feat(渲染编辑): 优化渲染编辑完善资源保存

2026-10-30 01:27:26 +08:00 · 2026-10-05 12:49:51 +08:00
1 parent 5dcd7a4e8d
commit 6c14eef1ee
@@ -10,7 +10,7 @@ require (
 	github.com/tdewolff/canvas v0.0.0-20260923214215-09804640d00c
 	github.com/tdewolff/font v0.0.0-20260913163313-54f98bb59ee6
 	github.com/xiaoqidun/jbig2 v1.1.1
-	github.com/xiaoqidun/pdfgo v0.0.0-20261004151751-bff53238d953
+	github.com/xiaoqidun/pdfgo v0.0.0-20261005044242-0007b05b8143
 	golang.org/x/crypto v0.57.0
 	golang.org/x/image v0.46.0
 	golang.org/x/text v0.42.0
@@ -62,8 +62,8 @@ github.com/tdewolff/test v1.0.12 h1:7F21DqIajswxuche0geHdrUZRCWE4oko4b7bcmkkrxk=
 github.com/tdewolff/test v1.0.12/go.mod h1:XPuWBzvdUzhCuxWO1ojpXsyzsA5bFoS3tO/Q3kFuTG8=
 github.com/xiaoqidun/jbig2 v1.1.1 h1:SjliS9ZO0e9HX5oYUfBuXrfFsXhWricrSz+XY4zORDI=
 github.com/xiaoqidun/jbig2 v1.1.1/go.mod h1:8QonVWtENmh2kc6l+leqeGl7psb79DWlHDkCpdChjfQ=
-github.com/xiaoqidun/pdfgo v0.0.0-20261004151751-bff53238d953 h1:4fz7oPfhMZr8N2fWjv7Gn6+L2TInWGarQm3UGaqV+54=
-github.com/xiaoqidun/pdfgo v0.0.0-20261004151751-bff53238d953/go.mod h1:4Pf0JO0BEf1uzRY+uZjlMjN/OPn3PYa0YMi6cqPPTLQ=
+github.com/xiaoqidun/pdfgo v0.0.0-20261005044242-0007b05b8143 h1:OELHUuQnFJHk9qyHQX2gzdMvSQNbnEyD70YeB5Eoldg=
+github.com/xiaoqidun/pdfgo v0.0.0-20261005044242-0007b05b8143/go.mod h1:4Pf0JO0BEf1uzRY+uZjlMjN/OPn3PYa0YMi6cqPPTLQ=
 github.com/xyproto/randomstring v1.0.5 h1:YtlWPoRdgMu3NZtP45drfy1GKoojuR7hmRcnhZqKjWU=
 github.com/xyproto/randomstring v1.0.5/go.mod h1:rgmS5DeNXLivK7YprL0pY+lTuhNQW3iGxZ18UQApw/E=
 github.com/yuin/goldmark v1.8.5 h1:r6N5afV5qj/5S4UTch8agZHJ8UxNCMwX7WjkkJam2NA=
@@ -32,6 +32,13 @@ import (
 // 入参: parts 输出改动, reader 原包，可为空, removed 已移除条目
 // 返回: error 读取或取消错误
 func (e *Editor) compressResourceParts(parts map[string][]byte, reader *Reader, removed map[string]bool) error {
+	return e.compressResourceReferences(parts, reader, removed, nil)
+}
+
+// compressResourceReferences 优化资源，复用自产页面引用并流式读取暂存页面
+// 入参: parts 输出改动, reader 原包, removed 已移除条目, generated 本次自产页面及引用
+// 返回: error 读取或取消错误
+func (e *Editor) compressResourceReferences(parts map[string][]byte, reader *Reader, removed map[string]bool, generated map[string]editorGeneratedReferences) error {
 	if e.output == nil || e.output.options.Mode == CompressionUnchanged || e.output.protected {
 		return nil
 	}
@@ -68,9 +75,18 @@ func (e *Editor) compressResourceParts(parts map[string][]byte, reader *Reader,
 			return nil
 		}
 		actual[key] = name
+		data, replaced := parts[name]
+		if known, ok := generated[name]; ok && replaced && known.refs != nil && known.matches(data) {
+			refs.merge(known.refs)
+			continue
+		}
 		var input io.ReadCloser
-		if data, ok := parts[name]; ok {
-			input = io.NopCloser(bytes.NewReader(data))
+		if replaced {
+			if known := generated[name]; data == nil && known.staged != nil {
+				input = known.staged.open()
+			} else {
+				input = io.NopCloser(bytes.NewReader(data))
+			}
 		} else {
 			var err error
 			input, err = reader.openFile(name)
@@ -45,30 +45,31 @@ const editorPathValidationLimit = 1024
 // 阶段可重复，准备完成不代表保存成功
 // 回调不可重入修改编辑器，可返回context.Canceled等调用方停止原因
 type Editor struct {
-	OnWriteProgress func(stage string, completed, total int) error
-	Info            DocInfo
-	pages           []PageContent
-	resources       []editorResource
-	fonts           map[string]*FontResource
-	fontDirs        []string
-	fontFS          []fs.FS
-	fontRenderer    *Renderer
-	fontMetrics     map[string]FontMetrics
-	backends        RenderBackends
-	images          map[string]image.Point
-	resourceID      map[editorResourceKey]string
-	maxID           int
-	history         []editorChange
-	historyIndex    int
-	historyLimit    int
-	revision        uint64
-	serial          uint64
-	source          *editorSource
-	origins         map[string]*editorObjectOrigin
-	removedPages    map[string]bool
-	outlines        []byte
-	encryption      *encryptionState
-	output          *outputOptimization
+	OnWriteProgress  func(stage string, completed, total int) error
+	Info             DocInfo
+	pages            []PageContent
+	resources        []editorResource
+	fonts            map[string]*FontResource
+	fontDirs         []string
+	fontFS           []fs.FS
+	fontRenderer     *Renderer
+	fontSourcesCache *fontSourceCache
+	fontMetrics      map[string]FontMetrics
+	backends         RenderBackends
+	images           map[string]image.Point
+	resourceID       map[editorResourceKey]string
+	maxID            int
+	history          []editorChange
+	historyIndex     int
+	historyLimit     int
+	revision         uint64
+	serial           uint64
+	source           *editorSource
+	origins          map[string]*editorObjectOrigin
+	removedPages     map[string]bool
+	outlines         []byte
+	encryption       *encryptionState
+	output           *outputOptimization
 }

 // editorResource 文档内嵌资源
@@ -100,6 +101,7 @@ type editorResourceKey struct {
 func (e *Editor) SetFontDirs(dirs ...string) {
 	e.fontDirs = slices.Clone(dirs)
 	e.fontRenderer = nil
+	e.fontSourcesCache = nil
 	e.fontMetrics = make(map[string]FontMetrics)
 }

@@ -108,6 +110,7 @@ func (e *Editor) SetFontDirs(dirs ...string) {
 func (e *Editor) SetFontFS(fsys ...fs.FS) {
 	e.fontFS = slices.Clone(fsys)
 	e.fontRenderer = nil
+	e.fontSourcesCache = nil
 	e.fontMetrics = make(map[string]FontMetrics)
 }

@@ -687,7 +690,12 @@ func (e *Editor) LayoutText(obj *TextObject, value string, options TextLayout) e
 // 入参: reader 当前文档快照
 // 返回: *Renderer 度量器
 func (e *Editor) newRenderer(reader *Reader) *Renderer {
-	return NewRenderer(reader, WithFontDirs(e.fontDirs...), WithFontFS(e.fontFS...), WithRenderBackends(e.backends))
+	if e.fontSourcesCache == nil {
+		e.fontSourcesCache = newFontSourceCache()
+	}
+	renderer := NewRenderer(reader, WithFontDirs(e.fontDirs...), WithFontFS(e.fontFS...), WithRenderBackends(e.backends))
+	renderer.fontSourcesCache = e.fontSourcesCache.child()
+	return renderer
 }

 // definition 获取独立定义资源的标识
@@ -1091,14 +1099,12 @@ func (e *Editor) prepareText(obj *TextObject) error {
 	if obj.VScale != 0 || obj.Decoration != "" {
 		return &EditError{Code: EditUnsupportedObject, Err: fmt.Errorf("text extensions are not supported for creation")}
 	}
-	if external {
-		if len(obj.CGTransform) != 0 {
-			return fmt.Errorf("external font text cannot use glyph indices")
-		}
-	} else {
-		if err := validateTextGlyphs(*obj, sfnt.NumGlyphs()); err != nil {
-			return err
-		}
+	glyphCount := 1 << 16
+	if !external {
+		glyphCount = int(sfnt.NumGlyphs())
+	}
+	if err := validateTextGlyphs(*obj, glyphCount); err != nil {
+		return err
 	}
 	if !finite(obj.HScale) || obj.HScale < 0 || !finite(obj.LineWidth) || obj.LineWidth < 0 || !finite(obj.MiterLimit) || obj.MiterLimit < 0 {
 		return fmt.Errorf("invalid text dimensions")
@@ -155,13 +155,85 @@ func (e *Editor) CopyObjectsContext(ctx context.Context, page int, objects []Gra
 		for end < len(prepared) && e.copiedLayerStyle(objects[end]) == style {
 			end++
 		}
-		layers[target].Objects = append(layers[target].Objects, prepared[i:end]...)
+		if len(layers[target].Objects) == 0 {
+			layers[target].Objects = prepared[i:end:end]
+		} else {
+			layers[target].Objects = append(layers[target].Objects, prepared[i:end]...)
+		}
 		i = end
 	}
-	e.replaceLayers(page, layers)
+	e.replacePreparedLayers(page, layers)
 	return result, nil
 }

+// appendOwnedObjects 接收本库生成对象的所有权，全部校验后提交一次撤销记录
+// 入参: ctx 取消上下文, page 目标页, objects 无原文和标识且调用方不再修改的对象
+// 返回: error 校验或取消错误，失败时不提交页面内容
+func (e *Editor) appendOwnedObjects(ctx context.Context, page int, objects []GraphicObject) error {
+	if err := ctx.Err(); err != nil {
+		return err
+	}
+	if _, err := e.page(page); err != nil {
+		return err
+	}
+	if len(objects) == 0 {
+		return nil
+	}
+	if err := e.prepareSourceIDs(); err != nil {
+		return err
+	}
+	validation := &editorValidation{Editor: e}
+	for i, object := range objects {
+		if err := ctx.Err(); err != nil {
+			return err
+		}
+		if object.origin != nil || editorObjectID(object) != "" {
+			return fmt.Errorf("owned objects must not have identifiers or original content")
+		}
+		prepared, err := validation.prepareObject("", object)
+		if err != nil {
+			return err
+		}
+		objects[i] = prepared
+	}
+	if err := ctx.Err(); err != nil {
+		return err
+	}
+	for i := range objects {
+		id := strconv.Itoa(e.maxID + i + 1)
+		switch objects[i].Type {
+		case "TextObject":
+			objects[i].TextObject.ID = id
+		case "PathObject":
+			objects[i].PathObject.ID = id
+		case "ImageObject":
+			objects[i].ImageObject.ID = id
+		case "CompositeObject":
+			objects[i].CompositeGraphicUnit.ID = id
+		}
+	}
+	layers := copyEditorPage(e.pages[page]).Content.Layer
+	target := 0
+	if e.originalPage(page) {
+		target = len(e.source.pages[e.pages[page].ID].original.Content.Layer)
+	}
+	if target < len(layers) {
+		target = len(layers) - 1
+	}
+	e.maxID += len(objects)
+	if target == len(layers) || layers[target].DrawParam != "" {
+		layers = append(layers, Layer{ID: e.nextID(), Type: "Body"})
+		target = len(layers) - 1
+	}
+	if len(layers[target].Objects) == 0 {
+		layers[target].Objects = objects[:len(objects):len(objects)]
+	} else {
+		layers[target].Objects = append(layers[target].Objects, objects...)
+	}
+	e.replacePreparedLayers(page, layers)
+	return nil
+}
+
 // PageObjectInfo 批量获取页面对象能力与位置，不重复扫描同一容器
 // 每次按当前页面计算，变更和撤销后无需失效旧缓存
 // 入参: page 页面索引
@@ -543,12 +615,18 @@ func (e *Editor) replaceLayers(page int, layers []Layer) {
 		return
 	}
 	after := copyEditorPage(PageContent{Content: Content{Layer: layers}}).Content.Layer
+	e.replacePreparedLayers(page, after)
+}
+
+// replacePreparedLayers 提交已隔离的图层容器，历史启用时仍保留独立快照
+// 入参: page 页面索引, after 已准备且不再由调用方修改的图层列表
+func (e *Editor) replacePreparedLayers(page int, after []Layer) {
 	if e.historyLimit == 0 {
 		e.pages[page].Content.Layer = after
 		e.recordChange()
 		return
 	}
-	before = copyEditorPage(PageContent{Content: Content{Layer: before}}).Content.Layer
+	before := copyEditorPage(e.pages[page]).Content.Layer
 	apply := func(e *Editor, layers []Layer) {
 		e.pages[page].Content.Layer = copyEditorPage(PageContent{Content: Content{Layer: layers}}).Content.Layer
 	}
@@ -0,0 +1,123 @@
+// Copyright 2025-2026 肖其顿 (XIAO QI DUN)
+//
+// 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 ofdgo
+
+import (
+	"reflect"
+	"sync"
+)
+
+// editorClonePlans 仅共享类型的可变字段索引，不持有文档对象或资源
+var editorClonePlans sync.Map
+
+// cloneEditorData 深复制编辑快照，未导出的排版及原文来源保持不可变共享
+// 入参: value 原值
+// 返回: T 独立副本
+func cloneEditorData[T any](value T) T {
+	source := reflect.ValueOf(value)
+	if !source.IsValid() {
+		return value
+	}
+	return cloneEditorValue(source).Interface().(T)
+}
+
+// cloneEditorValue 复制公开指针、切片及数组，纯值数据不创建额外结构
+// 入参: value 原值
+// 返回: reflect.Value 独立副本
+func cloneEditorValue(value reflect.Value) reflect.Value {
+	if !value.IsValid() || !editorCloneMutable(value.Type()) {
+		return value
+	}
+	switch value.Kind() {
+	case reflect.Pointer:
+		if value.IsNil() {
+			return value
+		}
+		result := reflect.New(value.Type().Elem())
+		cloneEditorInto(result.Elem(), value.Elem())
+		return result
+	case reflect.Slice:
+		if value.IsNil() {
+			return value
+		}
+		result := reflect.MakeSlice(value.Type(), value.Len(), value.Len())
+		if !editorCloneMutable(value.Type().Elem()) {
+			reflect.Copy(result, value)
+		} else {
+			for i := range value.Len() {
+				cloneEditorInto(result.Index(i), value.Index(i))
+			}
+		}
+		return result
+	default:
+		result := reflect.New(value.Type()).Elem()
+		cloneEditorInto(result, value)
+		return result
+	}
+}
+
+// cloneEditorInto 直接填充目标结构，避免指针及切片成员的二次结构分配
+// 入参: target 可写目标, source 原值
+func cloneEditorInto(target, source reflect.Value) {
+	switch source.Kind() {
+	case reflect.Struct:
+		target.Set(source)
+		for _, i := range editorCloneFields(source.Type()) {
+			cloneEditorInto(target.Field(i), source.Field(i))
+		}
+	case reflect.Array:
+		target.Set(source)
+		if editorCloneMutable(source.Type().Elem()) {
+			for i := range source.Len() {
+				cloneEditorInto(target.Index(i), source.Index(i))
+			}
+		}
+	default:
+		target.Set(cloneEditorValue(source))
+	}
+}
+
+// editorCloneMutable 判断类型是否含有需要隔离的公开可变成员
+// 入参: kind 数据类型
+// 返回: bool 是否需要独立复制
+func editorCloneMutable(kind reflect.Type) bool {
+	switch kind.Kind() {
+	case reflect.Pointer, reflect.Slice:
+		return true
+	case reflect.Array:
+		return editorCloneMutable(kind.Elem())
+	case reflect.Struct:
+		return len(editorCloneFields(kind)) != 0
+	}
+	return false
+}
+
+// editorCloneFields 按类型复用公开可变字段，索引发布后保持只读
+// 入参: kind 结构类型
+// 返回: []int 可变字段索引
+func editorCloneFields(kind reflect.Type) []int {
+	if cached, ok := editorClonePlans.Load(kind); ok {
+		return cached.([]int)
+	}
+	var fields []int
+	for i := range kind.NumField() {
+		field := kind.Field(i)
+		if field.IsExported() && editorCloneMutable(field.Type) {
+			fields = append(fields, i)
+		}
+	}
+	actual, _ := editorClonePlans.LoadOrStore(kind, fields)
+	return actual.([]int)
+}
@@ -49,6 +49,7 @@ func (e *Editor) Transaction(edit func(*Editor) error) error {
 	changed := e.revision != before.revision
 	e.history, e.historyIndex, e.historyLimit = before.history, before.historyIndex, before.historyLimit
 	e.backends, e.fontDirs, e.fontFS = before.backends, before.fontDirs, before.fontFS
+	e.fontSourcesCache = before.fontSourcesCache
 	if e.sameDocument(&before) {
 		e.fontRenderer, e.fontMetrics = before.fontRenderer, before.fontMetrics
 	} else {
@@ -159,6 +160,7 @@ func (e *Editor) restoreTransaction(state Editor) {
 	state.history, state.historyIndex, state.historyLimit = e.history, e.historyIndex, e.historyLimit
 	state.serial = max(e.serial, state.serial)
 	state.backends, state.fontDirs, state.fontFS = e.backends, e.fontDirs, e.fontFS
+	state.fontSourcesCache = e.fontSourcesCache
 	state.encryption, state.output, state.OnWriteProgress = e.encryption, e.output, e.OnWriteProgress
 	*e = state
 }
@@ -912,45 +912,6 @@ func (e *Editor) editorImage(id string) (image.Point, error) {
 	return size, nil
 }

