[chore] add back exif-terminator and use only for jpeg,png,webp (#3161)

* add back exif-terminator and use only for jpeg,png,webp

* fix arguments passed to terminateExif()

* pull in latest exif-terminator

* fix test

* update processed img

---------

Co-authored-by: tobi <tobi.smethurst@protonmail.com>

1 files changed, 1212 insertions, 0 deletions

diff --git a/vendor/github.com/golang/geo/s2/polygon.go b/vendor/github.com/golang/geo/s2/polygon.go
new file mode 100644
index 000000000..57c43fbcf
--- /dev/null
+++ b/vendor/github.com/golang/geo/s2/polygon.go
@@ -0,0 +1,1212 @@
+// Copyright 2015 Google Inc. All rights reserved.
+//
+// Licensed under the Apache License, Version 2.0 (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.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package s2
+
+import (
+	"fmt"
+	"io"
+	"math"
+)
+
+// Polygon represents a sequence of zero or more loops; recall that the
+// interior of a loop is defined to be its left-hand side (see Loop).
+//
+// When the polygon is initialized, the given loops are automatically converted
+// into a canonical form consisting of "shells" and "holes". Shells and holes
+// are both oriented CCW, and are nested hierarchically. The loops are
+// reordered to correspond to a pre-order traversal of the nesting hierarchy.
+//
+// Polygons may represent any region of the sphere with a polygonal boundary,
+// including the entire sphere (known as the "full" polygon). The full polygon
+// consists of a single full loop (see Loop), whereas the empty polygon has no
+// loops at all.
+//
+// Use FullPolygon() to construct a full polygon. The zero value of Polygon is
+// treated as the empty polygon.
+//
+// Polygons have the following restrictions:
+//
+//  - Loops may not cross, i.e. the boundary of a loop may not intersect
+//    both the interior and exterior of any other loop.
+//
+//  - Loops may not share edges, i.e. if a loop contains an edge AB, then
+//    no other loop may contain AB or BA.
+//
+//  - Loops may share vertices, however no vertex may appear twice in a
+//    single loop (see Loop).
+//
+//  - No loop may be empty. The full loop may appear only in the full polygon.
+type Polygon struct {
+	loops []*Loop
+
+	// index is a spatial index of all the polygon loops.
+	index *ShapeIndex
+
+	// hasHoles tracks if this polygon has at least one hole.
+	hasHoles bool
+
+	// numVertices keeps the running total of all of the vertices of the contained loops.
+	numVertices int
+
+	// numEdges tracks the total number of edges in all the loops in this polygon.
+	numEdges int
+
+	// bound is a conservative bound on all points contained by this loop.
+	// If l.ContainsPoint(P), then l.bound.ContainsPoint(P).
+	bound Rect
+
+	// Since bound is not exact, it is possible that a loop A contains
+	// another loop B whose bounds are slightly larger. subregionBound
+	// has been expanded sufficiently to account for this error, i.e.
+	// if A.Contains(B), then A.subregionBound.Contains(B.bound).
+	subregionBound Rect
+
+	// A slice where element i is the cumulative number of edges in the
+	// preceding loops in the polygon. This field is used for polygons that
+	// have a large number of loops, and may be empty for polygons with few loops.
+	cumulativeEdges []int
+}
+
+// PolygonFromLoops constructs a polygon from the given set of loops. The polygon
+// interior consists of the points contained by an odd number of loops. (Recall
+// that a loop contains the set of points on its left-hand side.)
+//
+// This method determines the loop nesting hierarchy and assigns every loop a
+// depth. Shells have even depths, and holes have odd depths.
+//
+// Note: The given set of loops are reordered by this method so that the hierarchy
+// can be traversed using Parent, LastDescendant and the loops depths.
+func PolygonFromLoops(loops []*Loop) *Polygon {
+	p := &Polygon{}
+	// Empty polygons do not contain any loops, even the Empty loop.
+	if len(loops) == 1 && loops[0].IsEmpty() {
+		p.initLoopProperties()
+		return p
+	}
+	p.loops = loops
+	p.initNested()
+	return p
+}
+
+// PolygonFromOrientedLoops returns a Polygon from the given set of loops,
+// like PolygonFromLoops. It expects loops to be oriented such that the polygon
+// interior is on the left-hand side of all loops. This implies that shells
+// and holes should have opposite orientations in the input to this method.
+// (During initialization, loops representing holes will automatically be
+// inverted.)
+func PolygonFromOrientedLoops(loops []*Loop) *Polygon {
+	// Here is the algorithm:
+	//
+	// 1. Remember which of the given loops contain OriginPoint.
+	//
+	// 2. Invert loops as necessary to ensure that they are nestable (i.e., no
+	//    loop contains the complement of any other loop). This may result in a
+	//    set of loops corresponding to the complement of the given polygon, but
+	//    we will fix that problem later.
+	//
+	//    We make the loops nestable by first normalizing all the loops (i.e.,
+	//    inverting any loops whose turning angle is negative). This handles
+	//    all loops except those whose turning angle is very close to zero
+	//    (within the maximum error tolerance). Any such loops are inverted if
+	//    and only if they contain OriginPoint(). (In theory this step is only
+	//    necessary if there are at least two such loops.) The resulting set of
+	//    loops is guaranteed to be nestable.
+	//
+	// 3. Build the polygon. This yields either the desired polygon or its
+	//    complement.
