;
+ if (!are_near(_final_point, other->_final_point, precision)) return false;
+ if (isChord() && other->isChord()) return true;
+
+ if (sweep() != other->sweep()) return false;
+ if (!are_near(_ellipse, other->_ellipse, precision)) return false;
+ return true;
+}
+
+void EllipticalArc::feed(PathSink &sink, bool moveto_initial) const
+{
+ if (moveto_initial) {
+ sink.moveTo(_initial_point);
+ }
+ sink.arcTo(ray(X), ray(Y), rotationAngle(), _large_arc, sweep(), _final_point);
+}
+
+int EllipticalArc::winding(Point const &p) const
+{
+ using std::swap;
+
+ double sinrot, cosrot;
+ sincos(rotationAngle(), sinrot, cosrot);
+
+ Angle ymin_a = std::atan2( ray(Y) * cosrot, ray(X) * sinrot );
+ Angle ymax_a = ymin_a + M_PI;
+
+ Point ymin = pointAtAngle(ymin_a);
+ Point ymax = pointAtAngle(ymax_a);
+ if (ymin[Y] > ymax[Y]) {
+ swap(ymin, ymax);
+ swap(ymin_a, ymax_a);
+ }
+
+ Interval yspan(ymin[Y], ymax[Y]);
+ if (!yspan.lowerContains(p[Y])) return 0;
+
+ bool left = cross(ymax - ymin, p - ymin) > 0;
+ bool inside = _ellipse.contains(p);
+ bool includes_ymin = _angles.contains(ymin_a);
+ bool includes_ymax = _angles.contains(ymax_a);
+
+ AngleInterval rarc(ymin_a, ymax_a, true),
+ larc(ymax_a, ymin_a, true);
+
+ // we'll compute the result for an arc in the direction of increasing angles
+ // and then negate if necessary
+ Angle ia = initialAngle(), fa = finalAngle();
+ Point ip = _initial_point, fp = _final_point;
+ if (!sweep()) {
+ swap(ia, fa);
+ swap(ip, fp);
+ }
+
+ bool initial_left = larc.contains(ia);
+ bool initial_right = !initial_left; // rarc.contains(ia);
+ bool final_left = larc.contains(fa);
+ bool final_right = !final_left; // rarc.contains(fa);
+
+ int result = 0;
+ if (inside || left) {
+ if (includes_ymin && final_right) {
+ Interval ival(ymin[Y], fp[Y]);
+ if (ival.lowerContains(p[Y])) {
+ ++result;
+ }
+ }
+ if (initial_right && final_right && !largeArc()) {
+ Interval ival(ip[Y], fp[Y]);
+ if (ival.lowerContains(p[Y])) {
+ ++result;
+ }
+ }
+ if (initial_right && includes_ymax) {
+ Interval ival(ip[Y], ymax[Y]);
+ if (ival.lowerContains(p[Y])) {
+ ++result;
+ }
+ }
+ if (!initial_right && !final_right && includes_ymin && includes_ymax) {
+ Interval ival(ymax[Y], ymin[Y]);
+ if (ival.lowerContains(p[Y])) {
+ ++result;
+ }
+ }
+ }
+ if (left && !inside) {
+ if (includes_ymin && initial_left) {
+ Interval ival(ymin[Y], ip[Y]);
+ if (ival.lowerContains(p[Y])) {
+ --result;
+ }
+ }
+ if (initial_left && final_left && !largeArc()) {
+ Interval ival(ip[Y], fp[Y]);
+ if (ival.lowerContains(p[Y])) {
+ --result;
+ }
+ }
+ if (final_left && includes_ymax) {
+ Interval ival(fp[Y], ymax[Y]);
+ if (ival.lowerContains(p[Y])) {
+ --result;
+ }
+ }
+ if (!initial_left && !final_left && includes_ymin && includes_ymax) {
+ Interval ival(ymax[Y], ymin[Y]);
+ if (ival.lowerContains(p[Y])) {
+ --result;
+ }
+ }
+ }
+ return sweep() ? result : -result;
+}
+
+std::ostream &operator<<(std::ostream &out, EllipticalArc const &ea)
+{
+ out << "EllipticalArc("
+ << ea.initialPoint() << ", "
+ << format_coord_nice(ea.ray(X)) << ", " << format_coord_nice(ea.ray(Y)) << ", "
+ << format_coord_nice(ea.rotationAngle()) << ", "
+ << "large_arc=" << (ea.largeArc() ? "true" : "false") << ", "
+ << "sweep=" << (ea.sweep() ? "true" : "false") << ", "
+ << ea.finalPoint() << ")";
+ return out;
+}
+
+} // end namespace Geom
+
+/*
+ Local Variables:
+ mode:c++
+ c-file-style:"stroustrup"
+ c-file-offsets:((innamespace . 0)(inline-open . 0)(case-label . +))
+ indent-tabs-mode:nil
+ fill-column:99
+ End:
+*/
+// vim: filetype=cpp:expandtab:shiftwidth=4:tabstop=8:softtabstop=4:fileencoding=utf-8:textwidth=99 :
+