-// cloneEditorData 深复制编辑快照，未导出的排版及原文来源保持不可变共享
-// 入参: value 原值
-// 返回: T 独立副本
-func cloneEditorData[T any](value T) T {
-	return cloneEditorValue(reflect.ValueOf(value)).Interface().(T)
-}
-
-// cloneEditorValue 复制结构体中的可变指针和切片
-// 入参: value 原值
-// 返回: reflect.Value 独立副本
-func cloneEditorValue(value reflect.Value) reflect.Value {
-	switch value.Kind() {
-	case reflect.Pointer:
-		if !value.IsNil() {
-			result := reflect.New(value.Type().Elem())
-			result.Elem().Set(cloneEditorValue(value.Elem()))
-			return result
-		}
-	case reflect.Slice:
-		if !value.IsNil() {
-			result := reflect.MakeSlice(value.Type(), value.Len(), value.Len())
-			for i := range value.Len() {
-				result.Index(i).Set(cloneEditorValue(value.Index(i)))
-			}
-			return result
-		}
-	case reflect.Struct:
-		result := reflect.New(value.Type()).Elem()
-		result.Set(value)
-		for i := range value.NumField() {
-			if result.Field(i).CanSet() && value.Type().Field(i).IsExported() {
-				result.Field(i).Set(cloneEditorValue(value.Field(i)))
-			}
-		}
-		return result
-	}
-	return value
-}
-
 // editorObjectXML 编码可编辑对象，包含独立命名空间声明
 // 入参: object 对象
 // 返回: []byte XML片段, error 错误信息
@@ -31,7 +31,22 @@ import (
 // 入参: progress 保存进度回调，预览时为nil
 // 返回: map[string][]byte 替换及新增条目, error 错误信息
 func (e *Editor) sourceParts(progress editorProgress) (map[string][]byte, error) {
+	return e.sourcePartsReferences(progress, nil)
+}
+
+// sourcePartsReferences 生成保存快照，同步收集完全自产页面的实际编码引用
+// 入参: progress 保存进度回调, generated 本次自产页面引用，可为nil
+// 返回: map[string][]byte 输出条目, error 编码错误
+func (e *Editor) sourcePartsReferences(progress editorProgress, generated map[string]editorGeneratedReferences) (map[string][]byte, error) {
+	return e.sourcePartsPrepared(progress, generated, false)
+}
+
+// sourcePartsPrepared 准备保存条目，按需压缩暂存自产页面
+// 入参: progress 保存进度回调, generated 本次页面及引用, stage 是否压缩暂存
+// 返回: map[string][]byte 输出条目, error 编码错误
+func (e *Editor) sourcePartsPrepared(progress editorProgress, generated map[string]editorGeneratedReferences, stage bool) (map[string][]byte, error) {
 	parts := make(map[string][]byte)
+	var stager editorPageStager
 	source := e.source
 	reader := source.reader
 	pageRefs := make([]Page, len(e.pages))
@@ -56,11 +71,31 @@ func (e *Editor) sourceParts(progress editorProgress) (map[string][]byte, error)
 			}
 		} else {
 			name := e.packageName(packagePagePath("", page.ID))
-			data, err := e.sourceNewPageXML(page)
-			if err != nil {
-				return nil, err
+			var scan *editorReferenceScan
+			if generated != nil {
+				scan = &editorReferenceScan{refs: &editorResourceRefs{ids: make(map[string]bool), files: make(map[string]bool), fonts: make(map[string]*editorFontUsage)}, name: name, safe: true}
+			}
+			if stage && generated != nil && !e.sourcePagePreserved(page) {
+				data, err := stager.encode(e, page, scan)
+				if err != nil {
+					return nil, err
+				}
+				parts[name] = nil
+				known := editorGeneratedReferences{staged: data}
+				if scan != nil && scan.safe && scan.seen && len(scan.stack) == 0 {
+					known.refs = scan.refs
+				}
+				generated[name] = known
+			} else {
+				data, err := e.sourcePageData(page, scan)
+				if err != nil {
+					return nil, err
+				}
+				parts[name] = data
+				if scan != nil && scan.safe && scan.seen && len(scan.stack) == 0 {
+					generated[name] = editorGeneratedReferences{data: data, refs: scan.refs}
+				}
 			}
-			parts[name] = data
 			pageRefs[i] = Page{ID: page.ID, BaseLoc: "/" + name}
 		}
 	}
@@ -80,7 +115,7 @@ func (e *Editor) sourceParts(progress editorProgress) (map[string][]byte, error)
 		})
 	}
 	if len(fonts)+len(images)+len(spaces) != 0 {
-		data, err := encodeOFDXML(func(x *ofdXML) { x.resources(fonts, images, spaces) })
+		data, err := e.encodeXML(func(x *ofdXML) { x.resources(fonts, images, spaces) })
 		if err != nil {
 			return nil, err
 		}
@@ -155,7 +190,7 @@ func (e *Editor) sourceParts(progress editorProgress) (map[string][]byte, error)
 				if original != nil {
 					entries = append(entries, data[original.start:original.end])
 				} else {
-					item, err := encodeOFDXML(func(x *ofdXML) {
+					item, err := e.encodeXML(func(x *ofdXML) {
 						x.root("Page", ofdAttrs{{Name: xml.Name{Local: "ID"}, Value: page.ID}, {Name: xml.Name{Local: "BaseLoc"}, Value: page.BaseLoc}})
 						x.end("Page")
 					})
@@ -215,7 +250,7 @@ func (e *Editor) sourceParts(progress editorProgress) (map[string][]byte, error)
 		parts[name] = updated
 	}
 	if originalCount != len(source.pages) || len(source.annotationPages) != 0 || len(e.removedPages) != 0 {
-		if err := e.prunePageReferences(parts); err != nil {
+		if err := e.prunePageReferences(parts, generated); err != nil {
 			return nil, err
 		}
 	}
@@ -225,6 +260,10 @@ func (e *Editor) sourceParts(progress editorProgress) (map[string][]byte, error)
 			if actual != name {
 				delete(parts, name)
 				parts[actual] = data
+				if known, ok := generated[name]; ok {
+					delete(generated, name)
+					generated[actual] = known
+				}
 			}
 		}
 	}
@@ -392,11 +431,14 @@ func (e *Editor) sourcePageXML(index int, source *editorSourcePage) ([]byte, err
 // 返回: []byte 图层XML, error 错误信息
 func (e *Editor) sourceLayerXML(layer Layer) ([]byte, error) {
 	if !slices.ContainsFunc(layer.Objects, func(object GraphicObject) bool { return e.objectOrigin(editorObjectID(object)) != nil }) {
-		data, err := encodeOFDXML(func(x *ofdXML) {
+		data, err := e.encodeXML(func(x *ofdXML) {
 			attrs := ofdAttrs{{Name: xml.Name{Local: "ID"}, Value: layer.ID}, {Name: xml.Name{Local: "Type"}, Value: layer.Type}}
 			attrs.add("DrawParam", layer.DrawParam)
 			x.root("Layer", attrs)
 			for _, object := range layer.Objects {
+				if x.err != nil {
+					return
+				}
 				x.object(object, false)
 			}
 			x.end("Layer")
@@ -429,16 +471,23 @@ func (e *Editor) sourceLayerXML(layer Layer) ([]byte, error) {
 // 入参: page 页面
 // 返回: []byte 页面XML, error 错误信息
 func (e *Editor) sourceNewPageXML(page PageContent) ([]byte, error) {
-	preserved := false
-	for _, layer := range page.Content.Layer {
-		if slices.ContainsFunc(layer.Objects, func(object GraphicObject) bool { return e.objectOrigin(editorObjectID(object)) != nil }) {
-			preserved = true
-			break
+	return e.sourcePageData(page, nil)
+}
+
+// sourcePageData 编码新增页面，仅在未保留原文时收集编码引用
+// 入参: page 页面内容, references 本次引用记录，可为nil
+// 返回: []byte 页面XML, error 编码错误
+func (e *Editor) sourcePageData(page PageContent, references *editorReferenceScan) ([]byte, error) {
+	if !e.sourcePagePreserved(page) {
+		var data bytes.Buffer
+		x := newOFDXML(&data)
+		if e.output != nil {
+			x.ctx = e.output.ctx
 		}
-	}
-	if !preserved {
-		data, err := encodeOFDXML(func(x *ofdXML) { x.page(page) })
-		return bytes.TrimPrefix(data, []byte(xml.Header)), err
+		x.references = references
+		x.page(page)
+		err := x.finish()
+		return bytes.TrimPrefix(data.Bytes(), []byte(xml.Header)), err
 	}
 	area, err := editorXMLContainer("Area", nil, editorXMLText("PhysicalBox", page.Area.PhysicalBox))
 	if err != nil {
@@ -458,7 +507,7 @@ func (e *Editor) sourceNewPageXML(page PageContent) ([]byte, error) {
 	}
 	data := append(area, content...)
 	if len(page.Actions) != 0 {
-		actions, err := encodeOFDXML(func(x *ofdXML) { x.actions(page.Actions) })
+		actions, err := e.encodeXML(func(x *ofdXML) { x.actions(page.Actions) })
 		if err != nil {
 			return nil, err
 		}
@@ -467,16 +516,29 @@ func (e *Editor) sourceNewPageXML(page PageContent) ([]byte, error) {
 	return editorXMLContainer("Page", nil, data)
 }

+// sourcePagePreserved 判断新增页面是否仍有对象原文需要保留
+// 入参: page 页面内容
+// 返回: bool 是否包含原文对象
+func (e *Editor) sourcePagePreserved(page PageContent) bool {
+	for _, layer := range page.Content.Layer {
+		if slices.ContainsFunc(layer.Objects, func(object GraphicObject) bool { return e.objectOrigin(editorObjectID(object)) != nil }) {
+			return true
+		}
+	}
+	return false
+}
+
 // writeSource 将未修改ZIP条目直接复制到新包，逐项写入改动，不持有原资源解压副本
 // 入参: writer 输出流, fonts 新增字体子集, progress 保存进度回调
 // 返回: int64 写入字节数, error 错误信息
 func (e *Editor) writeSource(writer io.Writer, fonts map[string][]byte, progress editorProgress) (int64, error) {
-	parts, err := e.sourceParts(progress)
+	generated := make(map[string]editorGeneratedReferences)
+	parts, err := e.sourcePartsPrepared(progress, generated, true)
 	if err != nil {
 		return 0, err
 	}
 	maps.Copy(parts, fonts)
-	removed, err := e.compactSourceResources(parts, progress)
+	removed, err := e.compactSourceReferences(parts, progress, generated)
 	if err != nil {
 		return 0, err
 	}
@@ -484,7 +546,7 @@ func (e *Editor) writeSource(writer io.Writer, fonts map[string][]byte, progress
 	if removed == nil {
 		removed = make(map[string]bool)
 	}
-	if err := e.compressResourceParts(parts, reader, removed); err != nil {
+	if err := e.compressResourceReferences(parts, reader, removed, generated); err != nil {
 		return 0, err
 	}
 	if len(parts) != 0 || len(removed) != 0 || e.output != nil && e.output.options.Mode != CompressionUnchanged && !e.output.protected {
@@ -537,7 +599,11 @@ func (e *Editor) writeSource(writer io.Writer, fonts map[string][]byte, progress
 			}
 			if data, ok := remaining[name]; ok {
 				header := file.FileHeader
-				if err := e.writeOutputEntry(archive, header, data); err != nil {
+				if staged := generated[name].staged; data == nil && staged != nil {
+					if err := e.writeStagedPage(archive, header, staged); err != nil {
+						return output.count, err
+					}
+				} else if err := e.writeOutputEntry(archive, header, data); err != nil {
 					return output.count, err
 				}
 				delete(remaining, name)
@@ -575,7 +641,12 @@ func (e *Editor) writeSource(writer io.Writer, fonts map[string][]byte, progress
 		if strings.HasSuffix(strings.ToLower(name), ".xml") {
 			method = zip.Deflate
 		}
-		if err := e.writeOutputEntry(archive, zip.FileHeader{Name: name, Method: method}, remaining[name]); err != nil {
+		header := zip.FileHeader{Name: name, Method: method}
+		if staged := generated[name].staged; remaining[name] == nil && staged != nil {
+			if err := e.writeStagedPage(archive, header, staged); err != nil {
+				return output.count, err
+			}
+		} else if err := e.writeOutputEntry(archive, header, remaining[name]); err != nil {
 			return output.count, err
 		}
 	}
@@ -16,15 +16,16 @@ package ofdgo

 import (
 	"fmt"
+	"io"
 	"maps"
 	"strings"
 )

 // prunePageReferences 在输出快照中删除已移除页面的标准引用，不修改历史快照或其他文档
 // 沿当前文档、页面、模板、资源、注释和签名索引检查，不进入附件、私有扩展和签名原始凭据
-// 入参: parts 输出覆盖条目
+// 入参: parts 输出覆盖条目, generated 本次自产页面及引用，可为nil
 // 返回: error 读取或解析错误
-func (e *Editor) prunePageReferences(parts map[string][]byte) error {
+func (e *Editor) prunePageReferences(parts map[string][]byte, generated map[string]editorGeneratedReferences) error {
 	removed := maps.Clone(e.removedPages)
 	if removed == nil {
 		removed = make(map[string]bool)
@@ -54,6 +55,11 @@ func (e *Editor) prunePageReferences(parts map[string][]byte) error {
 	}
 	seen := make(map[string]bool)
 	for len(queue) > 0 {
+		if e.output != nil {
+			if err := e.output.ctx.Err(); err != nil {
+				return err
+			}
+		}
 		name := queue[0]
 		queue = queue[1:]
 		if seen[strings.ToLower(name)] {
@@ -67,6 +73,17 @@ func (e *Editor) prunePageReferences(parts map[string][]byte) error {
 			if err != nil {
 				return err
 			}
+		} else if known := generated[name]; data == nil && known.staged != nil {
+			input := known.staged.open()
+			if e.output != nil {
+				input = imageInput{ReadCloser: input, context: e.output.ctx}
+			}
+			var err error
+			data, err = io.ReadAll(input)
+			input.Close()
+			if err != nil {
+				return err
+			}
 		}
 		root, err := parseEditorXML(data)
 		if err != nil {
@@ -75,6 +92,11 @@ func (e *Editor) prunePageReferences(parts map[string][]byte) error {
 		var patches []editorXMLPatch
 		var walk func(*editorXML) error
 		walk = func(node *editorXML) error {
+			if e.output != nil {
+				if err := e.output.ctx.Err(); err != nil {
+					return err
+				}
+			}
 			if node.name.Space != "" && node.name.Space != ofdNamespace && node.name.Space != "http://www.ofdspec.org" {
 				return nil
 			}
@@ -153,6 +175,7 @@ func (e *Editor) prunePageReferences(parts map[string][]byte) error {
 		}
 		if len(patches) != 0 {
 			parts[name] = editorPatchXML(data, patches)
+			delete(generated, name)
 		}
 	}
 	return nil
@@ -0,0 +1,232 @@
+// Copyright 2025-2026 肖其顿 (XIAO QI DUN)
+//
+// 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 ofdgo
+
+import (
+	"bytes"
+	"encoding/xml"
+	"maps"
+	"strings"
+	"unicode/utf8"
+)
+
+// editorGeneratedReferences 保存本次编码页面及其引用，不跨写出复用
+type editorGeneratedReferences struct {
+	data   []byte
+	refs   *editorResourceRefs
+	staged *editorStagedPage
+}
+
+// matches 判断条目仍为本次编码结果，空条目与暂存占位不混用
+// 入参: data 当前条目数据
+// 返回: bool 是否可以复用本次引用
+func (r editorGeneratedReferences) matches(data []byte) bool {
+	if r.staged != nil {
+		return data == nil
+	}
+	return bytes.Equal(data, r.data)
+}
+
+// editorReferenceScan 按XML事件收集资源和字体用字，未知内容保持完整扫描
+type editorReferenceScan struct {
+	refs     *editorResourceRefs
+	name     string
+	base     string
+	stack    []string
+	fonts    []*editorFontUsage
+	text     strings.Builder
+	resource bool
+	capture  bool
+	safe     bool
+	seen     bool
+}
+
+// accept 共用读取和自产页面的引用判断，不将对象自身标识视为引用
+// 入参: token 已解析或已编码的XML事件
+// 返回: bool 是否仍可完整判断引用
+func (s *editorReferenceScan) accept(token xml.Token) bool {
+	if !s.safe {
+		return false
+	}
+	switch token := token.(type) {
+	case xml.StartElement:
+		if token.Name.Space != "" && token.Name.Space != ofdNamespace {
+			s.safe = false
+			return false
+		}
+		if token.Name.Local == "CustomTags" || token.Name.Local == "Extensions" || token.Name.Local == "ExtendData" || token.Name.Local == "Data" {
+			s.safe = false
+			return false
+		}
+		if len(s.stack) == 0 {
+			s.seen = true
+			s.resource = token.Name.Local == "Res"
+			if s.resource {
+				for _, attr := range token.Attr {
+					if attr.Name.Local == "BaseLoc" {
+						s.base = attr.Value
+					}
+				}
+			}
+		}
+		var usage *editorFontUsage
+		if len(s.fonts) != 0 {
+			usage = s.fonts[len(s.fonts)-1]
+			if usage != nil && (s.stack[len(s.stack)-1] == "TextCode" || s.stack[len(s.stack)-1] == "Glyphs") {
+				usage.unsafe = true
+			}
+		}
+		for _, attr := range token.Attr {
+			if attr.Name.Space == "xmlns" || attr.Name.Local == "xmlns" || attr.Name.Local == "xmlns:ofd" || attr.Name.Space == "http://www.w3.org/XML/1998/namespace" {
+				continue
+			}
+			if attr.Name.Space != "" {
+				s.safe = false
+				return false
+			}
+			if attr.Name.Local != "ID" {
+				s.refs.reference(s.name, s.base, attr.Name.Local, attr.Value, s.resource)
+			}
+			if attr.Name.Local == "Font" {
+				usage = s.refs.fontUsage(attr.Value)
+				if token.Name.Local != "TextObject" && token.Name.Local != "Text" {
+					usage.unsafe = true
+				}
+			} else if attr.Name.Local == "Substitution" {
+				s.refs.fontUsage(attr.Value).unsafe = true
+			}
+		}
+		s.stack = append(s.stack, token.Name.Local)
+		s.fonts = append(s.fonts, usage)
+		s.text.Reset()
+		s.capture = editorResourceReferenceKind(token.Name.Local, s.resource) != editorReferenceNone || usage != nil && (token.Name.Local == "TextCode" || token.Name.Local == "Glyphs")
+	case xml.CharData:
+		if s.capture {
+			s.text.Write(token)
+		}
+	case xml.EndElement:
+		if len(s.stack) == 0 || s.stack[len(s.stack)-1] != token.Name.Local {
+			s.safe = false
+			return false
+		}
+		s.refs.reference(s.name, s.base, token.Name.Local, s.text.String(), s.resource)
+		if (token.Name.Local == "Substitution" || token.Name.Local == "Font") && strings.TrimSpace(s.text.String()) != "" {
+			s.refs.fontUsage(s.text.String()).unsafe = true
+		}
+		if usage := s.fonts[len(s.fonts)-1]; usage != nil {
+			usage.text(token.Name.Local, s.text.String())
+		}
+		s.stack = s.stack[:len(s.stack)-1]
+		s.fonts = s.fonts[:len(s.fonts)-1]
+		s.text.Reset()
+		s.capture = false
+		if len(s.stack) != 0 {
+			key, usage := s.stack[len(s.stack)-1], s.fonts[len(s.fonts)-1]
+			s.capture = editorResourceReferenceKind(key, s.resource) != editorReferenceNone || usage != nil && (key == "TextCode" || key == "Glyphs")
+		}
+	case xml.Directive:
+		s.safe = false
+	}
+	return s.safe
+}
+
+// merge 合并本次编码已确认的引用，不共享字体用字的可变集合
+// 入参: other 独立页面引用
+func (r *editorResourceRefs) merge(other *editorResourceRefs) {
+	maps.Copy(r.ids, other.ids)
+	maps.Copy(r.files, other.files)
+	for id, usage := range other.fonts {
+		r.fontUsage(id).merge(usage, false)
+	}
+}
+
+// referenceToken 记录自产OFD事件，编码会替换的字符使本页回到完整扫描
+// 入参: token 已写入的XML事件
+func (x *ofdXML) referenceToken(token xml.Token) {
+	if x.references == nil || x.err != nil {
+		return
+	}
+	s := x.references
+	switch value := token.(type) {
+	case xml.StartElement:
+		value.Name = xml.Name{Space: ofdNamespace, Local: strings.TrimPrefix(value.Name.Local, "ofd:")}
+		for _, attr := range value.Attr {
+			if editorResourceReferenceKind(attr.Name.Local, s.resource) != editorReferenceNone && !editorReferenceTextSafe(attr.Value) {
+				s.safe = false
+			}
+		}
+		token = value
+	case xml.EndElement:
+		value.Name = xml.Name{Space: ofdNamespace, Local: strings.TrimPrefix(value.Name.Local, "ofd:")}
+		token = value
+	case xml.CharData:
+		if s.capture && !editorReferenceTextSafe(string(value)) {
+			s.safe = false
+		}
+	}
+	if !s.accept(token) {
+		x.references = nil
+	}
+}
+
+// referenceElement 收集已知叶节点的引用，其他结构仍完整扫描
+// 入参: name 元素名称, value 已编码的子树内容
+func (x *ofdXML) referenceElement(name string, value any) {
+	if x.references == nil || x.err != nil {
+		return
+	}
+	var attrs ofdAttrs
+	switch value := value.(type) {
+	case ShdColor:
+		attrs.add("ColorSpace", value.ColorSpace)
+	case *ShdColor:
+		if value == nil {
+			return
+		}
+		attrs.add("ColorSpace", value.ColorSpace)
+	case *Sound:
+		if value == nil {
+			return
+		}
+		attrs.add("ResourceID", value.ResourceID)
+	case *Movie:
+		if value == nil {
+			return
+		}
+		attrs.add("ResourceID", value.ResourceID)
+	case *GotoBookmark, *URI, *GotoA:
+	default:
+		x.references.safe = false
+		x.references = nil
+		return
+	}
+	x.referenceToken(xml.StartElement{Name: xml.Name{Local: "ofd:" + name}, Attr: attrs})
+	x.referenceToken(xml.EndElement{Name: xml.Name{Local: "ofd:" + name}})
+}
+
+// editorReferenceTextSafe 判断XML编码是否保持引用文本的Unicode值
+// 入参: value 原始引用或字体用字
+// 返回: bool 是否无需字符替换
+func editorReferenceTextSafe(value string) bool {
+	if !utf8.ValidString(value) {
+		return false
+	}
+	for _, char := range value {
+		if char != '\t' && char != '\n' && char != '\r' && (char < 0x20 || char == 0xfffe || char == 0xffff) {
+			return false
+		}
+	}
+	return true
+}
@@ -27,6 +27,12 @@ import (
 	"strings"
 )