+	//
+	// 4. If there is at least one loop, we find a loop L that is adjacent to
+	//    OriginPoint() (where "adjacent" means that there exists a path
+	//    connecting OriginPoint() to some vertex of L such that the path does
+	//    not cross any loop). There may be a single such adjacent loop, or
+	//    there may be several (in which case they should all have the same
+	//    contains_origin() value). We choose L to be the loop containing the
+	//    origin whose depth is greatest, or loop(0) (a top-level shell) if no
+	//    such loop exists.
+	//
+	// 5. If (L originally contained origin) != (polygon contains origin), we
+	//    invert the polygon. This is done by inverting a top-level shell whose
+	//    turning angle is minimal and then fixing the nesting hierarchy. Note
+	//    that because we normalized all the loops initially, this step is only
+	//    necessary if the polygon requires at least one non-normalized loop to
+	//    represent it.
+
+	containedOrigin := make(map[*Loop]bool)
+	for _, l := range loops {
+		containedOrigin[l] = l.ContainsOrigin()
+	}
+
+	for _, l := range loops {
+		angle := l.TurningAngle()
+		if math.Abs(angle) > l.turningAngleMaxError() {
+			// Normalize the loop.
+			if angle < 0 {
+				l.Invert()
+			}
+		} else {
+			// Ensure that the loop does not contain the origin.
+			if l.ContainsOrigin() {
+				l.Invert()
+			}
+		}
+	}
+
+	p := PolygonFromLoops(loops)
+
+	if p.NumLoops() > 0 {
+		originLoop := p.Loop(0)
+		polygonContainsOrigin := false
+		for _, l := range p.Loops() {
+			if l.ContainsOrigin() {
+				polygonContainsOrigin = !polygonContainsOrigin
+
+				originLoop = l
+			}
+		}
+		if containedOrigin[originLoop] != polygonContainsOrigin {
+			p.Invert()
+		}
+	}
+
+	return p
+}
+
+// Invert inverts the polygon (replaces it by its complement).
+func (p *Polygon) Invert() {
+	// Inverting any one loop will invert the polygon.  The best loop to invert
+	// is the one whose area is largest, since this yields the smallest area
+	// after inversion. The loop with the largest area is always at depth 0.
+	// The descendents of this loop all have their depth reduced by 1, while the
+	// former siblings of this loop all have their depth increased by 1.
+
+	// The empty and full polygons are handled specially.
+	if p.IsEmpty() {
+		*p = *FullPolygon()
+		return
+	}
+	if p.IsFull() {
+		*p = Polygon{}
+		return
+	}
+
+	// Find the loop whose area is largest (i.e., whose turning angle is
+	// smallest), minimizing calls to TurningAngle(). In particular, for
+	// polygons with a single shell at level 0 there is no need to call
+	// TurningAngle() at all. (This method is relatively expensive.)
+	best := 0
+	const none = 10.0 // Flag that means "not computed yet"
+	bestAngle := none
+	for i := 1; i < p.NumLoops(); i++ {
+		if p.Loop(i).depth != 0 {
+			continue
+		}
+		// We defer computing the turning angle of loop 0 until we discover
+		// that the polygon has another top-level shell.
+		if bestAngle == none {
+			bestAngle = p.Loop(best).TurningAngle()
+		}
+		angle := p.Loop(i).TurningAngle()
+		// We break ties deterministically in order to avoid having the output
+		// depend on the input order of the loops.
+		if angle < bestAngle || (angle == bestAngle && compareLoops(p.Loop(i), p.Loop(best)) < 0) {
+			best = i
+			bestAngle = angle
+		}
+	}
+	// Build the new loops vector, starting with the inverted loop.
+	p.Loop(best).Invert()
+	newLoops := make([]*Loop, 0, p.NumLoops())
+	// Add the former siblings of this loop as descendants.
+	lastBest := p.LastDescendant(best)
+	newLoops = append(newLoops, p.Loop(best))
+	for i, l := range p.Loops() {
+		if i < best || i > lastBest {
+			l.depth++
+			newLoops = append(newLoops, l)
+		}
+	}
+	// Add the former children of this loop as siblings.
+	for i, l := range p.Loops() {
+		if i > best && i <= lastBest {
+			l.depth--
+			newLoops = append(newLoops, l)
+		}
+	}
+	p.loops = newLoops
+	p.initLoopProperties()
+}
+
+// Defines a total ordering on Loops that does not depend on the cyclic
+// order of loop vertices. This function is used to choose which loop to
+// invert in the case where several loops have exactly the same area.
+func compareLoops(a, b *Loop) int {
+	if na, nb := a.NumVertices(), b.NumVertices(); na != nb {
+		return na - nb
+	}
+	ai, aDir := a.CanonicalFirstVertex()
+	bi, bDir := b.CanonicalFirstVertex()
+	if aDir != bDir {
+		return aDir - bDir
+	}
+	for n := a.NumVertices() - 1; n >= 0; n, ai, bi = n-1, ai+aDir, bi+bDir {
+		if cmp := a.Vertex(ai).Cmp(b.Vertex(bi).Vector); cmp != 0 {
+			return cmp
+		}
+	}
+	return 0
+}
+
+// PolygonFromCell returns a Polygon from a single loop created from the given Cell.
+func PolygonFromCell(cell Cell) *Polygon {
+	return PolygonFromLoops([]*Loop{LoopFromCell(cell)})
+}
+
+// initNested takes the set of loops in this polygon and performs the nesting
+// computations to set the proper nesting and parent/child relationships.
+func (p *Polygon) initNested() {
+	if len(p.loops) == 1 {
+		p.initOneLoop()
+		return
+	}
+
+	lm := make(loopMap)
+
+	for _, l := range p.loops {
+		lm.insertLoop(l, nil)
+	}
+	// The loops have all been added to the loopMap for ordering. Clear the
+	// loops slice because we add all the loops in-order in initLoops.
