优化,阅读问题处理

This commit is contained in:
truewhile
2026-10-10 11:26:47 +08:00
parent 374ee28f01
commit 4ed2edab5b
36 changed files with 3921 additions and 140 deletions
+375
View File
@@ -0,0 +1,375 @@
package rule
import (
"encoding/binary"
"fmt"
"strings"
)
// 本文件移植 legado 的 QueryTTF:解析字体(sfnt)的 cmap / glyf / loca / maxp 表,
// 建立「Unicode 码点 → 字形」「字形 → Unicode 码点」两张表,供
// java.queryTTF / java.replaceFont 还原被字体混淆的正文。
//
// 阅读站点的常见套路是:正文用一套打乱过的字体渲染,页面上给出「错误字体」(把
// 每个码点映射到错字形)与「正确字体」(字形到真实码点的映射)。replaceFont 按
// 字形轮廓把错误字体里的字符逐个换成正确字体里同字形的码点,从而还原原文。
//
// 解析器全程做边界检查:字体字节来自书源(不可信),坏字体只应报错,不能 panic。
// queryTTFFont 一个已解析的字体。
type queryTTFFont struct {
// unicodeToGlyphID 码点 → 字形在 glyf 表里的下标。
unicodeToGlyphID map[rune]uint16
// unicodeToGlyph 码点 → 字形轮廓(用于跨字体比较字形)。
unicodeToGlyph map[rune]string
// glyphToUnicode 字形轮廓 → 码点(正确字体用来查回真实字符)。
glyphToUnicode map[string]rune
}
// sfnt 表标签。
var (
ttfTagCmap = [4]byte{'c', 'm', 'a', 'p'}
ttfTagGlyf = [4]byte{'g', 'l', 'y', 'f'}
ttfTagLoca = [4]byte{'l', 'o', 'c', 'a'}
ttfTagMaxp = [4]byte{'m', 'a', 'x', 'p'}
ttfTagHead = [4]byte{'h', 'e', 'a', 'd'}
)
// parseQueryTTFFont 解析字体字节。支持 sfnt(TTF/OTF)与 ttc 的第一套字体。
func parseQueryTTFFont(data []byte) (font *queryTTFFont, err error) {
defer func() {
if r := recover(); r != nil {
font, err = nil, fmt.Errorf("字体解析失败: %v", r)
}
}()
if len(data) < 12 {
return nil, fmt.Errorf("字体数据过短")
}
// ttc:取第一套字体的偏移。
if string(data[:4]) == "ttcf" {
if len(data) < 16 {
return nil, fmt.Errorf("ttc 头部不完整")
}
off := int(binary.BigEndian.Uint32(data[12:16]))
if off <= 0 || off >= len(data) {
return nil, fmt.Errorf("ttc 字体偏移非法")
}
data = data[off:]
}
numTables := int(binary.BigEndian.Uint16(data[4:6]))
if numTables <= 0 || 12+numTables*16 > len(data) {
return nil, fmt.Errorf("sfnt 表目录非法")
}
tables := map[[4]byte][]byte{}
for i := 0; i < numTables; i++ {
rec := data[12+i*16 : 12+i*16+16]
var tag [4]byte
copy(tag[:], rec[:4])
off := int(binary.BigEndian.Uint32(rec[8:12]))
length := int(binary.BigEndian.Uint32(rec[12:16]))
if off < 0 || length < 0 || off > len(data) {
continue
}
if off+length > len(data) {
length = len(data) - off
}
tables[tag] = data[off : off+length]
}
cmap := tables[ttfTagCmap]
if cmap == nil {
return nil, fmt.Errorf("字体缺少 cmap 表")
}
mapping, err := parseTTFCmap(cmap)
if err != nil {
return nil, err
}
f := &queryTTFFont{
unicodeToGlyphID: mapping,
unicodeToGlyph: map[rune]string{},
glyphToUnicode: map[string]rune{},
}
// 有 glyf + loca + head + maxp 才能算字形轮廓;缺了(如 CFF 字体)只保留码点映射。
head := tables[ttfTagHead]
maxp := tables[ttfTagMaxp]
loca := tables[ttfTagLoca]
glyf := tables[ttfTagGlyf]
if head == nil || maxp == nil || loca == nil || glyf == nil || len(head) < 54 {
return f, nil
}
indexToLocFormat := int16(binary.BigEndian.Uint16(head[50:52]))
numGlyphs := int(binary.BigEndian.Uint16(maxp[4:6]))
offsets, err := parseTTFLoca(loca, numGlyphs, indexToLocFormat)
if err != nil {
return f, nil
}
glyphCache := map[uint16]string{}
for cp, gid := range mapping {
outline := ttfGlyphOutline(glyf, offsets, gid, glyphCache, 0)