+const (
+	editorReferenceNone = iota
+	editorReferenceID
+	editorReferenceFile
+)
+
 // editorResourceRefs 保存全包资源标识和直接文件引用，不以页面是否加载判断资源存活
 type editorResourceRefs struct {
 	ids   map[string]bool
@@ -41,6 +47,13 @@ type editorResourceRefs struct {
 // 入参: parts 已修改和新增的包内条目, progress 保存进度回调
 // 返回: map[string]bool 可移除的二进制条目, error 读取错误
 func (e *Editor) compactSourceResources(parts map[string][]byte, progress editorProgress) (map[string]bool, error) {
+	return e.compactSourceReferences(parts, progress, nil)
+}
+
+// compactSourceReferences 合并自产页面的已知引用，其他条目仍完整扫描
+// 入参: parts 输出条目, progress 保存进度回调, generated 自产页面及引用
+// 返回: map[string]bool 可移除条目, error 读取或取消错误
+func (e *Editor) compactSourceReferences(parts map[string][]byte, progress editorProgress, generated map[string]editorGeneratedReferences) (map[string]bool, error) {
 	if e.source.reader.DocumentCount() > 1 || e.output != nil && e.output.protected {
 		return nil, nil
 	}
@@ -68,9 +81,18 @@ func (e *Editor) compactSourceResources(parts map[string][]byte, progress editor
 		if strings.HasSuffix(name, "/") {
 			continue
 		}
+		data, replaced := parts[name]
+		if known, ok := generated[name]; ok && replaced && known.refs != nil && known.matches(data) {
+			refs.merge(known.refs)
+			continue
+		}
 		var input io.ReadCloser
-		if data, ok := parts[name]; ok {
-			input = io.NopCloser(bytes.NewReader(data))
+		if replaced {
+			if known := generated[name]; data == nil && known.staged != nil {
+				input = known.staged.open()
+			} else {
+				input = io.NopCloser(bytes.NewReader(data))
+			}
 		} else {
 			var err error
 			input, err = reader.openFile(name)
@@ -78,8 +100,14 @@ func (e *Editor) compactSourceResources(parts map[string][]byte, progress editor
 				return nil, err
 			}
 		}
+		if e.output != nil {
+			input = imageInput{ReadCloser: input, context: e.output.ctx}
+		}
 		resource, safe := refs.scan(input, name)
 		input.Close()
+		if e.output != nil && e.output.ctx.Err() != nil {
+			return nil, e.output.ctx.Err()
+		}
 		if !safe {
 			return nil, nil
 		}
@@ -224,79 +252,16 @@ func (r *editorResourceRefs) scan(input io.Reader, name string) (bool, bool) {
 		return false, !strings.EqualFold(path.Ext(name), ".xml")
 	}
 	decoder := xml.NewDecoder(buffer)
-	var stack []string
-	var fonts []*editorFontUsage
-	var text strings.Builder
-	resource, base := false, ""
+	scan := editorReferenceScan{refs: r, name: name, safe: true}
 	for {
 		token, err := decoder.Token()
 		if err == io.EOF {
-			return resource, len(stack) == 0
+			return scan.resource, len(scan.stack) == 0
 		}
 		if err != nil {
 			return false, false
 		}
-		switch token := token.(type) {
-		case xml.StartElement:
-			if token.Name.Space != "" && token.Name.Space != ofdNamespace {
-				return false, false
-			}
-			if token.Name.Local == "CustomTags" || token.Name.Local == "Extensions" || token.Name.Local == "ExtendData" || token.Name.Local == "Data" {
-				return false, false
-			}
-			if len(stack) == 0 {
-				resource = token.Name.Local == "Res"
-				if resource {
-					for _, attr := range token.Attr {
-						if attr.Name.Local == "BaseLoc" {
-							base = attr.Value
-						}
-					}
-				}
-			}
-			var usage *editorFontUsage
-			if len(fonts) != 0 {
-				usage = fonts[len(fonts)-1]
-				if usage != nil && (stack[len(stack)-1] == "TextCode" || stack[len(stack)-1] == "Glyphs") {
-					usage.unsafe = true
-				}
-			}
-			for _, attr := range token.Attr {
-				if attr.Name.Space == "xmlns" || attr.Name.Local == "xmlns" || attr.Name.Space == "http://www.w3.org/XML/1998/namespace" {
-					continue
-				}
-				if attr.Name.Space != "" {
-					return false, false
-				}
-				if attr.Name.Local != "ID" {
-					r.reference(name, base, attr.Name.Local, attr.Value, resource)
-				}
-				if attr.Name.Local == "Font" {
-					usage = r.fontUsage(attr.Value)
-					if token.Name.Local != "TextObject" && token.Name.Local != "Text" {
-						usage.unsafe = true
-					}
-				} else if attr.Name.Local == "Substitution" {
-					r.fontUsage(attr.Value).unsafe = true
-				}
-			}
-			stack = append(stack, token.Name.Local)
-			fonts = append(fonts, usage)
-			text.Reset()
-		case xml.CharData:
-			text.Write(token)
-		case xml.EndElement:
-			r.reference(name, base, token.Name.Local, text.String(), resource)
-			if (token.Name.Local == "Substitution" || token.Name.Local == "Font") && strings.TrimSpace(text.String()) != "" {
-				r.fontUsage(text.String()).unsafe = true
-			}
-			if usage := fonts[len(fonts)-1]; usage != nil {
-				usage.text(token.Name.Local, text.String())
-			}
-			stack = stack[:len(stack)-1]
-			fonts = fonts[:len(fonts)-1]
-			text.Reset()
-		case xml.Directive:
+		if !scan.accept(token) {
 			return false, false
 		}
 	}
@@ -305,32 +270,44 @@ func (r *editorResourceRefs) scan(input io.Reader, name string) (bool, bool) {
 // reference 记录标准资源标识和路径引用，不将普通数字或资源自身ID视为引用
 // 入参: name 当前文件, base 资源目录, key 字段, value 值, resource 是否资源索引
 func (r *editorResourceRefs) reference(name, base, key, value string, resource bool) {
+	kind := editorResourceReferenceKind(key, resource)
+	if kind == editorReferenceNone {
+		return
+	}
 	value = strings.TrimSpace(value)
 	if value == "" {
 		return
 	}
-	switch key {
-	case "Font", "ResourceID", "Substitution", "ImageMask", "Relative", "DrawParam", "ColorSpace", "DefaultCS", "Thumbnail", "TemplateID", "PageID", "PageRef", "RefId":
+	if kind == editorReferenceID {
 		if id := editorResourceID(value); id != "" {
 			r.ids[id] = true
 		}
 		return
-	case "DocRoot", "Signatures", "Cover", "PublicRes", "DocumentRes", "PageRes", "Annotations", "Attachments", "FileLoc", "FileRef", "Profile", "SignedValue", "SchemaLoc", "ExtendData", "File":
-	case "BaseLoc":
-		if resource {
-			return
-		}
-	case "FontFile", "MediaFile":
-		if resource {
-			return
-		}
-	default:
-		return
 	}
 	r.files[editorResourceLocation(name, base, value)] = true
 	r.files[strings.ToLower(cleanPackagePath(resolveResourcePath(name, base, value)))] = true
 }

+// editorResourceReferenceKind 区分标准标识和文件引用，其他文本不参与资源扫描
+// 入参: key 字段名称, resource 是否资源索引
+// 返回: int 引用类型
+func editorResourceReferenceKind(key string, resource bool) int {
+	switch key {
+	case "Font", "ResourceID", "Substitution", "ImageMask", "Relative", "DrawParam", "ColorSpace", "DefaultCS", "Thumbnail", "TemplateID", "PageID", "PageRef", "RefId":
+		return editorReferenceID
+	case "DocRoot", "Signatures", "Cover", "PublicRes", "DocumentRes", "PageRes", "Annotations", "Attachments", "FileLoc", "FileRef", "Profile", "SignedValue", "SchemaLoc", "ExtendData", "File":
+		return editorReferenceFile
+	case "BaseLoc":
+	case "FontFile", "MediaFile":
+	default:
+		return editorReferenceNone
+	}
+	if resource {
+		return editorReferenceNone
+	}
+	return editorReferenceFile
+}
+
 // editorResourceID 规范化ST_ID与ST_RefID的十进制表示
 // 入参: value 标识文本
 // 返回: string 标准十进制标识，无效时为空
@@ -0,0 +1,147 @@
+// Copyright 2025-2026 肖其顿 (XIAO QI DUN)
+//
+// 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 ofdgo
+
+import (
+	"archive/zip"
+	"bytes"
+	"compress/flate"
+	"encoding/xml"
+	"hash/crc32"
+	"io"
+	"unicode/utf8"
+)
+
+// editorStagedPage 保存单次写出的压缩页面及校验值，不持有未压缩XML
+type editorStagedPage struct {
+	data     []byte
+	size     uint64
+	checksum uint32
+}
+
+// editorPageStager 复用单次保存的页面压缩器，不跨文档共享状态
+type editorPageStager struct {
+	compressor *flate.Writer
+}
+
+// editorPageEncoding 同步统计编码字节及CRC，保留实际写入错误
+type editorPageEncoding struct {
+	writer   io.Writer
+	size     uint64
+	checksum uint32
+}
+
+// Write 写入页面字节并统计实际接收的部分
+// 入参: data 未压缩编码数据
+// 返回: int 实际字节数, error 写入或短写错误
+func (w *editorPageEncoding) Write(data []byte) (int, error) {
+	n, err := w.writer.Write(data)
+	w.size += uint64(n)
+	w.checksum = crc32.Update(w.checksum, crc32.IEEETable, data[:n])
+	if n != len(data) && err == nil {
+		err = io.ErrShortWrite
+	}
+	return n, err
+}
+
+// encode 边编码边压缩自产页面，同步收集引用及用字
+// 入参: editor 保存快照, page 页面内容, references 本次引用扫描
+// 返回: *editorStagedPage 压缩页面, error 编码或取消错误
+func (s *editorPageStager) encode(editor *Editor, page PageContent, references *editorReferenceScan) (*editorStagedPage, error) {
+	var data bytes.Buffer
+	if s.compressor == nil {
+		level := flate.DefaultCompression
+		if editor.output != nil && editor.output.options.Mode != CompressionUnchanged {
+			level = flate.BestCompression
+		}
+		var err error
+		s.compressor, err = flate.NewWriter(&data, level)
+		if err != nil {
+			return nil, err
+		}
+	} else {
+		s.compressor.Reset(&data)
+	}
+	output := editorPageEncoding{writer: s.compressor}
+	x := &ofdXML{encoder: xml.NewEncoder(&output), references: references}
+	if editor.output != nil {
+		x.ctx = editor.output.ctx
+	}
+	x.page(page)
+	if err := x.finish(); err != nil {
+		return nil, err
+	}
+	if err := s.compressor.Close(); err != nil {
+		return nil, err
+	}
+	if !x.checkContext() {
+		return nil, x.err
+	}
+	return &editorStagedPage{data: data.Bytes(), size: output.size, checksum: output.checksum}, nil
+}
+
+// open 流式读取暂存页面，仅用于无法完整收集编码引用的分支
+// 返回: io.ReadCloser 页面解压流
+func (p *editorStagedPage) open() io.ReadCloser {
+	return flate.NewReader(bytes.NewReader(p.data))
+}
+
+// writeStagedPage 直接写入已经压缩的自产页面，保持ZIP条目属性和取消语义
+// 入参: archive 输出归档, header 条目属性, page 暂存页面
+// 返回: error 写入、短写或取消错误
+func (e *Editor) writeStagedPage(archive *zip.Writer, header zip.FileHeader, page *editorStagedPage) error {
+	if e.output != nil && e.output.options.Mode != CompressionUnchanged {
+		if err := editorProgress(e.OnWriteProgress).report("compress", e.output.completed, 0); err != nil {
+			return err
+		}
+	}
+	header.Method, header.ReaderVersion = zip.Deflate, 20
+	header.CreatorVersion = header.CreatorVersion&0xff00 | 20
+	header.CRC32, header.UncompressedSize64, header.CompressedSize64 = page.checksum, page.size, uint64(len(page.data))
+	header.Flags &^= 8
+	if header.NonUTF8 {
+		header.Flags &^= 0x800
+	} else if utf8.ValidString(header.Name) && utf8.ValidString(header.Comment) {
+		for _, value := range header.Name + header.Comment {
+			if value < 0x20 || value > 0x7d || value == 0x5c {
+				header.Flags |= 0x800
+				break
+			}
+		}
+	}
+	entry, err := archive.CreateRaw(&header)
+	if err != nil {
+		return err
+	}
+	for offset := 0; offset < len(page.data); {
+		if e.output != nil {
+			if err := e.output.ctx.Err(); err != nil {
+				return err
+			}
+		}
+		end := min(offset+64<<10, len(page.data))
+		if n, err := entry.Write(page.data[offset:end]); err != nil {
+			return err
+		} else if n != end-offset {
+			return io.ErrShortWrite
+		}
+		offset = end
+	}
+	if e.output != nil && e.output.options.Mode != CompressionUnchanged {
+		e.output.completed++
+		return editorProgress(e.OnWriteProgress).report("compress", e.output.completed, 0)
+	}
+	return nil
+}
@@ -133,6 +133,12 @@ type resolvedFontResult struct {
 	err  error
 }

+// fontDataKey 区分只读字体的起点和长度，指针保留源数据直到预算淘汰
+type fontDataKey struct {
+	data *byte
+	size int
+}
+
 // ResolveFont 使用配置的字体后端解析资源，保留精确匹配与回退的区别
 // 入参: id 字体资源标识, exact 是否要求精确匹配
 // 返回: ResolvedFont 字体与来源, error 解析错误
@@ -165,7 +171,7 @@ func (r *Renderer) PrepareFont(id string) (*PreparedFont, error) {
 	prepared := &PreparedFont{ResolvedFont: resolved}
 	definition := r.Reader.fontCache[id]
 	if definition != nil && definition.FontFile != "" {
-		key := sha256.Sum256(resolved.Data)
+		key := r.fontDigest(resolved.Data)
 		if content, ok := r.fontPreparations.get(key); ok {
 			prepared.Data, prepared.Glyphs, prepared.CIDs, prepared.digest = content.Data, content.Glyphs, content.CIDs, content.digest
 			r.preparedFonts[id] = prepared
@@ -188,7 +194,7 @@ func (r *Renderer) PrepareFont(id string) (*PreparedFont, error) {
 				}
 			}
 		}
-		prepared.digest = sha256.Sum256(prepared.Data)
+		prepared.digest = r.fontDigest(prepared.Data)
 		cost := min(len(prepared.Data), r.fontPreparations.limit)*2 + (len(prepared.Glyphs)+len(prepared.CIDs))*48 + 256
 		r.fontPreparations.put(key, PreparedFont{ResolvedFont: ResolvedFont{Data: prepared.Data}, Glyphs: prepared.Glyphs, CIDs: prepared.CIDs, digest: prepared.digest}, cost)
 	}
@@ -196,6 +202,22 @@ func (r *Renderer) PrepareFont(id string) (*PreparedFont, error) {
 	return prepared, nil
 }

+// fontDigest 在预算内复用只读字体摘要，内容副本仍按摘要复用解析结果
+// 入参: data 不得修改的字体数据
+// 返回: [32]byte 字体摘要
+func (r *Renderer) fontDigest(data []byte) [32]byte {
+	if len(data) == 0 {
+		return sha256.Sum256(nil)
+	}
+	key := fontDataKey{data: &data[0], size: len(data)}
+	if digest, ok := r.fontDigests.get(key); ok {
+		return digest
+	}
+	digest := sha256.Sum256(data)
+	r.fontDigests.put(key, digest, min(cap(data), r.fontDigests.limit)+256)
+	return digest
+}
+
 // resetFontCache 重置字体解析及字形缓存，保留图片和非字体后端状态
 func (r *Renderer) resetFontCache() {
 	r.fontSourcesCache = newFontSourceCache()
@@ -206,6 +228,7 @@ func (r *Renderer) resetFontCache() {
 func (r *Renderer) resetFontBackendCache() {
 	r.sharedBackendStates = make(map[any]any)
 	r.preparedFonts = make(map[string]*PreparedFont)
+	r.fontDigests = &renderCache[fontDataKey, [32]byte]{limit: 4 << 20}
 	r.fontPreparations = &renderCache[[32]byte, PreparedFont]{limit: 16 << 20}
 	r.resolvedFonts = make(map[resolvedFontKey]resolvedFontResult)
 	if r.fontSourcesCache != nil {
@@ -19,6 +19,8 @@ import (
 	"encoding/binary"
 	"fmt"
 	"sort"
+	"unicode/utf16"
+	"unicode/utf8"
 )

 // cmapSegment cmap表段结构
@@ -29,7 +31,7 @@ type cmapSegment struct {
 	offset     uint16
 }

-// normalizeCmap 转换旧式双字节子表，保留各编码记录及原字形编号
+// normalizeCmap 转换扩展字符子表，保留各编码记录及原字形编号
 // 入参: data 字符映射表
 // 返回: []byte 等价字符映射表, error 子表错误
 func normalizeCmap(data []byte) ([]byte, error) {
@@ -48,7 +50,8 @@ func normalizeCmap(data []byte) ([]byte, error) {
 		if uint64(offset)+2 > uint64(len(data)) {
 			return nil, fmt.Errorf("invalid cmap subtable offset")
 		}
-		if binary.BigEndian.Uint16(data[offset:]) != 2 {
+		format := binary.BigEndian.Uint16(data[offset:])
+		if format != 2 && format != 8 && format != 10 && format != 13 {
 			continue
 		}
 		if out == nil {
@@ -56,17 +59,35 @@ func normalizeCmap(data []byte) ([]byte, error) {
 		}
 		next, ok := converted[offset]
 		if !ok {
-			mapping, err := parseCmapFormat2(data[offset:])
+			var mapping map[rune]uint16
+			var err error
+			language := uint32(0)
+			if format == 2 {
+				mapping, err = parseCmapFormat2(data[offset:])
+				if err == nil {
+					language = uint32(binary.BigEndian.Uint16(data[offset+4:]))
+				}
+			} else {
+				mapping, err = parseCmapExtended(data[offset:])
+				if err == nil {
+					language = binary.BigEndian.Uint32(data[offset+8:])
+				}
+			}
 			if err != nil {
 				return nil, err
 			}
 			table := buildCmapTable(0, mapping)
+			if language > 65535 {
+				table = buildCmapTableFormat12(0, mapping)
+			}
 			subtable := table[binary.BigEndian.Uint32(table[8:]):]
-			language := binary.BigEndian.Uint16(data[offset+4:])
 			if binary.BigEndian.Uint16(subtable) == 12 {
-				binary.BigEndian.PutUint32(subtable[8:], uint32(language))
+				binary.BigEndian.PutUint32(subtable[8:], language)
 			} else {
-				binary.BigEndian.PutUint16(subtable[4:], language)
+				binary.BigEndian.PutUint16(subtable[4:], uint16(language))
+			}
+			if uint64(len(out))+uint64(len(subtable)) > 1<<32-1 {
+				return nil, fmt.Errorf("normalized cmap exceeds offset range")
 			}
 			next = uint32(len(out))
 			converted[offset] = next
@@ -80,6 +101,110 @@ func normalizeCmap(data []byte) ([]byte, error) {
 	return out, nil
 }