+	p.loops = nil
+
+	// Reorder the loops in depth-first traversal order.
+	p.initLoops(lm)
+	p.initLoopProperties()
+}
+
+// loopMap is a map of a loop to its immediate children with respect to nesting.
+// It is used to determine which loops are shells and which are holes.
+type loopMap map[*Loop][]*Loop
+
+// insertLoop adds the given loop to the loop map under the specified parent.
+// All children of the new entry are checked to see if the need to move up to
+// a different level.
+func (lm loopMap) insertLoop(newLoop, parent *Loop) {
+	var children []*Loop
+	for done := false; !done; {
+		children = lm[parent]
+		done = true
+		for _, child := range children {
+			if child.ContainsNested(newLoop) {
+				parent = child
+				done = false
+				break
+			}
+		}
+	}
+
+	// Now, we have found a parent for this loop, it may be that some of the
+	// children of the parent of this loop may now be children of the new loop.
+	newChildren := lm[newLoop]
+	for i := 0; i < len(children); {
+		child := children[i]
+		if newLoop.ContainsNested(child) {
+			newChildren = append(newChildren, child)
+			children = append(children[0:i], children[i+1:]...)
+		} else {
+			i++
+		}
+	}
+
+	lm[newLoop] = newChildren
+	lm[parent] = append(children, newLoop)
+}
+
+// loopStack simplifies access to the loops while being initialized.
+type loopStack []*Loop
+
+func (s *loopStack) push(v *Loop) {
+	*s = append(*s, v)
+}
+func (s *loopStack) pop() *Loop {
+	l := len(*s)
+	r := (*s)[l-1]
+	*s = (*s)[:l-1]
+	return r
+}
+
+// initLoops walks the mapping of loops to all of their children, and adds them in
+// order into to the polygons set of loops.
+func (p *Polygon) initLoops(lm loopMap) {
+	var stack loopStack
+	stack.push(nil)
+	depth := -1
+
+	for len(stack) > 0 {
+		loop := stack.pop()
+		if loop != nil {
+			depth = loop.depth
+			p.loops = append(p.loops, loop)
+		}
+		children := lm[loop]
+		for i := len(children) - 1; i >= 0; i-- {
+			child := children[i]
+			child.depth = depth + 1
+			stack.push(child)
+		}
+	}
+}
+
+// initOneLoop set the properties for a polygon made of a single loop.
+// TODO(roberts): Can this be merged with initLoopProperties
+func (p *Polygon) initOneLoop() {
+	p.hasHoles = false
+	p.numVertices = len(p.loops[0].vertices)
+	p.bound = p.loops[0].RectBound()
+	p.subregionBound = ExpandForSubregions(p.bound)
+	// Ensure the loops depth is set correctly.
+	p.loops[0].depth = 0
+
+	p.initEdgesAndIndex()
+}
+
+// initLoopProperties sets the properties for polygons with multiple loops.
+func (p *Polygon) initLoopProperties() {
+	// the loops depths are set by initNested/initOriented prior to this.
+	p.bound = EmptyRect()
+	p.hasHoles = false
+	for _, l := range p.loops {
+		if l.IsHole() {
+			p.hasHoles = true
+		} else {
+			p.bound = p.bound.Union(l.RectBound())
+		}
+		p.numVertices += l.NumVertices()
+	}
+	p.subregionBound = ExpandForSubregions(p.bound)
+
+	p.initEdgesAndIndex()
+}
+
+// initEdgesAndIndex performs the shape related initializations and adds the final
+// polygon to the index.
+func (p *Polygon) initEdgesAndIndex() {
+	if p.IsFull() {
+		return
+	}
+	const maxLinearSearchLoops = 12 // Based on benchmarks.
+	if len(p.loops) > maxLinearSearchLoops {
+		p.cumulativeEdges = make([]int, 0, len(p.loops))
+	}
+
+	for _, l := range p.loops {
+		if p.cumulativeEdges != nil {
+			p.cumulativeEdges = append(p.cumulativeEdges, p.numEdges)
+		}
+		p.numEdges += len(l.vertices)
+	}
+
+	p.index = NewShapeIndex()
+	p.index.Add(p)
+}
+
+// FullPolygon returns a special "full" polygon.
+func FullPolygon() *Polygon {
+	ret := &Polygon{
+		loops: []*Loop{
+			FullLoop(),
+		},
+		numVertices:    len(FullLoop().Vertices()),
+		bound:          FullRect(),
+		subregionBound: FullRect(),
+	}
+	ret.initEdgesAndIndex()
+	return ret
+}
+
+// Validate checks whether this is a valid polygon,
+// including checking whether all the loops are themselves valid.
+func (p *Polygon) Validate() error {
+	for i, l := range p.loops {
+		// Check for loop errors that don't require building a ShapeIndex.
+		if err := l.findValidationErrorNoIndex(); err != nil {
+			return fmt.Errorf("loop %d: %v", i, err)
+		}
+		// Check that no loop is empty, and that the full loop only appears in the
+		// full polygon.
+		if l.IsEmpty() {
+			return fmt.Errorf("loop %d: empty loops are not allowed", i)
+		}
+		if l.IsFull() && len(p.loops) > 1 {
+			return fmt.Errorf("loop %d: full loop appears in non-full polygon", i)
+		}
+	}
+
+	// TODO(roberts): Uncomment the remaining checks when they are completed.
+
+	// Check for loop self-intersections and loop pairs that cross
+	// (including duplicate edges and vertices).