f.unicodeToGlyph[cp] = outline
if _, exists := f.glyphToUnicode[outline]; !exists {
f.glyphToUnicode[outline] = cp
}
}
return f, nil
}
// parseTTFCmap 解析 cmap 表,返回码点 → 字形下标。
// 支持 format 0 / 4 / 6(legado QueryTTF 同样只支持这三种)。
func parseTTFCmap(cmap []byte) (map[rune]uint16, error) {
if len(cmap) < 4 {
return nil, fmt.Errorf("cmap 表过短")
}
numTables := int(binary.BigEndian.Uint16(cmap[2:4]))
if 4+numTables*8 > len(cmap) {
return nil, fmt.Errorf("cmap 子表目录非法")
}
out := map[rune]uint16{}
for i := 0; i < numTables; i++ {
rec := cmap[4+i*8 : 4+i*8+8]
off := int(binary.BigEndian.Uint32(rec[4:8]))
if off < 0 || off+2 > len(cmap) {
continue
}
sub := cmap[off:]
switch binary.BigEndian.Uint16(sub[0:2]) {
case 0:
parseCmapFormat0(sub, out)
case 4:
parseCmapFormat4(sub, out)
case 6:
parseCmapFormat6(sub, out)
}
// 优先保留第一个子表解析到的映射;后续子表只补缺失项。
}
if len(out) == 0 {
return nil, fmt.Errorf("cmap 没有可用的 format 0/4/6 子表")
}
return out, nil
}
func parseCmapFormat0(sub []byte, out map[rune]uint16) {
if len(sub) < 262 {
return
}
length := int(binary.BigEndian.Uint16(sub[2:4]))
if length > len(sub) {
length = len(sub)
}
glyphs := sub[6:min(6+256, length)]
for i, gid := range glyphs {
if gid != 0 {
if _, exists := out[rune(i)]; !exists {
out[rune(i)] = uint16(gid)
}
}
}
}
func parseCmapFormat4(sub []byte, out map[rune]uint16) {
if len(sub) < 14 {
return
}
segCountX2 := int(binary.BigEndian.Uint16(sub[6:8]))
segCount := segCountX2 / 2
if segCount == 0 {
return
}
endBase := 14
startBase := endBase + segCountX2 + 2
deltaBase := startBase + segCountX2
rangeBase := deltaBase + segCountX2
if rangeBase+segCountX2 > len(sub) {
return
}
for i := 0; i < segCount; i++ {
end := int(binary.BigEndian.Uint16(sub[endBase+i*2:]))
start := int(binary.BigEndian.Uint16(sub[startBase+i*2:]))
delta := int16(binary.BigEndian.Uint16(sub[deltaBase+i*2:]))
rangeOffset := int(binary.BigEndian.Uint16(sub[rangeBase+i*2:]))
if start > end {
continue
}
for cp := start; cp <= end && cp <= 0xFFFF; cp++ {
if cp == 0xFFFF {
continue
}
var gid uint16
if rangeOffset == 0 {
gid = uint16(int(cp) + int(delta))
} else {
idx := rangeBase + i*2 + rangeOffset + (cp-start)*2
if idx+2 > len(sub) {
continue
}
g := binary.BigEndian.Uint16(sub[idx : idx+2])
if g == 0 {
continue
}
gid = uint16(int(g) + int(delta))
}
if gid != 0 {
if _, exists := out[rune(cp)]; !exists {
out[rune(cp)] = gid
}
}
}
}
}
func parseCmapFormat6(sub []byte, out map[rune]uint16) {
if len(sub) < 10 {
return
}
first := int(binary.BigEndian.Uint16(sub[6:8]))
count := int(binary.BigEndian.Uint16(sub[8:10]))
for i := 0; i < count; i++ {
idx := 10 + i*2
if idx+2 > len(sub) {
return
}
gid := binary.BigEndian.Uint16(sub[idx : idx+2])
if gid == 0 {
continue
}
cp := rune(first + i)
if _, exists := out[cp]; !exists {
out[cp] = gid
}
}
}
// parseTTFLoca 解析 loca 表,返回每个字形的字节区间起止。
func parseTTFLoca(loca []byte, numGlyphs int, indexToLocFormat int16) ([]int, error) {
if indexToLocFormat == 0 {
need := (numGlyphs + 1) * 2
if len(loca) < need {
return nil, fmt.Errorf("loca 表过短")
}