+// parseCmapExtended 读取混合字符、扩展数组及常量分组，限制到有效Unicode范围
+// 入参: data 格式8、10或13子表
+// 返回: map[rune]uint16 独立字符映射, error 结构或映射错误
+func parseCmapExtended(data []byte) (map[rune]uint16, error) {
+	if len(data) < 12 {
+		return nil, fmt.Errorf("truncated extended cmap")
+	}
+	length := uint64(binary.BigEndian.Uint32(data[4:]))
+	if length < 12 || length > uint64(len(data)) {
+		return nil, fmt.Errorf("invalid extended cmap length")
+	}
+	data = data[:int(length)]
+	format := binary.BigEndian.Uint16(data)
+	result := make(map[rune]uint16)
+	add := func(code uint32, glyph uint16) error {
+		char := rune(code)
+		if code > utf8.MaxRune {
+			high, low := rune(code>>16), rune(code&65535)
+			if format != 8 || high < 0xd800 || high > 0xdbff || low < 0xdc00 || low > 0xdfff {
+				return fmt.Errorf("invalid extended cmap character")
+			}
+			char = utf16.DecodeRune(high, low)
+		}
+		if glyph != 0 && utf8.ValidRune(char) {
+			if previous, exists := result[char]; exists && previous != glyph {
+				return fmt.Errorf("ambiguous extended cmap character")
+			}
+			result[char] = glyph
+		}
+		return nil
+	}
+	if format == 10 {
+		if len(data) < 20 {
+			return nil, fmt.Errorf("truncated cmap format 10")
+		}
+		first, count := uint64(binary.BigEndian.Uint32(data[12:])), uint64(binary.BigEndian.Uint32(data[16:]))
+		if count > uint64((len(data)-20)/2) || first+count > utf8.MaxRune+1 {
+			return nil, fmt.Errorf("invalid cmap format 10 range")
+		}
+		for index := uint64(0); index < count; index++ {
+			if err := add(uint32(first+index), binary.BigEndian.Uint16(data[20+int(index)*2:])); err != nil {
+				return nil, err
+			}
+		}
+		return result, nil
+	}
+	offset := 16
+	if format == 8 {
+		offset = 8208
+	} else if format != 13 {
+		return nil, fmt.Errorf("invalid extended cmap format %d", format)
+	}
+	if len(data) < offset {
+		return nil, fmt.Errorf("truncated cmap format %d", format)
+	}
+	count := uint64(binary.BigEndian.Uint32(data[offset-4:]))
+	if count > uint64((len(data)-offset)/12) {
+		return nil, fmt.Errorf("invalid extended cmap group count")
+	}
+	var previous uint32
+	for index := uint64(0); index < count; index++ {
+		group := data[offset+int(index)*12:]
+		first, last, glyph := binary.BigEndian.Uint32(group), binary.BigEndian.Uint32(group[4:]), uint64(binary.BigEndian.Uint32(group[8:]))
+		if first > last || index != 0 && first <= previous {
+			return nil, fmt.Errorf("invalid extended cmap group order")
+		}
+		previous = last
+		endGlyph := glyph
+		if format == 8 {
+			endGlyph += uint64(last) - uint64(first)
+			if first <= 65535 && last > 65535 {
+				return nil, fmt.Errorf("mixed cmap character range")
+			}
+			if last > utf8.MaxRune && (first>>16 != last>>16 || first>>16 < 0xd800 || last>>16 > 0xdbff || first&65535 < 0xdc00 || last&65535 > 0xdfff) {
+				return nil, fmt.Errorf("invalid UTF-16 cmap group")
+			}
+		} else if last > utf8.MaxRune {
+			return nil, fmt.Errorf("invalid cmap Unicode range")
+		}
+		if endGlyph > 65535 {
+			return nil, fmt.Errorf("extended cmap glyph index overflow")
+		}
+		for code := uint64(first); code <= uint64(last); code++ {
+			if format == 8 {
+				word := code
+				if code > 65535 {
+					word >>= 16
+				}
+				if (data[12+word/8]&(1<<(7-word%8)) != 0) != (code > 65535) {
+					return nil, fmt.Errorf("invalid cmap character width")
+				}
+			}
+			id := glyph
+			if format == 8 {
+				id += code - uint64(first)
+			}
+			if err := add(uint32(code), uint16(id)); err != nil {
+				return nil, err
+			}
+		}
+	}
+	return result, nil
+}
+
 // parseCmapFormat2 解析混合单字节与双字节字符的子头映射
 // 入参: data format 2子表
 // 返回: map[rune]uint16 编码值到字形编号, error 子头或字形数组越界
@@ -155,8 +280,8 @@ func parseCmapMappings(data []byte) map[rune]uint16 {
 		if platformID != 0 && platformID != 3 {
 			continue
 		}
-		offset := int(binary.BigEndian.Uint32(data[pos+4 : pos+8]))
-		if offset+2 > len(data) {
+		offset := uint64(binary.BigEndian.Uint32(data[pos+4 : pos+8]))
+		if offset+2 > uint64(len(data)) {
 			continue
 		}
 		switch binary.BigEndian.Uint16(data[offset : offset+2]) {
@@ -168,6 +293,19 @@ func parseCmapMappings(data []byte) map[rune]uint16 {
 			parseCmapFormat6(data[offset:], result)
 		case 12:
 			parseCmapFormat12(data[offset:], result)
+		case 2, 8, 10, 13:
+			var mapping map[rune]uint16
+			var err error
+			if binary.BigEndian.Uint16(data[offset:]) == 2 {
+				mapping, err = parseCmapFormat2(data[offset:])
+			} else {
+				mapping, err = parseCmapExtended(data[offset:])
+			}
+			if err == nil {
+				for char, glyph := range mapping {
+					result[char] = glyph
+				}
+			}
 		}
 	}
 	if len(result) == 0 {
@@ -263,27 +401,28 @@ func parseCmapFormat12(data []byte, result map[rune]uint16) {
 	if len(data) < 16 {
 		return
 	}
-	length := int(binary.BigEndian.Uint32(data[4:8]))
-	if length > len(data) {
-		length = len(data)
+	length := uint64(binary.BigEndian.Uint32(data[4:8]))
+	if length < 16 || length > uint64(len(data)) {
+		return
 	}
-	nGroups := int(binary.BigEndian.Uint32(data[12:16]))
-	for i := 0; i < nGroups; i++ {
-		pos := 16 + i*12
-		if pos+12 > length {
-			break
-		}
+	nGroups := uint64(binary.BigEndian.Uint32(data[12:16]))
+	if nGroups > (length-16)/12 {
+		return
+	}
+	for i := uint64(0); i < nGroups; i++ {
+		pos := 16 + int(i)*12
 		startChar := binary.BigEndian.Uint32(data[pos:])
 		endChar := binary.BigEndian.Uint32(data[pos+4:])
 		startGID := binary.BigEndian.Uint32(data[pos+8:])
-		for c := startChar; c <= endChar; c++ {
-			gid := startGID + c - startChar
-			if gid != 0 && gid <= 0xFFFF {
+		if startChar > endChar || startGID > 65535 || startChar > utf8.MaxRune {
+			continue
+		}
+		last := min(uint64(endChar), uint64(startChar)+65535-uint64(startGID), uint64(utf8.MaxRune))
+		for c := uint64(startChar); c <= last; c++ {
+			gid := uint64(startGID) + c - uint64(startChar)
+			if gid != 0 && utf8.ValidRune(rune(c)) {
 				result[rune(c)] = uint16(gid)
 			}
-			if c == endChar {
-				break
-			}
 		}
 	}
 }
@@ -104,6 +104,13 @@ func (r *Reader) Fonts() ([]Font, error) {
 // 入参: id 字体资源标识
 // 返回: []byte 字体数据, error 资源缺失或集合解析错误
 func (r *Reader) FontData(id string) ([]byte, error) {
+	return r.fontData(id, false)
+}
+
+// fontData 按调用方所有权读取内嵌字体，集合字体始终独立提取
+// 入参: id 字体资源标识, readonly 是否允许借用只读资源
+// 返回: []byte 字体数据, error 资源或集合错误
+func (r *Reader) fontData(id string, readonly bool) ([]byte, error) {
 	if r.fontCache[id] == nil {
 		if _, err := r.Fonts(); err != nil {
 			return nil, err
@@ -113,7 +120,11 @@ func (r *Reader) FontData(id string) ([]byte, error) {
 	if f == nil || f.FontFile == "" {
 		return nil, fmt.Errorf("embedded font %q not found", id)
 	}
-	data, err := r.ResData(f.FontFile)
+	read := r.readFile
+	if readonly {
+		read = r.readFileView
+	}
+	data, err := read(r.ResPath(f.FontFile))
 	if err != nil {
 		return nil, err
 	}
@@ -223,7 +234,7 @@ func (r *Reader) embeddedFontFace(of Font) (*FontFace, error) {
 	name := r.ResPath(of.FontFile)
 	faces, ok := r.fontFaces[name]
 	if !ok {
-		data, err := r.readFile(name)
+		data, err := r.readFileView(name)
 		if err != nil {
 			return nil, err
 		}
@@ -84,6 +84,14 @@ func newFontSourceCache() *fontSourceCache {
 	return &fontSourceCache{candidates: make(map[string][]fontSource), directories: make(map[string][]fontFileCandidate), filesystems: make(map[int][]fontFileCandidate), data: make(map[fontSourceKey]fontSourceData)}
 }

+// child 复用相同来源的文件索引与数据，匹配和回退状态保持独立
+// 返回: *fontSourceCache 独立文档缓存
+func (c *fontSourceCache) child() *fontSourceCache {
+	cache := newFontSourceCache()
+	cache.directories, cache.filesystems, cache.data = c.directories, c.filesystems, c.data
+	return cache
+}
+
 // readFontSource 读取并拆出独立字体，保留集合中的名称和样式选择
 // 入参: source 字体来源, definition 字体定义
 // 返回: []byte 字体数据, error 读取或集合错误
@@ -156,7 +164,7 @@ func (r *Renderer) resolveFontSource(backend FontBackend, id string, exact bool)
 		definition = &Font{ID: id}
 	}
 	if definition.FontFile != "" {
-		data, err := r.Reader.FontData(id)
+		data, err := r.Reader.fontData(id, true)
 		if err != nil {
 			return ResolvedFont{}, err
 		}
@@ -42,6 +42,9 @@ func (p GeometryPath) Transform(matrix Matrix) (GeometryPath, error) {
 		return Point{x, y}
 	}
 	var result GeometryPath
+	if len(p) > 0 {
+		result = make(GeometryPath, 0, len(p))
+	}
 	var current, start Point
 	for _, s := range p {
 		originalSegment := s
@@ -21,7 +21,6 @@ import (
 	"math"
 	"reflect"
 	"strconv"
-	"strings"

 	"github.com/xiaoqidun/pdfgo"
 )
@@ -62,6 +61,12 @@ type PDFImportReport struct {
 	Warnings     []pdfgo.Diagnostic
 }

+// pdfPointBounds 逐点汇总边界，不保留文字或路径的临时坐标数组
+type pdfPointBounds struct {
+	minX, minY, maxX, maxY float64
+	found                  bool
+}
+
 // pdfImporter 保存单次转换的对象与资源状态
 type pdfImporter struct {
 	ctx              context.Context
@@ -77,7 +82,7 @@ type pdfImporter struct {
 	clipTexts        map[*pdfgo.TextClip][]TextObject
 	clipPaths        *renderCache[[32]byte, pdfClipPath]
 	cmykSpace        string
-	objects          []GraphicObject
+	objects          *pdfObjectBuffer
 	pages            map[pdfgo.Reference]*pdfgo.Page
 	pageIDs          map[pdfgo.Reference]string
 	pageIndexes      map[pdfgo.Reference]int
@@ -143,6 +148,7 @@ func ImportPDF(ctx context.Context, source io.ReaderAt, size int64, options PDFI
 	editor := NewEditor()
 	editor.SetRenderBackends(renderer.Backends())
 	editor.fontDirs, editor.fontFS = renderer.FontSources()
+	editor.fontSourcesCache = renderer.fontSourcesCache
 	importer := pdfImporter{ctx: ctx, reader: reader, editor: editor, renderer: renderer, rasterDPI: renderer.DPI, fontIDs: map[*pdfgo.Font]string{}, fonts: map[*pdfgo.Font]*pdfImportedFont{}, pages: map[pdfgo.Reference]*pdfgo.Page{}, pageIDs: map[pdfgo.Reference]string{}}
 	importer.report.Warnings = diagnostics
 	importer.pageIndexes = make(map[pdfgo.Reference]int)
@@ -270,18 +276,21 @@ func ConvertPDF(ctx context.Context, source io.ReaderAt, size int64, output io.W
 		}
 		return nil
 	}
-	var written int64
-	_, err = editor.WriteTo(convertWriter{context: ctx, writer: output, count: &written})
+	_, err = editor.WriteToWithOptions(ctx, output, WriteOptions{})
 	return report, err
 }

 // commitObjects 将当前页面待转换对象写入编辑文档
 // 返回: error 对象写入错误
 func (p *pdfImporter) commitObjects() error {
-	if len(p.objects) == 0 {
+	if p.objects.len() == 0 {
 		return nil
 	}
-	if _, err := p.editor.CopyObjectsContext(p.ctx, p.page, p.objects, 0, 0); err != nil {
+	objects, err := p.objects.data(p.ctx)
+	if err != nil {
+		return err
+	}
+	if err := p.editor.appendOwnedObjects(p.ctx, p.page, objects); err != nil {
 		return err
 	}
 	p.objects = nil
@@ -313,11 +322,15 @@ func pdfPageMatrix(page *pdfgo.Page) (pdfgo.Matrix, float64, float64) {
 // 入参: values 数值列表
 // 返回: string 空格分隔的数值
 func pdfNumbers(values ...float64) string {
-	parts := make([]string, len(values))
+	var buffer [128]byte
+	data := buffer[:0]
 	for n, v := range values {
-		parts[n] = strconv.FormatFloat(v, 'g', -1, 64)
+		if n != 0 {
+			data = append(data, ' ')
+		}
+		data = strconv.AppendFloat(data, v, 'g', -1, 64)
 	}
-	return strings.Join(parts, " ")
+	return string(data)
 }

 // pdfBoundary 序列化对象边界
@@ -329,17 +342,31 @@ func pdfBoundary(b Box) string { return pdfNumbers(b.X, b.Y, b.W, b.H) }
 // 入参: points 坐标点集合
 // 返回: Box 包围框
 func pdfBounds(points []pdfgo.Point) Box {
-	if len(points) == 0 {
+	var bounds pdfPointBounds
+	for _, point := range points {
+		bounds.add(point)
+	}
+	return bounds.box()
+}
+
+// add 按原顺序累计一个控制点的最小和最大坐标
+// 入参: point 坐标点
+func (b *pdfPointBounds) add(point pdfgo.Point) {
+	if !b.found {
+		b.minX, b.minY, b.maxX, b.maxY = point.X, point.Y, point.X, point.Y
+		b.found = true
+	}
+	b.minX, b.minY = math.Min(b.minX, point.X), math.Min(b.minY, point.Y)
+	b.maxX, b.maxY = math.Max(b.maxX, point.X), math.Max(b.maxY, point.Y)
+}
+
+// box 取得原精度边界，空集合仍返回零值
+// 返回: Box 坐标范围
+func (b *pdfPointBounds) box() Box {
+	if !b.found {
 		return Box{}
 	}
-	minX, minY, maxX, maxY := points[0].X, points[0].Y, points[0].X, points[0].Y
-	for _, p := range points {
-		minX = math.Min(minX, p.X)
-		minY = math.Min(minY, p.Y)
-		maxX = math.Max(maxX, p.X)
-		maxY = math.Max(maxY, p.Y)
-	}
-	return Box{minX, minY, math.Max(maxX-minX, 1e-6), math.Max(maxY-minY, 1e-6)}
+	return Box{b.minX, b.minY, math.Max(b.maxX-b.minX, 1e-6), math.Max(b.maxY-b.minY, 1e-6)}
 }

 // pathBoundsContext 逐段计算变换后的控制点边界，不创建临时点集合
@@ -462,12 +489,12 @@ func (p *pdfImporter) clips(paths []pdfgo.Path, origin Box) (*Clips, error) {
 				if err := p.flushPath(); err != nil {
 					return nil, err
 				}
-				count := len(p.objects)
+				count := p.objects.len()
 				textCount := p.report.TextObjects
 				if err := p.text(pdfgo.TextMark{Font: mark.Font, Glyphs: mark.Glyphs, Positions: mark.Positions, Matrix: mark.Matrix, Size: mark.Size, HorizontalScale: mark.HorizontalScale, Mode: 7, Clip: mark, Style: pdfgo.Style{Fill: pdfgo.Paint{Alpha: 1}}}); err != nil {
 					return nil, err
 				}
-				p.objects = p.objects[:count]
+				p.objects.truncate(count)
 				p.report.TextObjects = textCount
 				objects = p.clipTexts[mark]
 			}
@@ -617,7 +644,7 @@ func (p *pdfImporter) appendPath(mark pdfgo.PathMark) error {
 		phase := mark.Style.DashPhase * scale
 		object.DashOffset = &phase
 	}
-	p.objects = append(p.objects, GraphicObject{Type: "PathObject", PathObject: object})
+	p.appendObject(GraphicObject{Type: "PathObject", PathObject: object})
 	p.report.PathObjects++
 	return nil
 }
@@ -737,14 +764,18 @@ func (p *pdfImporter) tilingPattern(source *pdfgo.TilingPattern, base pdfgo.Pain
 	if err := cell.flushPath(); err != nil {
 		return nil, err
 	}
-	pattern.CellContent.Objects = cell.objects
+	objects, err := cell.objects.data(p.ctx)
+	if err != nil {
+		return nil, err
+	}
+	pattern.CellContent.Objects = objects
 	if width > xstep || height > ystep {
 		columns, rows := math.Ceil(width/xstep), math.Ceil(height/ystep)
 		if !finite(columns*rows) || columns*rows > float64(int(^uint(0)>>1)) {
 			return nil, fmt.Errorf("PDF tiling pattern instance count overflow")
 		}
-		bounds := make([]Box, len(cell.objects))
-		for index, object := range cell.objects {
+		bounds := make([]Box, len(objects))
+		for index, object := range objects {
 			if object.Type == "PathObject" && object.PathObject.Stroke != nil && !*object.PathObject.Stroke {
 				var err error
 				bounds[index], err = ParseBox(object.PathObject.Boundary)
@@ -753,14 +784,19 @@ func (p *pdfImporter) tilingPattern(source *pdfgo.TilingPattern, base pdfgo.Pain
 				}
 			}
 		}
-		pattern.CellContent.Objects = nil
+		var expanded pdfObjectBuffer
 		for row := 1 - int(rows); row <= 0; row++ {
 			for column := 1 - int(columns); column <= 0; column++ {
 				if err := p.ctx.Err(); err != nil {
 					return nil, err
 				}
 				dx, dy := float64(column)*xstep*unit, float64(row)*ystep*unit
-				for index, object := range cell.objects {
+				for index, object := range objects {
+					if index%pdfObjectChunkLimit == 0 {
+						if err := p.ctx.Err(); err != nil {
+							return nil, err
+						}
+					}
 					b := bounds[index]
 					if b.W > 0 && b.H > 0 && (b.X+dx+b.W < 0 || b.Y+dy+b.H < 0 || b.X+dx > pattern.Width || b.Y+dy > pattern.Height) {
 						continue
@@ -769,10 +805,14 @@ func (p *pdfImporter) tilingPattern(source *pdfgo.TilingPattern, base pdfgo.Pain
 					if err != nil {
 						return nil, err
 					}
-					pattern.CellContent.Objects = append(pattern.CellContent.Objects, translated)
+					expanded.append(translated)
 				}
 			}
 		}
+		pattern.CellContent.Objects, err = expanded.data(p.ctx)
+		if err != nil {
+			return nil, err
+		}
 	}
 	return pattern, nil
 }
@@ -98,7 +98,11 @@ func (p *pdfImporter) appearanceAnnotation(ctx context.Context, page *pdfgo.Page
 		p.compositeNodes = stamp.compositeNodes
 		p.transferBackdrop = stamp.transferBackdrop
 	}
-	converted := Annotation{Type: pdfAnnotationType(annotation.Subtype), Subtype: string(annotation.Subtype), Remark: remark, Appearance: Appearance{Boundary: pdfBoundary(box), Objects: stamp.objects}}
+	objects, err := stamp.objects.data(ctx)
+	if err != nil {
+		return err
+	}
+	converted := Annotation{Type: pdfAnnotationType(annotation.Subtype), Subtype: string(annotation.Subtype), Remark: remark, Appearance: Appearance{Boundary: pdfBoundary(box), Objects: objects}}
 	if annotation.Subtype == "Redact" {
 		converted.Parameters = &AnnotationParameters{Parameter: []AnnotationParameter{{Name: "PDF.Redact.Pending", Value: "true"}}}
 	}
@@ -0,0 +1,119 @@
+// Copyright 2025-2026 肖其顿 (XIAO QI DUN)
+//
+// 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 ofdgo
+
+import "context"
+
+// pdfObjectChunkLimit 限制导入暂存片段的容量，不限制页面对象数量
+const pdfObjectChunkLimit = 256
+
+// pdfObjectBuffer 分块收集导入对象，提交前只整理一次连续数据
+type pdfObjectBuffer struct {
+	parts [][]GraphicObject
+	size  int
+}
+
+// appendObject 追加本次导入对象，按需建立独立暂存缓冲
+// 入参: object 独立导入对象
+func (p *pdfImporter) appendObject(object GraphicObject) {
+	if p.objects == nil {
+		p.objects = &pdfObjectBuffer{}
+	}
+	p.objects.append(object)
+}
+
+// append 追加对象，小批量连续存储，超过阈值后不复制前序片段
+// 入参: object 独立对象
+func (b *pdfObjectBuffer) append(object GraphicObject) {
+	index := len(b.parts) - 1
+	if index == 0 && len(b.parts[0]) == cap(b.parts[0]) && cap(b.parts[0]) < pdfObjectChunkLimit {
+		part := b.parts[0]
+		next := make([]GraphicObject, len(part), min(pdfObjectChunkLimit, max(1, cap(part)*2)))
+		copy(next, part)
+		b.parts[0] = next
+	}
+	if index < 0 || len(b.parts[index]) == cap(b.parts[index]) {
+		capacity := pdfObjectChunkLimit
+		if index < 0 {
+			capacity = 1
+		}
+		b.parts = append(b.parts, make([]GraphicObject, 0, capacity))
+		index++
+	}
+	b.parts[index] = append(b.parts[index], object)
+	b.size++
+}
+
+// len 获取暂存对象数量，空缓冲返回零
+// 返回: int 对象数量
+func (b *pdfObjectBuffer) len() int {
+	if b == nil {
+		return 0
+	}
+	return b.size
+}
+
+// truncate 回滚到已有对象位置，清理丢弃对象及片段引用
+// 入参: size 保留对象数量
+func (b *pdfObjectBuffer) truncate(size int) {
+	if size < 0 || size > b.len() {
+		panic("invalid PDF object buffer size")
+	}
+	if size == b.len() {
+		return
+	}
+	remaining := size
+	for index, part := range b.parts {
+		if remaining <= len(part) {
+			clear(part[remaining:])
+			b.parts[index] = part[:remaining]
+			clear(b.parts[index+1:])
+			b.parts = b.parts[:index+1]
+			b.size = size
+			return
+		}
+		remaining -= len(part)
+	}
+}
+
+// data 按绘制顺序整理独立对象数组，取消时保留原片段供重试
+// 入参: ctx 取消上下文
+// 返回: []GraphicObject 暂存数据, error 取消错误
+func (b *pdfObjectBuffer) data(ctx context.Context) ([]GraphicObject, error) {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	if b.len() == 0 {
+		return nil, nil
+	}
+	if len(b.parts) == 1 {
+		return b.parts[0][:b.size:b.size], nil
+	}
+	objects := make([]GraphicObject, b.size)
+	offset := 0
+	for _, part := range b.parts {
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
+		offset += copy(objects[offset:], part)
+	}
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	b.parts[0] = objects
+	clear(b.parts[1:])
+	b.parts = b.parts[:1]
+	return objects, nil
+}
@@ -50,6 +50,7 @@ type pdfCompositor struct {
 	width, height   int
 	inverse         pdfgo.Matrix
 	cache           *pdfCompositeCache
+	scratch         *pdfCompositeScratch
 	masks           map[*pdfgo.SoftMask][]float64
 	meshes          map[pdfMeshKey][]pdfShadingPixel
 	gradients       map[pdfGradientKey][]pdfShadingPixel
@@ -71,6 +72,15 @@ type pdfCompositeGeometry struct {
 	box   Box
 }