+	// if findSelfIntersection(p.index) {
+	//	return fmt.Errorf("polygon has loop pairs that cross")
+	// }
+
+	// Check whether initOriented detected inconsistent loop orientations.
+	// if p.hasInconsistentLoopOrientations {
+	// 	return fmt.Errorf("inconsistent loop orientations detected")
+	// }
+
+	// Finally, verify the loop nesting hierarchy.
+	return p.findLoopNestingError()
+}
+
+// findLoopNestingError reports if there is an error in the loop nesting hierarchy.
+func (p *Polygon) findLoopNestingError() error {
+	// First check that the loop depths make sense.
+	lastDepth := -1
+	for i, l := range p.loops {
+		depth := l.depth
+		if depth < 0 || depth > lastDepth+1 {
+			return fmt.Errorf("loop %d: invalid loop depth (%d)", i, depth)
+		}
+		lastDepth = depth
+	}
+	// Then check that they correspond to the actual loop nesting.  This test
+	// is quadratic in the number of loops but the cost per iteration is small.
+	for i, l := range p.loops {
+		last := p.LastDescendant(i)
+		for j, l2 := range p.loops {
+			if i == j {
+				continue
+			}
+			nested := (j >= i+1) && (j <= last)
+			const reverseB = false
+
+			if l.containsNonCrossingBoundary(l2, reverseB) != nested {
+				nestedStr := ""
+				if !nested {
+					nestedStr = "not "
+				}
+				return fmt.Errorf("invalid nesting: loop %d should %scontain loop %d", i, nestedStr, j)
+			}
+		}
+	}
+	return nil
+}
+
+// IsEmpty reports whether this is the special "empty" polygon (consisting of no loops).
+func (p *Polygon) IsEmpty() bool {
+	return len(p.loops) == 0
+}
+
+// IsFull reports whether this is the special "full" polygon (consisting of a
+// single loop that encompasses the entire sphere).
+func (p *Polygon) IsFull() bool {
+	return len(p.loops) == 1 && p.loops[0].IsFull()
+}
+
+// NumLoops returns the number of loops in this polygon.
+func (p *Polygon) NumLoops() int {
+	return len(p.loops)
+}
+
+// Loops returns the loops in this polygon.
+func (p *Polygon) Loops() []*Loop {
+	return p.loops
+}
+
+// Loop returns the loop at the given index. Note that during initialization,
+// the given loops are reordered according to a pre-order traversal of the loop
+// nesting hierarchy. This implies that every loop is immediately followed by
+// its descendants. This hierarchy can be traversed using the methods Parent,
+// LastDescendant, and Loop.depth.
+func (p *Polygon) Loop(k int) *Loop {
+	return p.loops[k]
+}
+
+// Parent returns the index of the parent of loop k.
+// If the loop does not have a parent, ok=false is returned.
+func (p *Polygon) Parent(k int) (index int, ok bool) {
+	// See where we are on the depth hierarchy.
+	depth := p.loops[k].depth
+	if depth == 0 {
+		return -1, false
+	}
+
+	// There may be several loops at the same nesting level as us that share a
+	// parent loop with us. (Imagine a slice of swiss cheese, of which we are one loop.
+	// we don't know how many may be next to us before we get back to our parent loop.)
+	// Move up one position from us, and then begin traversing back through the set of loops
+	// until we find the one that is our parent or we get to the top of the polygon.
+	for k--; k >= 0 && p.loops[k].depth <= depth; k-- {
+	}
+	return k, true
+}
+
+// LastDescendant returns the index of the last loop that is contained within loop k.
+// If k is negative, it returns the last loop in the polygon.
+// Note that loops are indexed according to a pre-order traversal of the nesting
+// hierarchy, so the immediate children of loop k can be found by iterating over
+// the loops (k+1)..LastDescendant(k) and selecting those whose depth is equal
+// to Loop(k).depth+1.
+func (p *Polygon) LastDescendant(k int) int {
+	if k < 0 {
+		return len(p.loops) - 1
+	}
+
+	depth := p.loops[k].depth
+
+	// Find the next loop immediately past us in the set of loops, and then start
+	// moving down the list until we either get to the end or find the next loop
+	// that is higher up the hierarchy than we are.
+	for k++; k < len(p.loops) && p.loops[k].depth > depth; k++ {
+	}
+	return k - 1
+}
+
+// CapBound returns a bounding spherical cap.
+func (p *Polygon) CapBound() Cap { return p.bound.CapBound() }
+
+// RectBound returns a bounding latitude-longitude rectangle.
+func (p *Polygon) RectBound() Rect { return p.bound }
+
+// ContainsPoint reports whether the polygon contains the point.
+func (p *Polygon) ContainsPoint(point Point) bool {
+	// NOTE: A bounds check slows down this function by about 50%. It is
+	// worthwhile only when it might allow us to delay building the index.
+	if !p.index.IsFresh() && !p.bound.ContainsPoint(point) {
+		return false
+	}
+
+	// For small polygons, and during initial construction, it is faster to just
+	// check all the crossing.
+	const maxBruteForceVertices = 32
+	if p.numVertices < maxBruteForceVertices || p.index == nil {
+		inside := false
+		for _, l := range p.loops {
+			// use loops bruteforce to avoid building the index on each loop.
+			inside = inside != l.bruteForceContainsPoint(point)
+		}
+		return inside
+	}
+
+	// Otherwise, look up the ShapeIndex cell containing this point.
+	it := p.index.Iterator()
+	if !it.LocatePoint(point) {
+		return false
+	}
+
+	return p.iteratorContainsPoint(it, point)
+}
+
+// ContainsCell reports whether the polygon contains the given cell.