out := make([]int, numGlyphs+1)
for i := 0; i <= numGlyphs; i++ {
out[i] = int(binary.BigEndian.Uint16(loca[i*2:])) * 2
}
return out, nil
}
need := (numGlyphs + 1) * 4
if len(loca) < need {
return nil, fmt.Errorf("loca 表过短")
}
out := make([]int, numGlyphs+1)
for i := 0; i <= numGlyphs; i++ {
out[i] = int(binary.BigEndian.Uint32(loca[i*4:]))
}
return out, nil
}
// ttfGlyphOutline 把字形转成轮廓字符串(对应 legado QueryTTF.Glyf.toString)。
// 复合字形递归展开组件,深度上限防自引用。
func ttfGlyphOutline(glyf []byte, offsets []int, gid uint16, cache map[uint16]string, depth int) string {
if depth > 8 {
return fmt.Sprintf("glyph%d", gid)
}
if v, ok := cache[gid]; ok {
return v
}
if int(gid)+1 >= len(offsets) {
return fmt.Sprintf("glyph%d", gid)
}
start, end := offsets[gid], offsets[gid+1]
if start < 0 || end > len(glyf) || end <= start {
// 空字形(如空格):用下标本身当轮廓,保证不同码点不会互相误判。
out := fmt.Sprintf("glyph%d", gid)
cache[gid] = out
return out
}
numberOfContours := int16(binary.BigEndian.Uint16(glyf[start : start+2]))
if numberOfContours >= 0 {
out := fmt.Sprintf("simple:%d:%s", numberOfContours, ttfSimpleGlyphPoints(glyf[start:end]))
cache[gid] = out
return out
}
// 复合字形:逐组件展开。
var sb strings.Builder
sb.WriteString("composite")
pos := start + 10
for pos+4 <= end {
flags := binary.BigEndian.Uint16(glyf[pos : pos+2])
component := binary.BigEndian.Uint16(glyf[pos+2 : pos+4])
pos += 4
if flags&0x0001 != 0 { // ARG_1_AND_2_ARE_WORDS
pos += 4
} else {
pos += 2
}
switch {
case flags&0x0008 != 0: // WE_HAVE_A_SCALE
pos += 2
case flags&0x0040 != 0: // WE_HAVE_AN_X_AND_Y_SCALE
pos += 4
case flags&0x0080 != 0: // WE_HAVE_A_TWO_BY_TWO
pos += 8
}
sb.WriteString("+")
sb.WriteString(ttfGlyphOutline(glyf, offsets, component, cache, depth+1))
if flags&0x0020 == 0 { // MORE_COMPONENTS
break
}
}
out := sb.String()
cache[gid] = out
return out
}
// ttfSimpleGlyphPoints 取简单字形的轮廓点(标志与坐标的紧凑编码)。
func ttfSimpleGlyphPoints(data []byte) string {
if len(data) < 10 {
return ""
}
numberOfContours := int(binary.BigEndian.Uint16(data[0:2]))
if numberOfContours <= 0 {
return ""
}
endPtsPos := 10
if endPtsPos+numberOfContours*2+2 > len(data) {
return ""
}
numPoints := int(binary.BigEndian.Uint16(data[endPtsPos+(numberOfContours-1)*2:])) + 1
if numPoints <= 0 {
return ""
}
pos := endPtsPos + numberOfContours*2 + 2 // + instructionLength
if pos > len(data) {
return ""
}
instrLen := int(binary.BigEndian.Uint16(data[pos-2 : pos]))
pos += instrLen
flags := make([]byte, 0, numPoints)
for len(flags) < numPoints && pos < len(data) {
flag := data[pos]
pos++
flags = append(flags, flag)
if flag&0x08 != 0 { // REPEAT
if pos >= len(data) {
break
}
repeat := int(data[pos])
pos++
for i := 0; i < repeat && len(flags) < numPoints; i++ {
flags = append(flags, flag)
}
}
}
// 解析 x / y 坐标(与点数等长的增量序列,这里只用于区分字形)。
var sb strings.Builder
sb.WriteString(fmt.Sprintf("n=%d;", numPoints))
for _, flag := range flags {
sb.WriteByte('0' + flag&0x0F)
}
sb.WriteString(";")
// 跳过坐标数据不影响「同名轮廓一致性」的判断:同字形的字体坐标编码一致。
if pos < len(data) {
sb.WriteString(fmt.Sprintf("d=%d", len(data)-pos))
}
return sb.String()
}