+// pdfCompositeColorSampler 在单次绘制内复用完全相同参数的轴向颜色，不近似函数
+type pdfCompositeColorSampler struct {
+	paint      pdfgo.Paint
+	space      *pdfgo.ColorSpace
+	intent     pdfgo.Name
+	conversion pdfgo.ColorConversion
+	values     map[float64][4]float64
+}
+
 // pdfCompositeCache 在单页内复用几何、图像分量及蒙版内容
 type pdfCompositeCache struct {
 	geometry     map[pdfCompositeKey]pdfCompositeGeometry
@@ -242,8 +252,17 @@ func (p *pdfImporter) compositeTextBounds(mark pdfgo.TextMark, stroke bool) Box
 // 返回: []pdfCompositeNode 原始图元, error 解析错误
 func (p *pdfImporter) collectCompositeNodes(walk func(pdfgo.Visitor) error) ([]pdfCompositeNode, error) {
 	active := make(map[*pdfgo.Font]map[string]bool)
+	var seenFonts map[*pdfgo.Font]bool
 	var externalNames map[string]bool
 	var embeddedFonts map[*pdfgo.Font]bool
+	var textMarks []pdfgo.TextMark
+	storeText := func(mark pdfgo.TextMark) *pdfgo.TextMark {
+		if len(textMarks) == cap(textMarks) {
+			textMarks = make([]pdfgo.TextMark, 0, min(64, max(8, cap(textMarks)*2)))
+		}
+		textMarks = append(textMarks, mark)
+		return &textMarks[len(textMarks)-1]
+	}
 	var collect func(func(pdfgo.Visitor) error) ([]pdfCompositeNode, error)
 	collect = func(walk func(pdfgo.Visitor) error) ([]pdfCompositeNode, error) {
 		var nodes []pdfCompositeNode
@@ -259,7 +278,11 @@ func (p *pdfImporter) collectCompositeNodes(walk func(pdfgo.Visitor) error) ([]p
 			return nil
 		}
 		v.Text = func(mark pdfgo.TextMark) error {
-			if mark.Font.Subtype != "Type3" {
+			if mark.Font.Subtype != "Type3" && !seenFonts[mark.Font] {
+				if seenFonts == nil {
+					seenFonts = make(map[*pdfgo.Font]bool)
+				}
+				seenFonts[mark.Font] = true
 				if len(mark.Font.Program) == 0 {
 					if externalNames == nil {
 						externalNames = make(map[string]bool)
@@ -277,7 +300,13 @@ func (p *pdfImporter) collectCompositeNodes(walk func(pdfgo.Visitor) error) ([]p
 					embeddedFonts[mark.Font] = true
 				}
 			}
-			var text []pdfCompositeNode
+			text := &nodes
+			if mark.Object != nil && mark.Object.Knockout {
+				if len(nodes) == 0 || nodes[len(nodes)-1].textObject != mark.Object {
+					nodes = append(nodes, pdfCompositeNode{group: &pdfgo.GroupMark{Alpha: 1, Knockout: true}, textObject: mark.Object})
+				}
+				text = &nodes[len(nodes)-1].children
+			}
 			if mark.Font.Subtype != "Type3" {
 				if err := p.halftone(mark.Style); err != nil {
 					return err
@@ -289,10 +318,10 @@ func (p *pdfImporter) collectCompositeNodes(walk func(pdfgo.Visitor) error) ([]p
 					for i := range mark.Glyphs {
 						glyph := mark
 						glyph.Glyphs, glyph.Positions, glyph.Clip = mark.Glyphs[i:i+1], mark.Positions[i:i+1], nil
-						text = append(text, pdfCompositeNode{text: &glyph})
+						*text = append(*text, pdfCompositeNode{text: storeText(glyph)})
 					}
 				} else {
-					text = append(text, pdfCompositeNode{text: &mark})
+					*text = append(*text, pdfCompositeNode{text: storeText(mark)})
 				}
 			} else {
 				glyphs := active[mark.Font]
@@ -305,25 +334,17 @@ func (p *pdfImporter) collectCompositeNodes(walk func(pdfgo.Visitor) error) ([]p
 						return fmt.Errorf("recursive PDF Type3 glyph %q", glyph.Name)
 					}
 					glyphs[glyph.Name] = true
-					children, err := collect(func(visitor pdfgo.Visitor) error { return p.reader.WalkType3Glyph(p.ctx, mark, index, visitor) })
+					source := mark
+					children, err := collect(func(visitor pdfgo.Visitor) error { return p.reader.WalkType3Glyph(p.ctx, source, index, visitor) })
 					delete(glyphs, glyph.Name)
 					if err != nil {
 						return err
 					}
 					part := mark
 					part.Glyphs, part.Positions = mark.Glyphs[index:index+1], mark.Positions[index:index+1]
-					text = append(text, pdfCompositeNode{group: &pdfgo.GroupMark{Alpha: 1}, children: children, glyph: &part})
+					*text = append(*text, pdfCompositeNode{group: &pdfgo.GroupMark{Alpha: 1}, children: children, glyph: &part})
 				}
 			}
-			if mark.Object != nil && mark.Object.Knockout {
-				if len(nodes) == 0 || nodes[len(nodes)-1].textObject != mark.Object {
-					nodes = append(nodes, pdfCompositeNode{group: &pdfgo.GroupMark{Alpha: 1, Knockout: true}, textObject: mark.Object})
-				}
-				node := &nodes[len(nodes)-1]
-				node.children = append(node.children, text...)
-			} else {
-				nodes = append(nodes, text...)
-			}
 			return nil
 		}
 		v.Image = func(mark pdfgo.ImageMark) error {
@@ -624,6 +645,9 @@ func (n pdfCompositeNode) coverage(p *pdfImporter, stroke bool) error {
 // 入参: backdrop 前序图元, node 当前效果, space 页面混合空间, flatten 是否合并页面背景
 // 返回: error 合成或保存错误
 func (p *pdfImporter) compositeRegion(backdrop []pdfCompositeNode, node pdfCompositeNode, space *pdfgo.ColorSpace, flatten bool) error {
+	if err := p.ctx.Err(); err != nil {
+		return err
+	}
 	scene, box, err := p.compileGroup(func(local *pdfImporter) error {
 		if node.group != nil {
 			return node.coverage(local, false)
@@ -650,21 +674,28 @@ func (p *pdfImporter) compositeRegion(backdrop []pdfCompositeNode, node pdfCompo
 	if err != nil {
 		return err
 	}
+	if err := p.ctx.Err(); err != nil {
+		return err
+	}
 	output := image.NewNRGBA(image.Rect(0, 0, w, h))
 	cache := p.compositingCache()
 	step := 25.4 / p.rasterDPI
-	buffer := make([]pdfCompositePixel, min(256, w)*min(256, h))
+	buffer := make([]pdfCompositePixel, min(pdfCompositeTileSize, w)*min(pdfCompositeTileSize, h))
+	scratch := &pdfCompositeScratch{}
 	var transfers []*pdfgo.TransferFunction
 	var previousTransfers []*pdfgo.TransferFunction
 	if pdfCompositeHasTransfer(backdrop) || pdfCompositeHasTransfer([]pdfCompositeNode{node}) {
 		transfers = make([]*pdfgo.TransferFunction, len(buffer))
 		previousTransfers = make([]*pdfgo.TransferFunction, len(buffer))
 	}
-	for y := 0; y < h; y += 256 {
-		for x := 0; x < w; x += 256 {
+	for y := 0; y < h; y += pdfCompositeTileSize {
+		for x := 0; x < w; x += pdfCompositeTileSize {
+			if err := p.ctx.Err(); err != nil {
+				return err
+			}
 			runtime.Gosched()
-			width, height := min(256, w-x), min(256, h-y)
-			c := pdfCompositor{importer: p, box: Box{X: box.X + float64(x)*step, Y: box.Y + float64(y)*step, W: float64(width) * step, H: float64(height) * step}, width: width, height: height, inverse: inverse, cache: cache, masks: map[*pdfgo.SoftMask][]float64{}}
+			width, height := min(pdfCompositeTileSize, w-x), min(pdfCompositeTileSize, h-y)
+			c := pdfCompositor{importer: p, box: Box{X: box.X + float64(x)*step, Y: box.Y + float64(y)*step, W: float64(width) * step, H: float64(height) * step}, width: width, height: height, inverse: inverse, cache: cache, scratch: scratch, masks: map[*pdfgo.SoftMask][]float64{}}
 			if transfers != nil {
 				c.transfers = transfers[:width*height]
 				clear(c.transfers)
@@ -682,6 +713,11 @@ func (p *pdfImporter) compositeRegion(backdrop []pdfCompositeNode, node pdfCompo
 				return err
 			}
 			for i, pixel := range pixels {
+				if i%pdfCompositeTileSize == 0 {
+					if err := p.ctx.Err(); err != nil {
+						return err
+					}
+				}
 				if pixel.effect == 0 && (c.transfers == nil || c.transfers[i] == previousTransfers[i]) {
 					continue
 				}
@@ -711,6 +747,7 @@ func (p *pdfImporter) compositeRegion(backdrop []pdfCompositeNode, node pdfCompo
 				}
 				output.SetNRGBA(x+i%width, y+i/width, color.NRGBA{R: uint8(math.Round(rgb[0] * 255)), G: uint8(math.Round(rgb[1] * 255)), B: uint8(math.Round(rgb[2] * 255)), A: uint8(math.Round(alpha * 255))})
 			}
+			c.releaseMasks()
 		}
 	}
 	return p.appendRasterGroup(output, box, 1)
@@ -15,6 +15,7 @@
 package ofdgo

 import (
+	"context"
 	"fmt"
 	"math"

@@ -25,6 +26,9 @@ import (
 // 入参: nodes 图元, pixels 背景与输出像素, space 混合空间
 // 返回: error 几何或颜色错误
 func (c *pdfCompositor) draw(nodes []pdfCompositeNode, pixels []pdfCompositePixel, space *pdfgo.ColorSpace) error {
+	if err := c.importer.ctx.Err(); err != nil {
+		return err
+	}
 	for _, node := range nodes {
 		if err := c.importer.ctx.Err(); err != nil {
 			return err
@@ -59,9 +63,13 @@ func (c *pdfCompositor) drawMark(node pdfCompositeNode, pixels []pdfCompositePix
 	result := pixels
 	grouped := fill && stroke && (style.FillOverprint || style.StrokeOverprint) && style.Fill.Alpha == style.Stroke.Alpha
 	if fill && stroke {
-		initial = append([]pdfCompositePixel(nil), pixels...)
+		initial = c.acquirePixels(len(pixels))
+		defer c.releasePixels(initial)
+		copy(initial, pixels)
 		if grouped {
-			result = append([]pdfCompositePixel(nil), pixels...)
+			result = c.acquirePixels(len(pixels))
+			defer c.releasePixels(result)
+			copy(result, pixels)
 			for i := range result {
 				result[i].effect = 0
 				result[i].shape = 0
@@ -88,7 +96,7 @@ func (c *pdfCompositor) drawMark(node pdfCompositeNode, pixels []pdfCompositePix
 		}
 	}
 	if grouped {
-		return pdfCompositeGroup(pixels, initial, result, space, space, style.Fill.Alpha, mask, style.BlendMode, style.RenderingIntent, style.ColorConversion)
+		return pdfCompositeGroup(c.importer.ctx, pixels, initial, result, space, space, style.Fill.Alpha, mask, style.BlendMode, style.RenderingIntent, style.ColorConversion)
 	}
 	return nil
 }
@@ -146,6 +154,7 @@ func (c *pdfCompositor) drawPaint(node pdfCompositeNode, outline bool, paint pdf
 		if err != nil {
 			return err
 		}
+		defer c.releasePixels(pattern)
 	}
 	var shading []pdfShadingPixel
 	if paint.Mesh != nil && (node.image == nil || node.image.Image.ImageMask) {
@@ -161,6 +170,11 @@ func (c *pdfCompositor) drawPaint(node pdfCompositeNode, outline bool, paint pdf
 	}
 	var uniformValues [4]float64
 	uniformReady := false
+	var sampler *pdfCompositeColorSampler
+	if g := paint.Axial; node.image == nil && shading == nil && g != nil && g.Bounds == nil && g.Background == nil &&
+		(g.Start.Y == g.End.Y && c.inverse[2] == 0 || g.Start.X == g.End.X && c.inverse[1] == 0) {
+		sampler = &pdfCompositeColorSampler{paint: paint, space: space, intent: style.RenderingIntent, conversion: style.ColorConversion}
+	}
 	geometry, err := c.geometry(node, outline)
 	if err != nil {
 		return err
@@ -231,7 +245,11 @@ func (c *pdfCompositor) drawPaint(node pdfCompositeNode, outline bool, paint pdf
 					opacity *= alpha
 				} else {
 					var visible bool
-					values, visible, err = pdfCompositeColor(paint, point, space, style.RenderingIntent, style.ColorConversion)
+					if sampler != nil {
+						values, visible, err = sampler.color(point)
+					} else {
+						values, visible, err = pdfCompositeColor(paint, point, space, style.RenderingIntent, style.ColorConversion)
+					}
 					if !visible {
 						shape = 0
 					}
@@ -311,11 +329,19 @@ func (c *pdfCompositor) drawGroup(node pdfCompositeNode, pixels []pdfCompositePi
 	if !g.Isolated && !g.Knockout && g.Alpha == 1 && g.SoftMask == nil && pdfNormalBlend(g.BlendMode) && space.Equal(parent) {
 		return c.draw(node.children, pixels, space)
 	}
-	initial := make([]pdfCompositePixel, len(pixels))
-	if !g.Isolated {
+	initial := c.acquirePixels(len(pixels))
+	defer c.releasePixels(initial)
+	if g.Isolated {
+		clear(initial)
+	} else {
 		sameSpace := space.Equal(parent)
 		components := parent.Components()
 		for i, pixel := range pixels {
+			if i%pdfCompositeTileSize == 0 {
+				if err := c.importer.ctx.Err(); err != nil {
+					return err
+				}
+			}
 			values := pixel.values
 			if !sameSpace {
 				var err error
@@ -327,7 +353,9 @@ func (c *pdfCompositor) drawGroup(node pdfCompositeNode, pixels []pdfCompositePi
 			initial[i] = pdfCompositePixel{values: values, alpha: pixel.alpha}
 		}
 	}
-	result := append([]pdfCompositePixel(nil), initial...)
+	result := c.acquirePixels(len(initial))
+	defer c.releasePixels(result)
+	copy(result, initial)
 	if g.Knockout {
 		if err := c.drawKnockout(node.children, initial, result, space); err != nil {
 			return err
@@ -343,20 +371,30 @@ func (c *pdfCompositor) drawGroup(node pdfCompositeNode, pixels []pdfCompositePi
 	}
 	if g.AlphaIsShape {
 		for i := range result {
+			if i%pdfCompositeTileSize == 0 {
+				if err := c.importer.ctx.Err(); err != nil {
+					return err
+				}
+			}
 			result[i].shape *= g.Alpha
 			if mask != nil {
 				result[i].shape *= mask[i]
 			}
 		}
 	}
-	return pdfCompositeGroup(pixels, initial, result, space, parent, g.Alpha, mask, g.BlendMode, g.RenderingIntent, g.ColorConversion)
+	return pdfCompositeGroup(c.importer.ctx, pixels, initial, result, space, parent, g.Alpha, mask, g.BlendMode, g.RenderingIntent, g.ColorConversion)
 }

 // drawKnockout 将每个对象与组初始背景合成，再按对象形状替换先前对象
 // 入参: nodes 组图元, initial 初始背景, pixels 组输出, space 混合空间
 // 返回: error 图元或合成错误
 func (c *pdfCompositor) drawKnockout(nodes []pdfCompositeNode, initial, pixels []pdfCompositePixel, space *pdfgo.ColorSpace) error {
-	candidate := make([]pdfCompositePixel, len(initial))
+	if err := c.importer.ctx.Err(); err != nil {
+		return err
+	}
+	candidate := c.acquirePixels(len(initial))
+	defer c.releasePixels(candidate)
+	clear(candidate)
 	for _, node := range nodes {
 		var box Box
 		var err error
@@ -373,6 +411,9 @@ func (c *pdfCompositor) drawKnockout(nodes []pdfCompositeNode, initial, pixels [
 			continue
 		}
 		for y := region.Min.Y; y < region.Max.Y; y++ {
+			if err := c.importer.ctx.Err(); err != nil {
+				return err
+			}
 			start, end := y*c.width+region.Min.X, y*c.width+region.Max.X
 			copy(candidate[start:end], initial[start:end])
 		}
@@ -407,9 +448,12 @@ func pdfCompositeKnockout(target *pdfCompositePixel, initial, source pdfComposit
 }