+func (p *Polygon) ContainsCell(cell Cell) bool {
+	it := p.index.Iterator()
+	relation := it.LocateCellID(cell.ID())
+
+	// If "cell" is disjoint from all index cells, it is not contained.
+	// Similarly, if "cell" is subdivided into one or more index cells then it
+	// is not contained, since index cells are subdivided only if they (nearly)
+	// intersect a sufficient number of edges.  (But note that if "cell" itself
+	// is an index cell then it may be contained, since it could be a cell with
+	// no edges in the loop interior.)
+	if relation != Indexed {
+		return false
+	}
+
+	// Otherwise check if any edges intersect "cell".
+	if p.boundaryApproxIntersects(it, cell) {
+		return false
+	}
+
+	// Otherwise check if the loop contains the center of "cell".
+	return p.iteratorContainsPoint(it, cell.Center())
+}
+
+// IntersectsCell reports whether the polygon intersects the given cell.
+func (p *Polygon) IntersectsCell(cell Cell) bool {
+	it := p.index.Iterator()
+	relation := it.LocateCellID(cell.ID())
+
+	// If cell does not overlap any index cell, there is no intersection.
+	if relation == Disjoint {
+		return false
+	}
+	// If cell is subdivided into one or more index cells, there is an
+	// intersection to within the S2ShapeIndex error bound (see Contains).
+	if relation == Subdivided {
+		return true
+	}
+	// If cell is an index cell, there is an intersection because index cells
+	// are created only if they have at least one edge or they are entirely
+	// contained by the loop.
+	if it.CellID() == cell.id {
+		return true
+	}
+	// Otherwise check if any edges intersect cell.
+	if p.boundaryApproxIntersects(it, cell) {
+		return true
+	}
+	// Otherwise check if the loop contains the center of cell.
+	return p.iteratorContainsPoint(it, cell.Center())
+}
+
+// CellUnionBound computes a covering of the Polygon.
+func (p *Polygon) CellUnionBound() []CellID {
+	// TODO(roberts): Use ShapeIndexRegion when it's available.
+	return p.CapBound().CellUnionBound()
+}
+
+// boundaryApproxIntersects reports whether the loop's boundary intersects cell.
+// It may also return true when the loop boundary does not intersect cell but
+// some edge comes within the worst-case error tolerance.
+//
+// This requires that it.Locate(cell) returned Indexed.
+func (p *Polygon) boundaryApproxIntersects(it *ShapeIndexIterator, cell Cell) bool {
+	aClipped := it.IndexCell().findByShapeID(0)
+
+	// If there are no edges, there is no intersection.
+	if len(aClipped.edges) == 0 {
+		return false
+	}
+
+	// We can save some work if cell is the index cell itself.
+	if it.CellID() == cell.ID() {
+		return true
+	}
+
+	// Otherwise check whether any of the edges intersect cell.
+	maxError := (faceClipErrorUVCoord + intersectsRectErrorUVDist)
+	bound := cell.BoundUV().ExpandedByMargin(maxError)
+	for _, e := range aClipped.edges {
+		edge := p.index.Shape(0).Edge(e)
+		v0, v1, ok := ClipToPaddedFace(edge.V0, edge.V1, cell.Face(), maxError)
+		if ok && edgeIntersectsRect(v0, v1, bound) {
+			return true
+		}
+	}
+
+	return false
+}
+
+// iteratorContainsPoint reports whether the iterator that is positioned at the
+// ShapeIndexCell that may contain p, contains the point p.
+func (p *Polygon) iteratorContainsPoint(it *ShapeIndexIterator, point Point) bool {
+	// Test containment by drawing a line segment from the cell center to the
+	// given point and counting edge crossings.
+	aClipped := it.IndexCell().findByShapeID(0)
+	inside := aClipped.containsCenter
+
+	if len(aClipped.edges) == 0 {
+		return inside
+	}
+
+	// This block requires ShapeIndex.
+	crosser := NewEdgeCrosser(it.Center(), point)
+	shape := p.index.Shape(0)
+	for _, e := range aClipped.edges {
+		edge := shape.Edge(e)
+		inside = inside != crosser.EdgeOrVertexCrossing(edge.V0, edge.V1)
+	}
+
+	return inside
+}
+
+// Shape Interface
+
+// NumEdges returns the number of edges in this shape.
+func (p *Polygon) NumEdges() int {
+	return p.numEdges
+}
+
+// Edge returns endpoints for the given edge index.
+func (p *Polygon) Edge(e int) Edge {
+	var i int
+
+	if len(p.cumulativeEdges) > 0 {
+		for i = range p.cumulativeEdges {
+			if i+1 >= len(p.cumulativeEdges) || e < p.cumulativeEdges[i+1] {
+				e -= p.cumulativeEdges[i]
+				break
+			}
+		}
+	} else {
+		// When the number of loops is small, use linear search. Most often
+		// there is exactly one loop and the code below executes zero times.
+		for i = 0; e >= len(p.Loop(i).vertices); i++ {
+			e -= len(p.Loop(i).vertices)
+		}
+	}
+
+	return Edge{p.Loop(i).OrientedVertex(e), p.Loop(i).OrientedVertex(e + 1)}
+}
+
+// ReferencePoint returns the reference point for this polygon.
+func (p *Polygon) ReferencePoint() ReferencePoint {
+	containsOrigin := false
+	for _, l := range p.loops {
+		containsOrigin = containsOrigin != l.ContainsOrigin()
+	}
+	return OriginReferencePoint(containsOrigin)
+}
+
+// NumChains reports the number of contiguous edge chains in the Polygon.
