rond-with-you/internal/fog/render_diff.go

package fog

import (
	"image"
	"math"
)

// 差异图的三种归属。改这里的色值要同步 templates/admin/fog-preview.html 的图例。
var (
	DiffBothColor = [4]byte{126, 134, 146, 148} // 两边都有：导入不影响
	DiffGainColor = [4]byte{36, 164, 90, 205}   // 只有快照有：叠加会新增
	DiffLostColor = [4]byte{214, 56, 56, 210}   // 只有库里有：整体替换会丢
)

const (
	diffNone uint8 = iota
	diffBoth
	diffGain
	diffLost
)

// RenderDiffTile 渲染一张 256x256 的差异瓦片，供站长在真实底图上核对
// 「这份快照能补上哪里」与「整体替换会丢掉哪里」。
//
// 与 RenderTile 一样按底图坐标系做 WGS84→GCJ02 纠偏；不做描边——三种颜色
// 已经够区分，再描边在粗缩放级下会糊成一片。
//
// 逐格判定只在放大到「一格不小于一像素」（z>=14）时做；更粗的级别下一个块还不到
// 一个像素，按块内占多数的类别整块涂色。绘制顺序 both→gain→lost，让「会新增」
// 和「会丢」压住底色，重叠处不至于被灰吃掉。
func RenderDiffTile(cur, inc map[[2]int][]byte, z, x, y int, crs string) *image.NRGBA {
	if z < 0 || z > 22 {
		return nil
	}
	step := WorldCells >> z
	scale := 256.0 / float64(step)
	ox, oy := float64(x*step), float64(y*step)

	layer := make([]uint8, 256*256)
	mark := func(px, py, w, h int, kind uint8) {
		for r := 0; r < h; r++ {
			yy := py + r
			if yy < 0 || yy > 255 {
				continue
			}
			row := yy * 256
			for c := 0; c < w; c++ {
				xx := px + c
				if xx < 0 || xx > 255 {
					continue
				}
				layer[row+xx] = kind
			}
		}
	}

	// 直接遍历块表而不是全球块范围：低缩放级下范围循环会膨胀到数十亿次空查询。
	// GCJ 偏移最多约 700m（74 格），粗裁剪时两侧各放宽两块保证不漏。
	const margin = 2 * BlockCells
	lo, hi := float64(x*step-margin), float64((x+1)*step+margin)
	rlo, rhi := float64(y*step-margin), float64((y+1)*step+margin)
	shift := crs == "gcj02"

	draw := func(pass uint8, gbx, gby int, cb, ib []byte) {
		bx, by := float64(gbx*BlockCells), float64(gby*BlockCells)
		if bx+BlockCells <= lo || bx >= hi || by+BlockCells <= rlo || by >= rhi {
			return
		}
		ddx, ddy := 0.0, 0.0
		if shift {
			ddx, ddy = blockShift(gbx, gby)
		}
		u0 := (bx + ddx - ox) * scale
		v0 := (by + ddy - oy) * scale

		if scale >= 1 {
			w := int(math.Ceil(scale))
			for row := 0; row < BlockCells; row++ {
				base := row * 8
				for colB := 0; colB < 8; colB++ {
					var cbv, ibv byte
					if cb != nil {
						cbv = cb[base+colB]
					}
					if ib != nil {
						ibv = ib[base+colB]
					}
					if cbv|ibv == 0 {
						continue
					}
					py := int(math.Floor(v0 + float64(row)*scale))
					for bit := 0; bit < 8; bit++ {
						m := byte(0x80 >> bit)
						kind := cellKind(cbv&m != 0, ibv&m != 0)
						if kind != pass {
							continue
						}
						px := int(math.Floor(u0 + float64(colB*8+bit)*scale))
						mark(px, py, w, w, kind)
					}
				}
			}
			return
		}

		// 一格不足一像素：整块按多数类涂，三种混在一起的块以数量最多的为准
		if dominantKind(blockDiffCounts(cb, ib)) != pass {
			return
		}
		u1 := (bx + BlockCells + ddx - ox) * scale
		v1 := (by + BlockCells + ddy - oy) * scale
		px0, py0 := int(math.Floor(u0)), int(math.Floor(v0))
		px1, py1 := int(math.Ceil(u1)), int(math.Ceil(v1))
		if px1-px0 < 1 {
			px1 = px0 + 1
		}
		if py1-py0 < 1 {
			py1 = py0 + 1
		}
		mark(px0, py0, px1-px0, py1-py0, pass)
	}

	for _, pass := range []uint8{diffBoth, diffGain, diffLost} {
		for key, cb := range cur {
			draw(pass, key[0], key[1], cb, inc[key])
		}
		// 只在快照里的块单独走一遍，两边都有的块不再重复画
		for key, ib := range inc {
			if _, dup := cur[key]; dup {
				continue
			}
			draw(pass, key[0], key[1], nil, ib)
		}
	}

	img := image.NewNRGBA(image.Rect(0, 0, 256, 256))
	for i, kind := range layer {
		if kind == diffNone {
			continue
		}
		var c [4]byte
		switch kind {
		case diffBoth:
			c = DiffBothColor
		case diffGain:
			c = DiffGainColor
		default:
			c = DiffLostColor
		}
		o := i * 4
		img.Pix[o], img.Pix[o+1], img.Pix[o+2], img.Pix[o+3] = c[0], c[1], c[2], c[3]
	}
	return img
}

func cellKind(inCur, inInc bool) uint8 {
	switch {
	case inCur && inInc:
		return diffBoth
	case inInc:
		return diffGain
	case inCur:
		return diffLost
	}
	return diffNone
}

// blockDiffCounts 数一个块里三类各占多少格。两个位图都可能缺失（只有一边有这个块）。
func blockDiffCounts(cb, ib []byte) (both, gain, lost int) {
	n := len(cb)
	if len(ib) > n {
		n = len(ib)
	}
	for i := 0; i < n; i++ {
		var c, in byte
		if i < len(cb) {
			c = cb[i]
		}
		if i < len(ib) {
			in = ib[i]
		}
		both += int(popcount[c&in])
		gain += int(popcount[in&^c])
		lost += int(popcount[c&^in])
	}
	return
}

// dominantKind 取三类里占得最多的一类，用来在粗缩放级下决定整块的着色。
func dominantKind(both, gain, lost int) uint8 {
	switch {
	case gain > 0 && gain >= both && gain >= lost:
		return diffGain
	case lost > 0 && lost >= both:
		return diffLost
	case both > 0:
		return diffBoth
	}
	return diffNone
}