 // pdfCompositeGroup 移除组初始背景贡献后，将组结果合成到父空间
-// 入参: pixels 父组输出, initial 初始背景, result 组结果, space 组空间, parent 父空间, opacity 组不透明度, mask 蒙版, mode 混合模式, intent 渲染意图, conversion 设备转换函数
+// 入参: ctx 取消上下文, pixels 父组输出, initial 初始背景, result 组结果, space 组空间, parent 父空间, opacity 组不透明度, mask 蒙版, mode 混合模式, intent 渲染意图, conversion 设备转换函数
 // 返回: error 颜色或混合错误
-func pdfCompositeGroup(pixels, initial, result []pdfCompositePixel, space, parent *pdfgo.ColorSpace, opacity float64, mask []float64, mode, intent pdfgo.Name, conversion pdfgo.ColorConversion) error {
+func pdfCompositeGroup(ctx context.Context, pixels, initial, result []pdfCompositePixel, space, parent *pdfgo.ColorSpace, opacity float64, mask []float64, mode, intent pdfgo.Name, conversion pdfgo.ColorConversion) error {
+	if err := ctx.Err(); err != nil {
+		return err
+	}
 	if len(pixels) != len(initial) || len(pixels) != len(result) || mask != nil && len(mask) != len(pixels) {
 		return fmt.Errorf("invalid PDF transparency group buffers")
 	}
@@ -420,6 +464,11 @@ func pdfCompositeGroup(pixels, initial, result []pdfCompositePixel, space, paren
 	}
 	components, parentComponents := space.Components(), parent.Components()
 	for i, pixel := range result {
+		if i%pdfCompositeTileSize == 0 {
+			if err := ctx.Err(); err != nil {
+				return err
+			}
+		}
 		alpha := opacity
 		if mask != nil {
 			alpha *= mask[i]
@@ -441,6 +490,9 @@ func pdfCompositeGroup(pixels, initial, result []pdfCompositePixel, space, paren
 // 入参: mask 蒙版或空值, inherited 未指定蒙版空间时继承的空间
 // 返回: []float64 蒙版透明度或空值, error 蒙版错误
 func (c *pdfCompositor) mask(mask *pdfgo.SoftMask, inherited *pdfgo.ColorSpace) ([]float64, error) {
+	if err := c.importer.ctx.Err(); err != nil {
+		return nil, err
+	}
 	if mask == nil {
 		return nil, nil
 	}
@@ -450,7 +502,7 @@ func (c *pdfCompositor) mask(mask *pdfgo.SoftMask, inherited *pdfgo.ColorSpace)
 	nodes, ok := c.cache.masks[mask]
 	var err error
 	if !ok {
-		nodes, err = c.importer.collectCompositeNodes(mask.Walk)
+		nodes, err = c.importer.collectCompositeNodes(func(v pdfgo.Visitor) error { return mask.WalkContext(c.importer.ctx, v) })
 		if err != nil {
 			return nil, err
 		}
@@ -460,9 +512,16 @@ func (c *pdfCompositor) mask(mask *pdfgo.SoftMask, inherited *pdfgo.ColorSpace)
 	if space == nil {
 		space = inherited
 	}
-	pixels := make([]pdfCompositePixel, c.width*c.height)
+	pixels := c.acquirePixels(c.width * c.height)
+	defer c.releasePixels(pixels)
+	clear(pixels)
 	if mask.Subtype == "Luminosity" {
 		for i := range pixels {
+			if i%pdfCompositeTileSize == 0 {
+				if err := c.importer.ctx.Err(); err != nil {
+					return nil, err
+				}
+			}
 			copy(pixels[i].values[:], mask.Backdrop)
 			pixels[i].alpha = 1
 		}
@@ -472,8 +531,19 @@ func (c *pdfCompositor) mask(mask *pdfgo.SoftMask, inherited *pdfgo.ColorSpace)
 	if err := local.draw(nodes, pixels, space); err != nil {
 		return nil, err
 	}
-	result := make([]float64, len(pixels))
+	result := c.scratch.masks.acquire(len(pixels))
+	complete := false
+	defer func() {
+		if !complete {
+			c.scratch.masks.release(result)
+		}
+	}()
 	for i, pixel := range pixels {
+		if i%pdfCompositeTileSize == 0 {
+			if err := c.importer.ctx.Err(); err != nil {
+				return nil, err
+			}
+		}
 		v := pixel.alpha
 		if mask.Subtype == "Luminosity" {
 			v, err = space.Luminosity(pixel.values[:space.Components()], "RelativeColorimetric")
@@ -486,6 +556,7 @@ func (c *pdfCompositor) mask(mask *pdfgo.SoftMask, inherited *pdfgo.ColorSpace)
 			return nil, err
 		}
 	}
+	complete = true
 	c.masks[mask] = result
 	return result, nil
 }
@@ -494,15 +565,6 @@ func (c *pdfCompositor) mask(mask *pdfgo.SoftMask, inherited *pdfgo.ColorSpace)
 // 入参: mark 图像, point 页面坐标, space 混合空间
 // 返回: [4]float64 源颜色, float64 图像透明度, error 解码或变换错误
 func (c *pdfCompositor) imageColor(mark *pdfgo.ImageMark, point pdfgo.Point, space *pdfgo.ColorSpace) ([4]float64, float64, error) {
-	source := c.cache.images[mark.Image]
-	if source == nil {
-		var err error
-		source, err = mark.Image.DecodeComponentsContext(c.importer.ctx)
-		if err != nil {
-			return [4]float64{}, 0, err
-		}
-		c.cache.images[mark.Image] = source
-	}
 	inverse, found := c.cache.imageMatrix[mark]
 	if !found {
 		var ok bool
@@ -519,6 +581,15 @@ func (c *pdfCompositor) imageColor(mark *pdfgo.ImageMark, point pdfgo.Point, spa
 	if point.X < 0 || point.X > 1 || point.Y < 0 || point.Y > 1 {
 		return [4]float64{}, 0, nil
 	}
+	source := c.cache.images[mark.Image]
+	if source == nil {
+		var err error
+		source, err = mark.Image.DecodeComponentsViewContext(c.importer.ctx)
+		if err != nil {
+			return [4]float64{}, 0, err
+		}
+		c.cache.images[mark.Image] = source
+	}
 	x, y := point.X*float64(source.Rect.Dx())-.5, (1-point.Y)*float64(source.Rect.Dy())-.5
 	_, nativeProcess := source.Process.CMYKMask()
 	process := mark.Style.FillOverprint && space.Model == "DeviceCMYK" && !space.Calibrated() && len(source.Colorants) > 0
@@ -724,6 +795,31 @@ func pdfCompositeColor(paint pdfgo.Paint, point pdfgo.Point, space *pdfgo.ColorS
 	return values, true, err
 }

+// color 复用精确参数对应的颜色，区域、背景及其他画刷保持原求值路径
+// 入参: point 页面坐标
+// 返回: [4]float64 混合分量, bool 是否着色, error 颜色错误
+func (s *pdfCompositeColorSampler) color(point pdfgo.Point) ([4]float64, bool, error) {
+	g := s.paint.Axial
+	if g == nil || g.Bounds != nil || g.Background != nil || s.paint.Function != nil || s.paint.Tiling != nil {
+		return pdfCompositeColor(s.paint, point, s.space, s.intent, s.conversion)
+	}
+	position, visible := pdfShadingPosition(s.paint, point)
+	if !visible {
+		return [4]float64{}, false, nil
+	}
+	if values, ok := s.values[position]; ok {
+		return values, true, nil
+	}
+	values, visible, err := pdfCompositeColor(s.paint, point, s.space, s.intent, s.conversion)
+	if err == nil && visible && len(s.values) < 2*pdfCompositeTileSize {
+		if s.values == nil {
+			s.values = make(map[float64][4]float64)
+		}
+		s.values[position] = values
+	}
+	return values, visible, err
+}
+
 // pdfCompositePaintOver 以兼容叠印计算零色料分量，再应用对象混合模式
 // 入参: backdrop 背景及输出, source 源分量, alpha 源透明度, space 混合空间, mode 混合模式, compatible 是否保留零色料背景, marked 原生过程通道，nil使用普通空间
 // 返回: error 未支持的混合模式
@@ -57,11 +57,12 @@ func (c *pdfCompositor) tiling(paint pdfgo.Paint, backdrop []pdfCompositePixel,
 			return nil, fmt.Errorf("PDF tiling pattern instance range overflow")
 		}
 	}
-	result := make([]pdfCompositePixel, len(backdrop))
+	result := c.acquirePixels(len(backdrop))
 	for i, pixel := range backdrop {
 		result[i] = pdfCompositePixel{values: pixel.values, alpha: pixel.alpha}
 	}
-	candidate := make([]pdfCompositePixel, len(backdrop))
+	candidate := c.acquirePixels(len(backdrop))
+	defer c.releasePixels(candidate)
 	disjoint := source.BBox.XMax-source.BBox.XMin <= xstep && source.BBox.YMax-source.BBox.YMin <= ystep
 	pixelStep := 25.4 / c.importer.rasterDPI
 	localInverse, _ := pattern.local.matrix.Inverse()
@@ -92,10 +93,11 @@ func (c *pdfCompositor) tiling(paint pdfgo.Paint, backdrop []pdfCompositePixel,
 				}
 			}
 			cell := pdfCompositor{importer: &pattern.local, box: Box{X: c.box.X + float64(left)*pixelStep - dx, Y: c.box.Y + float64(top)*pixelStep - dy, W: float64(width) * pixelStep, H: float64(height) * pixelStep},
-				width: width, height: height, inverse: localInverse, cache: pattern.local.compositingCache(), masks: map[*pdfgo.SoftMask][]float64{}}
+				width: width, height: height, inverse: localInverse, cache: pattern.local.compositingCache(), scratch: c.scratch, masks: map[*pdfgo.SoftMask][]float64{}}
 			if err := cell.draw(pattern.nodes, pixels, space); err != nil {
 				return nil, err
 			}
+			cell.releaseMasks()
 			for y := range height {
 				for x := range width {
 					index := (top+y)*c.width + left + x
@@ -105,6 +107,11 @@ func (c *pdfCompositor) tiling(paint pdfgo.Paint, backdrop []pdfCompositePixel,
 		}
 	}
 	for i := range result {
+		if i%pdfCompositeTileSize == 0 {
+			if err := c.importer.ctx.Err(); err != nil {
+				return nil, err
+			}
+		}
 		pixel := &result[i]
 		if pixel.effect != 0 {
 			for j := 0; j < space.Components(); j++ {
@@ -187,7 +194,7 @@ func (p *pdfImporter) directCompositeNode(node pdfCompositeNode, space *pdfgo.Co
 				source := p.compositingCache().images[node.image.Image]
 				if source == nil {
 					var err error
-					source, err = node.image.Image.DecodeComponentsContext(p.ctx)
+					source, err = node.image.Image.DecodeComponentsViewContext(p.ctx)
 					if err != nil {
 						return false, err
 					}
@@ -0,0 +1,87 @@
+// Copyright 2025-2026 肖其顿 (XIAO QI DUN)
+//
+// 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 ofdgo
+
+// pdfCompositeTileSize 限定单次合成块的像素边长
+const pdfCompositeTileSize = 256
+
+// pdfCompositeBuffers 在一次局部合成中保留至多八个块容量内的空闲缓冲
+type pdfCompositeBuffers[T any] [8][]T
+
+// pdfCompositeScratch 复用像素及蒙版缓冲，不跨页持有活动内容
+type pdfCompositeScratch struct {
+	pixels pdfCompositeBuffers[pdfCompositePixel]
+	masks  pdfCompositeBuffers[float64]
+}
+
+// acquire 借用尚未初始化的独立缓冲，嵌套调用不复用活动缓冲
+// 入参: count 所需元素数量
+// 返回: []T 可由当前调用独占的缓冲
+func (s *pdfCompositeBuffers[T]) acquire(count int) []T {
+	selected := -1
+	for i, buffer := range s {
+		if cap(buffer) >= count && (selected < 0 || cap(buffer) < cap(s[selected])) {
+			selected = i
+		}
+	}
+	if selected < 0 {
+		return make([]T, count)
+	}
+	buffer := s[selected]
+	s[selected] = nil
+	return buffer[:count]
+}
+
+// release 归还已不再访问的缓冲，超出块容量或空闲槽数量时直接释放
+// 入参: buffer 当前调用借用的缓冲
+func (s *pdfCompositeBuffers[T]) release(buffer []T) {
+	if len(buffer) == 0 || cap(buffer) > pdfCompositeTileSize*pdfCompositeTileSize {
+		return
+	}
+	for i, stored := range s {
+		if stored == nil {
+			s[i] = buffer
+			return
+		}
+	}
+}
+
+// acquirePixels 为当前合成器借用独立像素缓冲
+// 入参: count 所需像素数量
+// 返回: []pdfCompositePixel 尚未初始化的独占缓冲
+func (c *pdfCompositor) acquirePixels(count int) []pdfCompositePixel {
+	if c.scratch == nil {
+		c.scratch = &pdfCompositeScratch{}
+	}
+	return c.scratch.pixels.acquire(count)
+}
+
+// releasePixels 归还当前调用已不再访问的像素缓冲
+// 入参: pixels 当前调用借用的缓冲
+func (c *pdfCompositor) releasePixels(pixels []pdfCompositePixel) {
+	if c.scratch != nil {
+		c.scratch.pixels.release(pixels)
+	}
+}
+
+// releaseMasks 在当前块绘制结束后归还蒙版结果，不复用其他块的坐标缓存
+func (c *pdfCompositor) releaseMasks() {
+	if c.scratch != nil {
+		for _, mask := range c.masks {
+			c.scratch.masks.release(mask)
+		}
+	}
+	clear(c.masks)
+}
@@ -58,7 +58,7 @@ func (p *pdfImporter) embeddedTextFont(source *pdfgo.Font, glyphs []pdfgo.Glyph)
 		return nil
 	}
 	for _, glyph := range glyphs {
-		if utf8.RuneCountInString(glyph.Text) != 1 {
+		if glyph.Name == ".notdef" || utf8.RuneCountInString(glyph.Text) != 1 {
 			return nil
 		}
 		char, _ := utf8.DecodeRuneInString(glyph.Text)
@@ -297,7 +297,7 @@ func (p *pdfImporter) groupPaths(mark pdfgo.GroupMark, walk func(pdfgo.Visitor)
 			return err
 		}
 		object := PathObject{Boundary: pdfBoundary(box), AbbreviatedData: data, Fill: &fill, Stroke: &stroke, FillColor: p.color(region.paint), Clips: objectClips}
-		p.objects = append(p.objects, GraphicObject{Type: "PathObject", PathObject: object})
+		p.appendObject(GraphicObject{Type: "PathObject", PathObject: object})
 		p.report.PathObjects++
 	}
 	return nil
@@ -358,6 +358,7 @@ func (p *pdfImporter) compileGroup(build func(*pdfImporter) error) (*RasterPage,
 	editor.SetRenderBackends(p.editor.Backends())
 	editor.SetFontDirs(p.editor.fontDirs...)
 	editor.SetFontFS(p.editor.fontFS...)
+	editor.fontSourcesCache = p.renderer.fontSourcesCache
 	if _, err := editor.AddPage(p.pageWidth, p.pageHeight); err != nil {
 		return nil, box, err
 	}
@@ -377,7 +378,7 @@ func (p *pdfImporter) compileGroup(build func(*pdfImporter) error) (*RasterPage,
 	if err := local.flushPath(); err != nil {
 		return nil, box, err
 	}
-	if len(local.objects) == 0 {
+	if local.objects.len() == 0 {
 		return nil, box, nil
 	}
 	if err := local.commitObjects(); err != nil {
@@ -391,17 +392,23 @@ func (p *pdfImporter) compileGroup(build func(*pdfImporter) error) (*RasterPage,
 	page := &editor.pages[0]
 	renderer := p.renderer.childRenderer(reader)
 	renderer.TransparentBackground = true
-	for _, layer := range page.Content.Layer {
-		for _, object := range layer.Objects {
-			if err := p.ctx.Err(); err != nil {
-				return nil, box, err
-			}
-			bounds, err := renderer.ObjectBounds(object, layer.DrawParam)
-			if err != nil {
-				return nil, box, err
-			}
-			box = unionTextBox(box, bounds)
-		}
+	scene, err := renderer.CompilePage(page)
+	if canceled := p.ctx.Err(); canceled != nil {
+		return nil, box, canceled
+	}
+	if err != nil {
+		return nil, box, err
+	}
+	if scene == nil {
+		return nil, box, fmt.Errorf("PDF local compositing: invalid compiled scene")
+	}
+	geometry, err := renderer.Geometry()
+	if err != nil {
+		return nil, box, err
+	}
+	box, err = scene.BoundsContext(p.ctx, geometry)
+	if err != nil {
+		return nil, Box{}, err
 	}
 	if box.W <= 0 || box.H <= 0 {
 		return nil, box, nil
@@ -413,11 +420,7 @@ func (p *pdfImporter) compileGroup(build func(*pdfImporter) error) (*RasterPage,
 	if box.W <= 0 || box.H <= 0 {
 		return nil, box, nil
 	}
-	scene, err := renderer.CompilePage(page)
-	if canceled := p.ctx.Err(); canceled != nil {
-		return nil, box, canceled
-	}
-	return scene, box, err
+	return scene, box, nil
 }

 // renderGroupScene 在指定区域渲染已编译场景，不修改缓存的命令与裁剪
@@ -466,7 +469,7 @@ func (p *pdfImporter) appendRasterGroup(img image.Image, box Box, opacity float6
 	}
 	alpha := int(math.Round(opacity * 255))
 	object := ImageObject{ResourceID: id, Boundary: pdfBoundary(box), CTM: pdfNumbers(box.W, 0, 0, box.H, 0, 0), Alpha: &alpha}
-	p.objects = append(p.objects, GraphicObject{Type: "ImageObject", ImageObject: object})
+	p.appendObject(GraphicObject{Type: "ImageObject", ImageObject: object})
 	p.report.ImageObjects++
 	if !p.rasterWarned {
 		p.warning(pdfgo.Diagnostic{Message: fmt.Sprintf("PDF effect locally composited at %g DPI; other objects retained", p.rasterDPI)})
@@ -659,7 +662,7 @@ func (p *pdfImporter) maskClip(mask *pdfgo.SoftMask) (pdfgo.Path, error) {
 		region, err = geometry.Combine(region, shape, operation)
 		return err
 	}
-	if err := mask.Walk(visitor); err != nil {
+	if err := mask.WalkContext(p.ctx, visitor); err != nil {
 		return pdfgo.Path{}, err
 	}
 	path := pdfgo.Path{}
@@ -18,7 +18,6 @@ import (
 	"errors"
 	"fmt"
 	"image"
-	"image/color"
 	"math"

 	"github.com/xiaoqidun/pdfgo"
@@ -36,6 +35,9 @@ func (p *pdfImporter) rasterMaskAllowed(err error) bool {
 // 入参: source 源图像, box 页面区域, mask PDF软蒙版
 // 返回: image.Image 合成图像, error 蒙版解释或渲染错误
 func (p *pdfImporter) applySoftMask(source image.Image, box Box, mask *pdfgo.SoftMask) (image.Image, error) {
+	if err := p.ctx.Err(); err != nil {
+		return nil, err
+	}
 	inverse, ok := p.matrix.Inverse()
 	if !ok {
 		return nil, fmt.Errorf("invalid PDF mask transform")
@@ -44,20 +46,31 @@ func (p *pdfImporter) applySoftMask(source image.Image, box Box, mask *pdfgo.Sof
 	w, h := out.Rect.Dx(), out.Rect.Dy()
 	step := 25.4 / p.rasterDPI
 	cache := p.compositingCache()
-	for y := 0; y < h; y += 256 {
-		for x := 0; x < w; x += 256 {
-			width, height := min(256, w-x), min(256, h-y)
-			c := pdfCompositor{importer: p, box: Box{X: box.X + float64(x)*step, Y: box.Y + float64(y)*step, W: float64(width) * step, H: float64(height) * step}, width: width, height: height, inverse: inverse, cache: cache, masks: map[*pdfgo.SoftMask][]float64{}}
-			alpha, err := c.mask(mask, &pdfgo.ColorSpace{Model: "DeviceRGB"})
+	scratch := &pdfCompositeScratch{}
+	space := &pdfgo.ColorSpace{Model: "DeviceRGB"}
+	for y := 0; y < h; y += pdfCompositeTileSize {
+		for x := 0; x < w; x += pdfCompositeTileSize {
+			if err := p.ctx.Err(); err != nil {
+				return nil, err
+			}
+			width, height := min(pdfCompositeTileSize, w-x), min(pdfCompositeTileSize, h-y)
+			c := pdfCompositor{importer: p, box: Box{X: box.X + float64(x)*step, Y: box.Y + float64(y)*step, W: float64(width) * step, H: float64(height) * step}, width: width, height: height, inverse: inverse, cache: cache, scratch: scratch, masks: map[*pdfgo.SoftMask][]float64{}}
+			alpha, err := c.mask(mask, space)
 			if err != nil {
 				return nil, err
 			}
 			for i, opacity := range alpha {
+				if i%width == 0 {
+					if err := p.ctx.Err(); err != nil {
+						return nil, err
+					}
+				}
 				px, py := out.Rect.Min.X+x+i%width, out.Rect.Min.Y+y+i/width
-				pixel := color.NRGBAModel.Convert(source.At(px, py)).(color.NRGBA)
+				pixel := imageNRGBAAt(source, px, py)
 				pixel.A = uint8(math.Round(float64(pixel.A) * opacity))
 				out.SetNRGBA(px, py, pixel)
 			}
+			c.releaseMasks()
 		}
 	}
 	return out, nil
@@ -28,10 +28,11 @@ import (
 	"github.com/xiaoqidun/pdfgo"
 )