+func (p *Polygon) NumChains() int {
+	return p.NumLoops()
+}
+
+// Chain returns the i-th edge Chain (loop) in the Shape.
+func (p *Polygon) Chain(chainID int) Chain {
+	if p.cumulativeEdges != nil {
+		return Chain{p.cumulativeEdges[chainID], len(p.Loop(chainID).vertices)}
+	}
+	e := 0
+	for j := 0; j < chainID; j++ {
+		e += len(p.Loop(j).vertices)
+	}
+
+	// Polygon represents a full loop as a loop with one vertex, while
+	// Shape represents a full loop as a chain with no vertices.
+	if numVertices := p.Loop(chainID).NumVertices(); numVertices != 1 {
+		return Chain{e, numVertices}
+	}
+	return Chain{e, 0}
+}
+
+// ChainEdge returns the j-th edge of the i-th edge Chain (loop).
+func (p *Polygon) ChainEdge(i, j int) Edge {
+	return Edge{p.Loop(i).OrientedVertex(j), p.Loop(i).OrientedVertex(j + 1)}
+}
+
+// ChainPosition returns a pair (i, j) such that edgeID is the j-th edge
+// of the i-th edge Chain.
+func (p *Polygon) ChainPosition(edgeID int) ChainPosition {
+	var i int
+
+	if len(p.cumulativeEdges) > 0 {
+		for i = range p.cumulativeEdges {
+			if i+1 >= len(p.cumulativeEdges) || edgeID < p.cumulativeEdges[i+1] {
+				edgeID -= p.cumulativeEdges[i]
+				break
+			}
+		}
+	} else {
+		// When the number of loops is small, use linear search. Most often
+		// there is exactly one loop and the code below executes zero times.
+		for i = 0; edgeID >= len(p.Loop(i).vertices); i++ {
+			edgeID -= len(p.Loop(i).vertices)
+		}
+	}
+	// TODO(roberts): unify this and Edge since they are mostly identical.
+	return ChainPosition{i, edgeID}
+}
+
+// Dimension returns the dimension of the geometry represented by this Polygon.
+func (p *Polygon) Dimension() int { return 2 }
+
+func (p *Polygon) typeTag() typeTag { return typeTagPolygon }
+
+func (p *Polygon) privateInterface() {}
+
+// Contains reports whether this polygon contains the other polygon.
+// Specifically, it reports whether all the points in the other polygon
+// are also in this polygon.
+func (p *Polygon) Contains(o *Polygon) bool {
+	// If both polygons have one loop, use the more efficient Loop method.
+	// Note that Loop's Contains does its own bounding rectangle check.
+	if len(p.loops) == 1 && len(o.loops) == 1 {
+		return p.loops[0].Contains(o.loops[0])
+	}
+
+	// Otherwise if neither polygon has holes, we can still use the more
+	// efficient Loop's Contains method (rather than compareBoundary),
+	// but it's worthwhile to do our own bounds check first.
+	if !p.subregionBound.Contains(o.bound) {
+		// Even though Bound(A) does not contain Bound(B), it is still possible
+		// that A contains B. This can only happen when union of the two bounds
+		// spans all longitudes. For example, suppose that B consists of two
+		// shells with a longitude gap between them, while A consists of one shell
+		// that surrounds both shells of B but goes the other way around the
+		// sphere (so that it does not intersect the longitude gap).
+		if !p.bound.Lng.Union(o.bound.Lng).IsFull() {
+			return false
+		}
+	}
+
+	if !p.hasHoles && !o.hasHoles {
+		for _, l := range o.loops {
+			if !p.anyLoopContains(l) {
+				return false
+			}
+		}
+		return true
+	}
+
+	// Polygon A contains B iff B does not intersect the complement of A. From
+	// the intersection algorithm below, this means that the complement of A
+	// must exclude the entire boundary of B, and B must exclude all shell
+	// boundaries of the complement of A. (It can be shown that B must then
+	// exclude the entire boundary of the complement of A.) The first call
+	// below returns false if the boundaries cross, therefore the second call
+	// does not need to check for any crossing edges (which makes it cheaper).
+	return p.containsBoundary(o) && o.excludesNonCrossingComplementShells(p)
+}
+
+// Intersects reports whether this polygon intersects the other polygon, i.e.
+// if there is a point that is contained by both polygons.
+func (p *Polygon) Intersects(o *Polygon) bool {
+	// If both polygons have one loop, use the more efficient Loop method.
+	// Note that Loop Intersects does its own bounding rectangle check.
+	if len(p.loops) == 1 && len(o.loops) == 1 {
+		return p.loops[0].Intersects(o.loops[0])
+	}
+
+	// Otherwise if neither polygon has holes, we can still use the more
+	// efficient Loop.Intersects method. The polygons intersect if and
+	// only if some pair of loop regions intersect.
+	if !p.bound.Intersects(o.bound) {
+		return false
+	}
+
+	if !p.hasHoles && !o.hasHoles {
+		for _, l := range o.loops {
+			if p.anyLoopIntersects(l) {
+				return true
+			}
+		}
+		return false
+	}
+
+	// Polygon A is disjoint from B if A excludes the entire boundary of B and B
+	// excludes all shell boundaries of A. (It can be shown that B must then
+	// exclude the entire boundary of A.) The first call below returns false if
+	// the boundaries cross, therefore the second call does not need to check
+	// for crossing edges.
+	return !p.excludesBoundary(o) || !o.excludesNonCrossingShells(p)
+}
+
+// compareBoundary returns +1 if this polygon contains the boundary of B, -1 if A
+// excludes the boundary of B, and 0 if the boundaries of A and B cross.