-// pdfImageKey 区分同一图像流在不同渲染意图下的资源
+// pdfImageKey 区分同一图像流的渲染意图及模板填充色
 type pdfImageKey struct {
 	stream *pdfgo.Stream
 	intent pdfgo.Name
+	tint   color.NRGBA64
 }

 // pdfImageResource 保存图像的有效颜色空间及对应资源，避免跨资源重映射复用
@@ -84,6 +85,9 @@ func (s *pdfStencilImage) NRGBA64At(x, y int) color.NRGBA64 {
 // 入参: mark PDF图像绘制信息
 // 返回: error 错误信息
 func (p *pdfImporter) image(mark pdfgo.ImageMark) error {
+	if err := p.ctx.Err(); err != nil {
+		return err
+	}
 	if mark.Style.BlendMode != "" && mark.Style.BlendMode != "Normal" && mark.Style.BlendMode != "Compatible" {
 		return &pdfgo.UnsupportedError{Feature: "image blend mode " + string(mark.Style.BlendMode)}
 	}
@@ -111,14 +115,28 @@ func (p *pdfImporter) image(mark pdfgo.ImageMark) error {
 	}
 	source := mark.Image
 	key := pdfImageKey{stream: source.Stream, intent: source.Intent}
-	id := ""
-	if !source.ImageMask {
-		for _, resource := range p.imageIDs[key] {
-			if reflect.DeepEqual(resource.space, source.ColorSpace) {
-				id = resource.id
-				break
+	if source.ImageMask {
+		fill := mark.Style.Fill.RGB
+		if values := mark.Style.Fill.CMYK; values != nil {
+			space := &pdfgo.ColorSpace{Model: "DeviceCMYK"}
+			fill, err = space.RGB(values[:], mark.Style.RenderingIntent)
+			if err != nil {
+				return err
 			}
 		}
+		key.tint = color.NRGBA64{
+			R: uint16(math.Round(fill[0] * 65535)),
+			G: uint16(math.Round(fill[1] * 65535)),
+			B: uint16(math.Round(fill[2] * 65535)),
+			A: 65535,
+		}
+	}
+	id := ""
+	for _, resource := range p.imageIDs[key] {
+		if reflect.DeepEqual(resource.space, source.ColorSpace) {
+			id = resource.id
+			break
+		}
 	}
 	if id == "" {
 		jbig2Original, err := source.JBIG2FileContext(p.ctx)
@@ -137,41 +155,27 @@ func (p *pdfImporter) image(mark pdfgo.ImageMark) error {
 			}
 		}
 		if source.ImageMask {
-			fill := mark.Style.Fill.RGB
-			if values := mark.Style.Fill.CMYK; values != nil {
-				space := &pdfgo.ColorSpace{Model: "DeviceCMYK"}
-				fill, err = space.RGB(values[:], mark.Style.RenderingIntent)
-				if err != nil {
-					return err
-				}
-			}
-			tint := color.NRGBA64{
-				R: uint16(math.Round(fill[0] * 65535)),
-				G: uint16(math.Round(fill[1] * 65535)),
-				B: uint16(math.Round(fill[2] * 65535)),
-			}
-			decoded = &pdfStencilImage{source: decoded, fill: tint}
+			decoded = &pdfStencilImage{source: decoded, fill: key.tint}
 		}
-		var encoded bytes.Buffer
-		if len(jbig2Original) != 0 {
-			encoded.Write(jbig2Original)
-		} else if len(jpegOriginal) != 0 {
-			encoded.Write(jpegOriginal)
-		} else {
+		data := jbig2Original
+		if len(data) == 0 {
+			data = jpegOriginal
+		}
+		if len(data) == 0 {
+			var encoded bytes.Buffer
 			if err := pdfgo.EncodePNG(p.ctx, &encoded, decoded); err != nil {
 				return err
 			}
+			data = encoded.Bytes()
 		}
-		id, err = p.editor.AddImage(encoded.Bytes())
+		id, err = p.editor.AddImage(data)
 		if err != nil {
 			return err
 		}
-		if !source.ImageMask {
-			if p.imageIDs == nil {
-				p.imageIDs = map[pdfImageKey][]pdfImageResource{}
-			}
-			p.imageIDs[key] = append(p.imageIDs[key], pdfImageResource{space: source.ColorSpace, id: id})
+		if p.imageIDs == nil {
+			p.imageIDs = map[pdfImageKey][]pdfImageResource{}
 		}
+		p.imageIDs[key] = append(p.imageIDs[key], pdfImageResource{space: source.ColorSpace, id: id})
 	}
 	return p.appendImage(mark, id)
 }
@@ -188,7 +192,7 @@ func (p *pdfImporter) appendImage(mark pdfgo.ImageMark, id string) error {
 		return err
 	}
 	object := ImageObject{Boundary: pdfBoundary(box), ResourceID: id, CTM: pdfNumbers(m[0], m[1], m[2], m[3], m[4]-box.X, m[5]-box.Y), Alpha: &alpha, Clips: clips}
-	p.objects = append(p.objects, GraphicObject{Type: "ImageObject", ImageObject: object})
+	p.appendObject(GraphicObject{Type: "ImageObject", ImageObject: object})
 	p.report.ImageObjects++
 	return nil
 }
@@ -240,7 +244,8 @@ func (p *pdfImporter) text(mark pdfgo.TextMark) error {
 		if mark.Clip == nil && p.rasterMaskAllowed(err) {
 			mask := mark.Style.SoftMask
 			mark.Style.SoftMask = nil
-			return p.rasterGroup(pdfgo.GroupMark{Alpha: 1, SoftMask: mask}, func(v pdfgo.Visitor) error { return v.Text(mark) })
+			masked := mark
+			return p.rasterGroup(pdfgo.GroupMark{Alpha: 1, SoftMask: mask}, func(v pdfgo.Visitor) error { return v.Text(masked) })
 		}
 		return err
 	}
@@ -249,7 +254,8 @@ func (p *pdfImporter) text(mark pdfgo.TextMark) error {
 		return &pdfgo.UnsupportedError{Feature: "text color separation overprint"}
 	}
 	if p.warning != nil && ((paintMode == 0 || paintMode == 2) && pdfGradientError(mark.Style.Fill) != nil || (paintMode == 1 || paintMode == 2) && pdfGradientError(mark.Style.Stroke) != nil) {
-		return p.compositeRegion(nil, pdfCompositeNode{text: &mark}, &pdfgo.ColorSpace{Model: "DeviceRGB"}, false)
+		gradient := mark
+		return p.compositeRegion(nil, pdfCompositeNode{text: &gradient}, &pdfgo.ColorSpace{Model: "DeviceRGB"}, false)
 	}
 	font := mark.Font
 	embedded := len(font.Program) != 0
@@ -311,11 +317,18 @@ func (p *pdfImporter) text(mark pdfgo.TextMark) error {
 		outline = metrics.(FontOutlines)
 	}
 	fill, stroke, visible := paintMode == 0 || paintMode == 2, paintMode == 1 || paintMode == 2, paintMode != 3
-	if stroke && !embedded {
+	needOutline := stroke
+	for _, glyph := range mark.Glyphs {
+		needOutline = needOutline || glyph.Name == ".notdef"
+	}
+	if needOutline && !embedded {
 		if resolved, err := p.editor.editorFont(id); err == nil {
 			if provider, ok := resolved.(FontOutlines); ok {
 				metrics, outline = resolved, provider
 				for _, glyph := range mark.Glyphs {
+					if glyph.Name == ".notdef" {
+						continue
+					}
 					char, _ := utf8.DecodeRuneInString(glyph.Text)
 					if utf8.RuneCountInString(glyph.Text) != 1 || resolved.GlyphIndex(char) == 0 && char != 0 {
 						metrics, outline = nil, nil
@@ -329,7 +342,7 @@ func (p *pdfImporter) text(mark pdfgo.TextMark) error {
 	var glyphPaths []pdfgo.Path
 	const unit = 25.4 / 72
 	m := p.matrix.Mul(mark.Matrix).Mul(pdfgo.Matrix{1 / unit, 0, 0, -1 / unit, 0, 0})
-	points := make([]pdfgo.Point, 0, 4*len(mark.Glyphs))
+	var extent pdfPointBounds
 	object := TextObject{Font: id, Size: mark.Size * unit, HScale: mark.HorizontalScale, TextCode: make([]TextCode, 0, len(mark.Glyphs))}
 	if embedded {
 		object.CGTransform = make([]CGTransform, 0, len(mark.Glyphs))
@@ -340,16 +353,16 @@ func (p *pdfImporter) text(mark pdfgo.TextMark) error {
 			return err
 		}
 		gid := glyph.ID
+		undefined := glyph.Name == ".notdef"
+		if !embedded && undefined {
+			gid = 0
+		}
 		if embedded && !reused && font.ProgramType == "FontFile" {
 			gid = imported.type1Glyphs[glyph.Name]
 			glyph.HasID = true
 		}
-		if glyph.Text == "" && glyph.HasID {
-			character := getUnicodeFromName(glyph.Name)
-			if character == 0 {
-				character = packedGlyphRune(gid)
-			}
-			glyph.Text = string(character)
+		if glyph.Text == "" && (glyph.HasID || undefined) {
+			glyph.Text = "\u00a4"
 		}
 		if embedded && !glyph.HasID {
 			if utf8.RuneCountInString(glyph.Text) != 1 {
@@ -367,18 +380,18 @@ func (p *pdfImporter) text(mark pdfgo.TextMark) error {
 		if !embedded && glyph.Text == "" {
 			return &pdfgo.UnsupportedError{Feature: "unmapped external font character"}
 		}
-		if outlineStroke && !embedded {
+		if outlineStroke && !embedded && !undefined {
 			char, _ := utf8.DecodeRuneInString(glyph.Text)
 			gid = metrics.GlyphIndex(char)
 		}
 		origin := mark.Positions[n]
 		object.TextCode = append(object.TextCode, TextCode{X: pdfNumbers(origin.X * unit), Y: pdfNumbers(-origin.Y * unit), Value: glyph.Text})
-		if embedded {
-			count := utf8.RuneCountInString(glyph.Text)
+		count := utf8.RuneCountInString(glyph.Text)
+		if embedded || undefined {
 			object.CGTransform = append(object.CGTransform, CGTransform{CodePosition: position, CodeCount: count, GlyphCount: 1, Glyphs: strconv.Itoa(int(gid))})
-			position += count
 		}
-		if embedded || outlineStroke {
+		position += count
+		if embedded || outlineStroke || undefined && outline != nil {
 			var path GeometryPath
 			var bounds Box
 			var err error
@@ -408,7 +421,7 @@ func (p *pdfImporter) text(mark pdfgo.TextMark) error {
 				}
 			}
 			for _, point := range []pdfgo.Point{{X: bounds.X, Y: bounds.Y}, {X: bounds.X + bounds.W, Y: bounds.Y}, {X: bounds.X, Y: bounds.Y + bounds.H}, {X: bounds.X + bounds.W, Y: bounds.Y + bounds.H}} {
-				points = append(points, m.Apply(pdfgo.Point{X: origin.X*unit + point.X*mark.HorizontalScale, Y: -origin.Y*unit + point.Y}))
+				extent.add(m.Apply(pdfgo.Point{X: origin.X*unit + point.X*mark.HorizontalScale, Y: -origin.Y*unit + point.Y}))
 			}
 			if outlineStroke {
 				matrix := mark.Matrix.Mul(pdfgo.Matrix{mark.HorizontalScale / unit, 0, 0, -1 / unit, origin.X, origin.Y})
@@ -420,12 +433,22 @@ func (p *pdfImporter) text(mark pdfgo.TextMark) error {
 			}
 		} else {
 			width := glyph.Width / 1000 * object.Size * mark.HorizontalScale
-			for _, point := range []pdfgo.Point{{X: 0, Y: -object.Size}, {X: width, Y: -object.Size}, {X: 0, Y: object.Size / 4}, {X: width, Y: object.Size / 4}} {
-				points = append(points, m.Apply(pdfgo.Point{X: origin.X*unit + point.X, Y: -origin.Y*unit + point.Y}))
+			bounds := Box{X: 0, Y: -object.Size, W: width, H: object.Size * 1.25}
+			if undefined {
+				b := font.BoundingBox
+				if b == nil {
+					b = font.FallbackBoundingBox
+				}
+				if b != nil {
+					bounds = Box{X: b.XMin / 1000 * object.Size * mark.HorizontalScale, Y: -b.YMax / 1000 * object.Size, W: (b.XMax - b.XMin) / 1000 * object.Size * mark.HorizontalScale, H: (b.YMax - b.YMin) / 1000 * object.Size}
+				}
+			}
+			for _, point := range []pdfgo.Point{{X: bounds.X, Y: bounds.Y}, {X: bounds.X + bounds.W, Y: bounds.Y}, {X: bounds.X, Y: bounds.Y + bounds.H}, {X: bounds.X + bounds.W, Y: bounds.Y + bounds.H}} {
+				extent.add(m.Apply(pdfgo.Point{X: origin.X*unit + point.X, Y: -origin.Y*unit + point.Y}))
 			}
 		}
 	}
-	box := pdfBounds(points)
+	box := extent.box()
 	if stroke && !outlineStroke {
 		strokeMatrix := mark.StrokeMatrix
 		if strokeMatrix == (pdfgo.Matrix{}) {
@@ -508,7 +531,7 @@ func (p *pdfImporter) text(mark pdfgo.TextMark) error {
 		object.Visible = new(bool)
 		object.Fill, object.Stroke = new(bool), new(bool)
 	}
-	p.objects = append(p.objects, GraphicObject{Type: "TextObject", TextObject: object})
+	p.appendObject(GraphicObject{Type: "TextObject", TextObject: object})
 	p.report.TextObjects++
 	return nil
 }
@@ -68,21 +68,10 @@ func (p *pdfImporter) disjointText(node pdfCompositeNode) (bool, error) {
 		}
 		var box Box
 		for _, command := range scene.Commands {
-			path := geometryFromRaster(command.Path)
-			if command.Image != nil {
-				b := command.Image.Bounds()
-				path = geometryRectangle(Box{X: float64(b.Min.X), Y: float64(b.Min.Y), W: float64(b.Dx()), H: float64(b.Dy())})
-			}
-			path, err = geometry.Transform(path, MatrixFromValues([6]float64(command.Transform)))
+			path, err := rasterCommandGeometry(geometry, command)
 			if err != nil {
 				return false, err
 			}
-			if command.Stroke != nil {
-				path, err = geometry.Stroke(path, *command.Stroke)
-				if err != nil {
-					return false, err
-				}
-			}
 			path, err = geometry.Transform(path, MatrixFromValues([6]float64(inverse)))
 			if err != nil {
 				return false, err
@@ -102,14 +102,17 @@ type RasterSpread struct {
 	Period  float64
 }

-// PixelSize 检查页面尺寸并计算目标像素大小
+// PixelSize 检查页面尺寸并计算目标像素大小，合法正尺寸至少保留一个像素
 // 返回: int 宽度, int 高度, error 无效尺寸
 func (p *RasterPage) PixelSize() (int, int, error) {
 	if p == nil || !rasterPositive(p.Width) || !rasterPositive(p.Height) || !rasterPositive(p.DPI) {
 		return 0, 0, fmt.Errorf("invalid raster page size or DPI")
 	}
 	scale := p.DPI / 25.4
-	w, h := math.Floor(p.Width*scale+0.5), math.Floor(p.Height*scale+0.5)
+	if !rasterPositive(scale) {
+		return 0, 0, fmt.Errorf("invalid raster page size or DPI")
+	}
+	w, h := math.Max(1, math.Floor(p.Width*scale+0.5)), math.Max(1, math.Floor(p.Height*scale+0.5))
 	if !rasterPositive(w) || !rasterPositive(h) || w*h > math.MaxInt32/4 {
 		return 0, 0, fmt.Errorf("invalid raster pixel dimensions: %g x %g", w, h)
 	}
@@ -0,0 +1,121 @@
+// Copyright 2025-2026 肖其顿 (XIAO QI DUN)
+//
+// 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 ofdgo
+
+import (
+	"context"
+	"fmt"
+)
+
+// Bounds 度量已编译指令的页面坐标范围，不重新定位文字或读取图片像素
+// 包含描边与指令裁剪，不应用页面边界，不排除图片透明像素或渐变透明区域
+// 入参: geometry 几何后端
+// 返回: Box 毫米坐标范围, error 几何或指令错误
+func (p *RasterPage) Bounds(geometry GeometryBackend) (Box, error) {
+	return p.BoundsContext(context.Background(), geometry)
+}
+
+// BoundsContext 按上下文度量只读绘制指令，取消或失败时不返回部分范围
+// 入参: ctx 取消上下文, geometry 几何后端
+// 返回: Box 毫米坐标范围, error 取消、几何或指令错误
+func (p *RasterPage) BoundsContext(ctx context.Context, geometry GeometryBackend) (Box, error) {
+	if err := ctx.Err(); err != nil {
+		return Box{}, err
+	}
+	if p == nil {
+		return Box{}, fmt.Errorf("invalid raster page")
+	}
+	if geometry == nil {
+		return Box{}, fmt.Errorf("raster bounds: %w", ErrBackendUnavailable)
+	}
+	var box Box
+	for index, command := range p.Commands {
+		if err := ctx.Err(); err != nil {
+			return Box{}, err
+		}
+		path, err := rasterCommandGeometry(geometry, command)
+		if err == nil {
+			var bounds Box
+			bounds, err = geometry.Bounds(path)
+			box = unionTextBox(box, bounds)
+		}
+		if err != nil {
+			return Box{}, fmt.Errorf("raster command %d bounds: %w", index, err)
+		}
+	}
+	if err := ctx.Err(); err != nil {
+		return Box{}, err
+	}
+	return box, nil
+}
+
+// rasterCommandGeometry 共用指令的页面变换、描边与裁剪，不修改路径或样式
+// 入参: geometry 几何后端, command 绘制指令
+// 返回: GeometryPath 页面轮廓, error 几何或指令错误
+func rasterCommandGeometry(geometry GeometryBackend, command RasterCommand) (GeometryPath, error) {
+	if command.Clip != nil && len(command.Clip) == 0 {
+		return nil, nil
+	}
+	var path GeometryPath
+	if command.Image != nil {
+		bounds := command.Image.Bounds()
+		if bounds.Empty() {
+			return nil, nil
+		}
+		path = geometryRectangle(Box{X: float64(bounds.Min.X), Y: float64(bounds.Min.Y), W: float64(bounds.Dx()), H: float64(bounds.Dy())})
+	} else {
+		if len(command.Path) == 0 || command.Paint.Gradient == nil && command.Paint.Color.A == 0 {
+			return nil, nil
+		}
+		for _, segment := range command.Path {
+			if segment.Verb > RasterClose {
+				return nil, fmt.Errorf("unsupported raster path verb %d", segment.Verb)
+			}
+		}
+		path = geometryFromRaster(command.Path)
+	}
+	matrix := MatrixFromValues([6]float64(command.Transform))
+	path, err := geometry.Transform(path, matrix)
+	if err != nil {
+		return nil, err
+	}
+	if command.Image == nil && command.Stroke != nil {
+		if err := validateStroke(*command.Stroke); err != nil {
+			return nil, err
+		}
+		path, err = geometry.Stroke(path, *command.Stroke)
+		if err != nil {
+			return nil, err
+		}
+	} else if command.EvenOdd && command.Clip != nil {
+		path, err = geometry.Normalize(path, true)
+		if err != nil {
+			return nil, err
+		}
+	}
+	if command.Clip != nil {
+		for _, segment := range command.Clip {
+			if segment.Verb > RasterClose {
+				return nil, fmt.Errorf("unsupported raster clip verb %d", segment.Verb)
+			}
+		}
+		clip := geometryFromRaster(command.Clip)
+		path, err = clipGeometry(geometry, path, &clip)
+		if err != nil {
+			return nil, err
+		}
+	}
+	return path, nil
+}
@@ -524,20 +524,40 @@ func validPackagePath(name string) bool {
 // 入参: name 文件名
 // 返回: []byte 文件内容, error 错误信息
 func (r *Reader) readFile(name string) ([]byte, error) {
+	data, borrowed, err := r.readFileData(name)
+	if borrowed {
+		data = bytes.Clone(data)
+	}
+	return data, err
+}
+
+// readFileView 借用已加载文件或读取压缩文件，返回数据不得修改
+// 入参: name 文件名
+// 返回: []byte 只读文件内容, error 错误信息
+func (r *Reader) readFileView(name string) ([]byte, error) {
+	data, _, err := r.readFileData(name)
+	return data, err
+}
+
+// readFileData 统一解析包内路径并区分缓存借用与独立读取
+// 入参: name 文件名
+// 返回: []byte 文件内容, bool 是否借用缓存, error 错误信息
+func (r *Reader) readFileData(name string) ([]byte, bool, error) {
 	name = cleanPackagePath(name)
 	if actual, ok := r.fileNamesFold[strings.ToLower(name)]; ok {
 		name = actual
 	}
 	if data, ok := r.files[name]; ok {
-		return bytes.Clone(data), nil
+		return data, true, nil
 	}
 	if f, ok := r.packageFile(name); ok {
 		if data, ok := r.files[cleanPackagePath(f.Name)]; ok {
-			return bytes.Clone(data), nil
+			return data, true, nil
 		}
-		return readZipFile(f)
+		data, err := readZipFile(f)
+		return data, false, err
 	}
-	return nil, fmt.Errorf("file not found: %s", name)
+	return nil, false, fmt.Errorf("file not found: %s", name)
 }