+func (p *Polygon) compareBoundary(o *Loop) int {
+	result := -1
+	for i := 0; i < len(p.loops) && result != 0; i++ {
+		// If B crosses any loop of A, the result is 0. Otherwise the result
+		// changes sign each time B is contained by a loop of A.
+		result *= -p.loops[i].compareBoundary(o)
+	}
+	return result
+}
+
+// containsBoundary reports whether this polygon contains the entire boundary of B.
+func (p *Polygon) containsBoundary(o *Polygon) bool {
+	for _, l := range o.loops {
+		if p.compareBoundary(l) <= 0 {
+			return false
+		}
+	}
+	return true
+}
+
+// excludesBoundary reports whether this polygon excludes the entire boundary of B.
+func (p *Polygon) excludesBoundary(o *Polygon) bool {
+	for _, l := range o.loops {
+		if p.compareBoundary(l) >= 0 {
+			return false
+		}
+	}
+	return true
+}
+
+// containsNonCrossingBoundary reports whether polygon A contains the boundary of
+// loop B. Shared edges are handled according to the rule described in loops
+// containsNonCrossingBoundary.
+func (p *Polygon) containsNonCrossingBoundary(o *Loop, reverse bool) bool {
+	var inside bool
+	for _, l := range p.loops {
+		x := l.containsNonCrossingBoundary(o, reverse)
+		inside = (inside != x)
+	}
+	return inside
+}
+
+// excludesNonCrossingShells reports wheterh given two polygons A and B such that the
+// boundary of A does not cross any loop of B, if A excludes all shell boundaries of B.
+func (p *Polygon) excludesNonCrossingShells(o *Polygon) bool {
+	for _, l := range o.loops {
+		if l.IsHole() {
+			continue
+		}
+		if p.containsNonCrossingBoundary(l, false) {
+			return false
+		}
+	}
+	return true
+}
+
+// excludesNonCrossingComplementShells reports whether given two polygons A and B
+// such that the boundary of A does not cross any loop of B, if A excludes all
+// shell boundaries of the complement of B.
+func (p *Polygon) excludesNonCrossingComplementShells(o *Polygon) bool {
+	// Special case to handle the complement of the empty or full polygons.
+	if o.IsEmpty() {
+		return !p.IsFull()
+	}
+	if o.IsFull() {
+		return true
+	}
+
+	// Otherwise the complement of B may be obtained by inverting loop(0) and
+	// then swapping the shell/hole status of all other loops. This implies
+	// that the shells of the complement consist of loop 0 plus all the holes of
+	// the original polygon.
+	for j, l := range o.loops {
+		if j > 0 && !l.IsHole() {
+			continue
+		}
+
+		// The interior of the complement is to the right of loop 0, and to the
+		// left of the loops that were originally holes.
+		if p.containsNonCrossingBoundary(l, j == 0) {
+			return false
+		}
+	}
+	return true
+}
+
+// anyLoopContains reports whether any loop in this polygon contains the given loop.
+func (p *Polygon) anyLoopContains(o *Loop) bool {
+	for _, l := range p.loops {
+		if l.Contains(o) {
+			return true
+		}
+	}
+	return false
+}
+
+// anyLoopIntersects reports whether any loop in this polygon intersects the given loop.
+func (p *Polygon) anyLoopIntersects(o *Loop) bool {
+	for _, l := range p.loops {
+		if l.Intersects(o) {
+			return true
+		}
+	}
+	return false
+}
+
+// Area returns the area of the polygon interior, i.e. the region on the left side
+// of an odd number of loops. The return value is between 0 and 4*Pi.
+func (p *Polygon) Area() float64 {
+	var area float64
+	for _, loop := range p.loops {
+		area += float64(loop.Sign()) * loop.Area()
+	}
+	return area
+}
+
+// Encode encodes the Polygon
+func (p *Polygon) Encode(w io.Writer) error {
+	e := &encoder{w: w}
+	p.encode(e)
+	return e.err
+}
+
+// encode only supports lossless encoding and not compressed format.
+func (p *Polygon) encode(e *encoder) {
+	if p.numVertices == 0 {
+		p.encodeCompressed(e, maxLevel, nil)
+		return
+	}
+
+	// Convert all the polygon vertices to XYZFaceSiTi format.
+	vs := make([]xyzFaceSiTi, 0, p.numVertices)
+	for _, l := range p.loops {
+		vs = append(vs, l.xyzFaceSiTiVertices()...)
+	}
+
+	// Computes a histogram of the cell levels at which the vertices are snapped.
+	// (histogram[0] is the number of unsnapped vertices, histogram[i] the number
+	// of vertices snapped at level i-1).
+	histogram := make([]int, maxLevel+2)
+	for _, v := range vs {
+		histogram[v.level+1]++
+	}
+
+	// Compute the level at which most of the vertices are snapped.
+	// If multiple levels have the same maximum number of vertices
+	// snapped to it, the first one (lowest level number / largest
+	// area / smallest encoding length) will be chosen, so this
+	// is desired.
+	var snapLevel, numSnapped int
+	for level, h := range histogram[1:] {
+		if h > numSnapped {
+			snapLevel, numSnapped = level, h
+		}
+	}
+
+	// Choose an encoding format based on the number of unsnapped vertices and a
+	// rough estimate of the encoded sizes.
+	numUnsnapped := p.numVertices - numSnapped // Number of vertices that won't be snapped at snapLevel.
+	const pointSize = 3 * 8                    // s2.Point is an r3.Vector, which is 3 float64s. That's 3*8 = 24 bytes.
+	compressedSize := 4*p.numVertices + (pointSize+2)*numUnsnapped
+	losslessSize := pointSize * p.numVertices
+	if compressedSize < losslessSize {
+		p.encodeCompressed(e, snapLevel, vs)
+	} else {
+		p.encodeLossless(e)
+	}
+}
+
+// encodeLossless encodes the polygon's Points as float64s.
+func (p *Polygon) encodeLossless(e *encoder) {
+	e.writeInt8(encodingVersion)
+	e.writeBool(true) // a legacy c++ value. must be true.
+	e.writeBool(p.hasHoles)
+	e.writeUint32(uint32(len(p.loops)))
+
+	if e.err != nil {
+		return
+	}
+	if len(p.loops) > maxEncodedLoops {
+		e.err = fmt.Errorf("too many loops (%d; max is %d)", len(p.loops), maxEncodedLoops)
+		return
+	}
+	for _, l := range p.loops {
+		l.encode(e)
+	}
+
+	// Encode the bound.
+	p.bound.encode(e)
+}
+
+func (p *Polygon) encodeCompressed(e *encoder, snapLevel int, vertices []xyzFaceSiTi) {
+	e.writeUint8(uint8(encodingCompressedVersion))
+	e.writeUint8(uint8(snapLevel))
+	e.writeUvarint(uint64(len(p.loops)))
+
+	if e.err != nil {
+		return
+	}
+	if l := len(p.loops); l > maxEncodedLoops {
+		e.err = fmt.Errorf("too many loops to encode: %d; max is %d", l, maxEncodedLoops)
+		return
+	}
+
+	for _, l := range p.loops {
+		l.encodeCompressed(e, snapLevel, vertices[:len(l.vertices)])
+		vertices = vertices[len(l.vertices):]
+	}
+	// Do not write the bound, num_vertices, or has_holes_ as they can be
+	// cheaply recomputed by decodeCompressed.  Microbenchmarks show the
+	// speed difference is inconsequential.
+}
+
+// Decode decodes the Polygon.
+func (p *Polygon) Decode(r io.Reader) error {
+	d := &decoder{r: asByteReader(r)}
+	version := int8(d.readUint8())
+	var dec func(*decoder)
+	switch version {
+	case encodingVersion:
+		dec = p.decode
+	case encodingCompressedVersion:
+		dec = p.decodeCompressed
+	default:
+		return fmt.Errorf("unsupported version %d", version)
+	}
+	dec(d)
+	return d.err
+}
+
+// maxEncodedLoops is the biggest supported number of loops in a polygon during encoding.
+// Setting a maximum guards an allocation: it prevents an attacker from easily pushing us OOM.
+const maxEncodedLoops = 10000000
+
+func (p *Polygon) decode(d *decoder) {
+	*p = Polygon{}
+	d.readUint8() // Ignore irrelevant serialized owns_loops_ value.
+
+	p.hasHoles = d.readBool()
+
+	// Polygons with no loops are explicitly allowed here: a newly created
+	// polygon has zero loops and such polygons encode and decode properly.
+	nloops := d.readUint32()
+	if d.err != nil {
+		return
+	}
+	if nloops > maxEncodedLoops {
+		d.err = fmt.Errorf("too many loops (%d; max is %d)", nloops, maxEncodedLoops)
+		return
+	}
+	p.loops = make([]*Loop, nloops)
+	for i := range p.loops {
+		p.loops[i] = new(Loop)
+		p.loops[i].decode(d)
+		p.numVertices += len(p.loops[i].vertices)
+	}
+
+	p.bound.decode(d)
+	if d.err != nil {
+		return
+	}
+	p.subregionBound = ExpandForSubregions(p.bound)
+	p.initEdgesAndIndex()
+}
+
+func (p *Polygon) decodeCompressed(d *decoder) {
+	snapLevel := int(d.readUint8())
+
+	if snapLevel > maxLevel {
+		d.err = fmt.Errorf("snaplevel too big: %d", snapLevel)
+		return
+	}
+	// Polygons with no loops are explicitly allowed here: a newly created
+	// polygon has zero loops and such polygons encode and decode properly.
+	nloops := int(d.readUvarint())
+	if nloops > maxEncodedLoops {
+		d.err = fmt.Errorf("too many loops (%d; max is %d)", nloops, maxEncodedLoops)
+	}
+	p.loops = make([]*Loop, nloops)
+	for i := range p.loops {
+		p.loops[i] = new(Loop)
+		p.loops[i].decodeCompressed(d, snapLevel)
+	}
+	p.initLoopProperties()
+}
+
+// TODO(roberts): Differences from C++
+// Centroid
+// SnapLevel
+// DistanceToPoint
+// DistanceToBoundary
+// Project
+// ProjectToBoundary
+// ApproxContains/ApproxDisjoint for Polygons
+// InitTo{Intersection/ApproxIntersection/Union/ApproxUnion/Diff/ApproxDiff}
+// InitToSimplified
+// InitToSnapped
+// IntersectWithPolyline
+// ApproxIntersectWithPolyline
+// SubtractFromPolyline
+// ApproxSubtractFromPolyline
+// DestructiveUnion
+// DestructiveApproxUnion
+// InitToCellUnionBorder
+// IsNormalized
+// Equal/BoundaryEqual/BoundaryApproxEqual/BoundaryNear Polygons
+// BreakEdgesAndAddToBuilder
+//
+// clearLoops
+// findLoopNestingError
+// initToSimplifiedInternal
+// internalClipPolyline
+// clipBoundary