 // openFile 打开文档内的文件流
@@ -15,7 +15,6 @@
 package ofdgo

 import (
-	"crypto/sha256"
 	"fmt"
 	"math"
 	"strings"
@@ -262,7 +261,7 @@ func (r *Renderer) PositionText(object TextObject, state RenderState) (*Position
 // 返回: FontMetrics 字体度量, error 字体解析错误
 func (r *Renderer) preparedMetrics(prepared *PreparedFont) (FontMetrics, error) {
 	if prepared.digest == ([32]byte{}) {
-		prepared.digest = sha256.Sum256(prepared.Data)
+		prepared.digest = r.fontDigest(prepared.Data)
 	}
 	key := prepared.digest
 	if metrics, ok := r.fontMetrics.get(key); ok {
@@ -49,6 +49,7 @@ type Renderer struct {
 	sharedBackendStates   map[any]any
 	resolvedFonts         map[resolvedFontKey]resolvedFontResult
 	preparedFonts         map[string]*PreparedFont
+	fontDigests           *renderCache[fontDataKey, [32]byte]
 	fontPreparations      *renderCache[[32]byte, PreparedFont]
 	fontSourcesCache      *fontSourceCache
 	fontMetrics           *renderCache[[32]byte, FontMetrics]
@@ -84,7 +85,7 @@ func (r *Renderer) FontSources() ([]string, []fs.FS) {
 	return append([]string(nil), r.fontDirs...), append([]fs.FS(nil), r.fontFS...)
 }

-// childRenderer 创建继承当前配置的子渲染器
+// childRenderer 继承配置并复用字体文件索引与数据，文档匹配结果保持独立
 // 入参: reader 子阅读器
 // 返回: *Renderer 子渲染器
 func (r *Renderer) childRenderer(reader *Reader) *Renderer {
@@ -102,7 +103,9 @@ func (r *Renderer) childRenderer(reader *Reader) *Renderer {
 	renderer := NewRenderer(reader, opts...)
 	renderer.sharedBackendStates = r.sharedBackendStates
 	renderer.fontPreparations = r.fontPreparations
+	renderer.fontDigests = r.fontDigests
 	renderer.fontMetrics = r.fontMetrics
+	renderer.fontSourcesCache = r.fontSourcesCache.child()
 	renderer.decodeImages = r.decodeImages
 	renderer.TransparentBackground = r.TransparentBackground
 	return renderer
@@ -14,11 +14,7 @@

 package ofdgo

-import (
-	"crypto/sha256"
-
-	"github.com/tdewolff/canvas"
-)
+import "github.com/tdewolff/canvas"

 // loadFont 加载字体
 // 入参: fontID 字体ID
@@ -42,7 +38,7 @@ func (r *Renderer) loadFont(fontID string) *canvas.FontFamily {
 	fontData := resolved.Data
 	fontStyle := canvasFontStyle(of)
 	if resolved.digest == ([32]byte{}) {
-		resolved.digest = sha256.Sum256(fontData)
+		resolved.digest = r.fontDigest(fontData)
 	}
 	key := canvasFontKey{digest: resolved.digest, name: of.FontName, style: fontStyle}
 	cache := r.canvasFontFamilies()
@@ -67,6 +67,16 @@ func (CanvasBackend) Bounds(path GeometryPath) (Box, error) {
 // 入参: path 页面路径, matrix 页面变换
 // 返回: GeometryPath 新路径, error 无效路径错误
 func (CanvasBackend) Transform(path GeometryPath, matrix Matrix) (GeometryPath, error) {
+	curves := true
+	for _, segment := range path {
+		if segment.Verb == GeometryArc {
+			curves = false
+			break
+		}
+	}
+	if curves {
+		return path.Transform(matrix)
+	}
 	p, err := geometryToCanvasPath(&path)
 	if err != nil {
 		return nil, err
@@ -282,7 +282,7 @@ func shapedLineWidth(runes []rune, advances []float64, spacing float64) float64
 // validateTextGlyphs 校验显式字形编号及原文区间，不允许跨TextCode或重叠映射
 // 入参: obj 文字对象, glyphCount 字体字形数量
 // 返回: error 无效映射
-func validateTextGlyphs(obj TextObject, glyphCount uint16) error {
+func validateTextGlyphs(obj TextObject, glyphCount int) error {
 	if len(obj.CGTransform) == 0 {
 		return nil
 	}
@@ -17,6 +17,7 @@ package ofdgo
 import (
 	"archive/zip"
 	"bytes"
+	"context"
 	"encoding/hex"
 	"encoding/xml"
 	"fmt"
@@ -43,8 +44,13 @@ type ofdCountingWriter struct {

 // ofdXML 使用标准ofd命名空间写出XML
 type ofdXML struct {
-	encoder *xml.Encoder
-	err     error
+	encoder    *xml.Encoder
+	references *editorReferenceScan
+	ctx        context.Context
+	attrs      ofdAttrs
+	escaped    bytes.Buffer
+	spaces     bytes.Buffer
+	err        error
 }

 // ofdAttrs OFD节点属性
@@ -316,7 +322,7 @@ func (e *Editor) writePartsWithPages(write func(string, []byte, bool) error, pro
 	}
 	fonts, images, spaces, definitions := e.usedResources()
 	writeXML := func(name string, encode func(*ofdXML)) error {
-		data, err := encodeOFDXML(encode)
+		data, err := e.encodeXML(encode)
 		if err != nil {
 			return err
 		}
@@ -538,6 +544,9 @@ func (x *ofdXML) page(page PageContent) {
 	x.end("Area")
 	x.start("Content", nil)
 	for _, layer := range page.Content.Layer {
+		if x.err != nil {
+			return
+		}
 		x.layer(layer)
 	}
 	x.end("Content")
@@ -554,6 +563,9 @@ func (x *ofdXML) layer(layer Layer) {
 	attrs.add("DrawParam", layer.DrawParam)
 	x.start("Layer", attrs)
 	for _, object := range layer.Objects {
+		if x.err != nil {
+			return
+		}
 		x.object(object, false)
 	}
 	x.end("Layer")
@@ -566,6 +578,9 @@ func (x *ofdXML) resources(fonts, images []editorResource, spaces []ColorSpace)
 	if len(spaces) != 0 {
 		x.start("ColorSpaces", nil)
 		for _, space := range spaces {
+			if x.err != nil {
+				return
+			}
 			x.start("ColorSpace", ofdAttrs{{Name: xml.Name{Local: "ID"}, Value: space.ID}, {Name: xml.Name{Local: "Type"}, Value: space.Type}, {Name: xml.Name{Local: "BitsPerComponent"}, Value: strconv.Itoa(space.BitsPerComponent)}})
 			x.end("ColorSpace")
 		}
@@ -574,6 +589,9 @@ func (x *ofdXML) resources(fonts, images []editorResource, spaces []ColorSpace)
 	if len(fonts) != 0 {
 		x.start("Fonts", nil)
 		for _, resource := range fonts {
+			if x.err != nil {
+				return
+			}
 			font := resource.font
 			var attrs ofdAttrs
 			attrs.add("ID", font.ID)
@@ -600,6 +618,9 @@ func (x *ofdXML) resources(fonts, images []editorResource, spaces []ColorSpace)
 	if len(images) != 0 {
 		x.start("MultiMedias", nil)
 		for _, resource := range images {
+			if x.err != nil {
+				return
+			}
 			image := resource.image
 			var attrs ofdAttrs
 			attrs.add("ID", image.ID)
@@ -625,11 +646,21 @@ func newOFDXML(writer io.Writer) *ofdXML {
 // token 写入XML标记并保留首个错误
 // 入参: token XML标记
 func (x *ofdXML) token(token xml.Token) {
-	if x.err == nil {
+	if x.checkContext() {
 		x.err = x.encoder.EncodeToken(token)
+		x.referenceToken(token)
 	}
 }

+// checkContext 保留首个编码或取消错误
+// 返回: bool 是否可以继续编码
+func (x *ofdXML) checkContext() bool {
+	if x.err == nil && x.ctx != nil {
+		x.err = x.ctx.Err()
+	}
+	return x.err == nil
+}
+
 // root 写入声明命名空间的根节点
 // 入参: name 节点名, attrs 属性
 func (x *ofdXML) root(name string, attrs ofdAttrs) {
@@ -660,7 +691,7 @@ func (x *ofdXML) text(name, value string) {
 // finish 完成XML输出
 // 返回: error 错误信息
 func (x *ofdXML) finish() error {
-	if x.err != nil {
+	if !x.checkContext() {
 		return x.err
 	}
 	return x.encoder.Close()
@@ -702,8 +733,30 @@ func (a *ofdAttrs) alpha(value *int) {
 // 入参: encode 节点写入方法
 // 返回: []byte XML数据, error 错误信息
 func encodeOFDXML(encode func(*ofdXML)) ([]byte, error) {
+	return encodeOFDXMLContext(nil, encode)
+}
+
+// encodeXML 按本次保存上下文编码，不改变普通快照的输出
+// 入参: encode 节点写入方法
+// 返回: []byte XML数据, error 编码或取消错误
+func (e *Editor) encodeXML(encode func(*ofdXML)) ([]byte, error) {
+	var ctx context.Context
+	if e.output != nil {
+		ctx = e.output.ctx
+	}
+	return encodeOFDXMLContext(ctx, encode)
+}
+
+// encodeOFDXMLContext 编码文档节点，失败时不返回部分数据
+// 入参: ctx 取消上下文，可为nil, encode 节点写入方法
+// 返回: []byte XML数据, error 编码或取消错误
+func encodeOFDXMLContext(ctx context.Context, encode func(*ofdXML)) ([]byte, error) {
 	var data bytes.Buffer
 	x := newOFDXML(&data)
+	x.ctx = ctx
+	if !x.checkContext() {
+		return nil, x.err
+	}
 	encode(x)
 	if err := x.finish(); err != nil {
 		return nil, err
@@ -714,7 +767,10 @@ func encodeOFDXML(encode func(*ofdXML)) ([]byte, error) {
 // object 按OFD规定顺序写出基本对象
 // 入参: object 图形对象, root 是否为根节点
 func (x *ofdXML) object(object GraphicObject, root bool) {
-	var attrs ofdAttrs
+	if !x.checkContext() {
+		return
+	}
+	attrs := x.attrs[:0]
 	var fill, stroke *FillColor
 	var actions []Action
 	if root {
@@ -799,6 +855,7 @@ func (x *ofdXML) object(object GraphicObject, root bool) {
 		attrs.alpha(obj.Alpha)
 		actions = obj.Actions
 	}
+	x.attrs = attrs
 	x.start(object.Type, attrs)
 	x.actions(actions)
 	if object.Type == "ImageObject" {
@@ -817,24 +874,31 @@ func (x *ofdXML) object(object GraphicObject, root bool) {
 		x.color("StrokeColor", stroke)
 		offset, index := 0, 0
 		for _, code := range object.TextObject.TextCode {
+			if x.err != nil {
+				return
+			}
 			end := offset + len(textCodeRunes(code.Value))
 			for index < len(object.TextObject.CGTransform) && object.TextObject.CGTransform[index].CodePosition < end {
-				transform := object.TextObject.CGTransform[index]
-				attrs := ofdAttrs{
-					{Name: xml.Name{Local: "CodePosition"}, Value: strconv.Itoa(transform.CodePosition - offset)},
-					{Name: xml.Name{Local: "CodeCount"}, Value: strconv.Itoa(transform.CodeCount)},
-					{Name: xml.Name{Local: "GlyphCount"}, Value: strconv.Itoa(transform.GlyphCount)},
+				if x.err != nil {
+					return
 				}
+				transform := object.TextObject.CGTransform[index]
+				attrs := x.attrs[:0]
+				attrs.add("CodePosition", strconv.Itoa(transform.CodePosition-offset))
+				attrs.add("CodeCount", strconv.Itoa(transform.CodeCount))
+				attrs.add("GlyphCount", strconv.Itoa(transform.GlyphCount))
+				x.attrs = attrs
 				x.start("CGTransform", attrs)
 				x.text("Glyphs", transform.Glyphs)
 				x.end("CGTransform")
 				index++
 			}
-			var attrs ofdAttrs
+			attrs := x.attrs[:0]
 			attrs.add("X", code.X)
 			attrs.add("Y", code.Y)
 			attrs.add("DeltaX", code.DeltaX)
 			attrs.add("DeltaY", code.DeltaY)
+			x.attrs = attrs
 			x.textCode(code.Value, attrs)
 			offset = end
 		}
@@ -845,23 +909,35 @@ func (x *ofdXML) object(object GraphicObject, root bool) {
 // clips 写出对象裁剪，保留区域内的路径与文字
 // 入参: clips 裁剪集合，nil不输出节点
 func (x *ofdXML) clips(clips *Clips) {
-	if clips == nil {
+	if clips == nil || !x.checkContext() {
 		return
 	}
 	var attrs ofdAttrs
 	attrs.flag("TransFlag", clips.TransFlag)
 	x.start("Clips", attrs)
 	for _, clip := range clips.Clip {
+		if x.err != nil {
+			return
+		}
 		x.start("Clip", nil)
 		for _, area := range clip.Area {
+			if x.err != nil {
+				return
+			}
 			var attrs ofdAttrs
 			attrs.add("CTM", area.CTM)
 			attrs.add("DrawParam", area.DrawParam)
 			x.start("Area", attrs)
 			for _, path := range area.Path {
+				if x.err != nil {
+					return
+				}
 				x.object(GraphicObject{Type: "Path", PathObject: path}, false)
 			}
 			for _, text := range area.Text {
+				if x.err != nil {
+					return
+				}
 				x.object(GraphicObject{Type: "Text", TextObject: text}, false)
 			}
 			x.end("Area")
@@ -874,23 +950,41 @@ func (x *ofdXML) clips(clips *Clips) {
 // textCode 使用XML字符引用转义空格，避免依赖阅读器保留原始空白
 // 入参: value OFD文字内容, attrs 文字定位属性
 func (x *ofdXML) textCode(value string, attrs ofdAttrs) {
-	if x.err != nil {
+	if !x.checkContext() {
 		return
 	}
-	var escaped bytes.Buffer
-	if x.err = xml.EscapeText(&escaped, []byte(value)); x.err != nil {
+	x.escaped.Reset()
+	if x.err = xml.EscapeText(&x.escaped, []byte(value)); x.err != nil {
 		return
 	}
+	data := x.escaped.Bytes()
+	if next := bytes.IndexByte(data, ' '); next >= 0 {
+		x.spaces.Reset()
+		for replaced := 0; next >= 0; replaced++ {
+			if replaced&255 == 0 && !x.checkContext() {
+				return
+			}
+			x.spaces.Write(data[:next])
+			x.spaces.WriteString("&#x20;")
+			data = data[next+1:]
+			next = bytes.IndexByte(data, ' ')
+		}
+		x.spaces.Write(data)
+		data = x.spaces.Bytes()
+	}
 	content := struct {
 		Text []byte `xml:",innerxml"`
-	}{Text: bytes.ReplaceAll(escaped.Bytes(), []byte(" "), []byte("&#x20;"))}
+	}{Text: data}
 	x.err = x.encoder.EncodeElement(content, xml.StartElement{Name: xml.Name{Local: "ofd:TextCode"}, Attr: attrs})
+	x.referenceToken(xml.StartElement{Name: xml.Name{Local: "ofd:TextCode"}, Attr: attrs})
+	x.referenceToken(xml.CharData(value))
+	x.referenceToken(xml.EndElement{Name: xml.Name{Local: "ofd:TextCode"}})
 }

 // color 写出纯色、渐变和图案，保留颜色空间与透明度
 // 入参: name 节点名, color 颜色
 func (x *ofdXML) color(name string, color *FillColor) {
-	if color == nil {
+	if color == nil || !x.checkContext() {
 		return
 	}
 	var attrs ofdAttrs
@@ -940,6 +1034,9 @@ func (x *ofdXML) meshShading(name string, extend, columns int, points []ShdPoint
 	}
 	x.start(name, attrs)
 	for _, point := range points {
+		if x.err != nil {
+			return
+		}
 		var attrs ofdAttrs
 		attrs.add("X", ofdNumber(point.X))
 		attrs.add("Y", ofdNumber(point.Y))
@@ -974,6 +1071,9 @@ func shadingAttrs(mapType string, mapUnit float64, extend, start, end string) of
 func (x *ofdXML) shading(name string, attrs ofdAttrs, segments []ShdSegment) {
 	x.start(name, attrs)
 	for _, segment := range segments {
+		if x.err != nil {
+			return
+		}
 		var attrs ofdAttrs
 		if !segment.positionMissing {
 			attrs.add("Position", ofdNumber(segment.Position))
@@ -988,15 +1088,16 @@ func (x *ofdXML) shading(name string, attrs ofdAttrs, segments []ShdSegment) {
 // element 写出具有标准XML标签的子树，显式声明OFD命名空间
 // 入参: name 元素名称, value 元素内容
 func (x *ofdXML) element(name string, value any) {
-	if x.err == nil {
+	if x.checkContext() {
 		x.err = x.encoder.EncodeElement(value, xml.StartElement{Name: xml.Name{Space: ofdNamespace, Local: name}})
+		x.referenceElement(name, value)
 	}
 }

 // border 写出图片边框及其绘制颜色
 // 入参: border 边框，nil不输出节点
 func (x *ofdXML) border(border *ImageBorder) {
-	if border == nil {
+	if border == nil || !x.checkContext() {
 		return
 	}
 	var attrs ofdAttrs
@@ -1015,6 +1116,9 @@ func (x *ofdXML) border(border *ImageBorder) {
 // pattern 写出图案单元，保留对象绘制顺序
 // 入参: pattern 图案
 func (x *ofdXML) pattern(pattern *Pattern) {
+	if !x.checkContext() {
+		return
+	}
 	var attrs ofdAttrs
 	attrs.number("Width", pattern.Width)
 	attrs.number("Height", pattern.Height)
@@ -1028,6 +1132,9 @@ func (x *ofdXML) pattern(pattern *Pattern) {
 	content.add("Thumbnail", pattern.CellContent.Thumbnail)
 	x.start("CellContent", content)
 	for object := range pattern.CellContent.objects() {
+		if x.err != nil {
+			return
+		}
 		x.object(object, false)
 	}
 	x.end("CellContent")
@@ -1037,19 +1144,28 @@ func (x *ofdXML) pattern(pattern *Pattern) {
 // actions 写出对象动作及精确点击区域，不检查链接可达性
 // 入参: actions 动作列表
 func (x *ofdXML) actions(actions []Action) {
-	if len(actions) == 0 {
+	if len(actions) == 0 || !x.checkContext() {
 		return
 	}
 	x.start("Actions", nil)
 	for _, action := range actions {
+		if x.err != nil {
+			return
+		}
 		var attrs ofdAttrs
 		attrs.add("Event", action.Event)
 		x.start("Action", attrs)
 		if action.Region != nil {
 			x.start("Region", nil)
 			for _, area := range action.Region.Area {
+				if x.err != nil {
+					return
+				}
 				x.start("Area", ofdAttrs{{Name: xml.Name{Local: "Start"}, Value: area.Start}})
 				for _, command := range area.Command {
+					if x.err != nil {
+						return
+					}
 					var attrs ofdAttrs
 					for _, pair := range [][2]string{{"Point1", command.Point1}, {"Point2", command.Point2}, {"Point3", command.Point3}, {"EllipseSize", command.EllipseSize}, {"RotationAngle", command.RotationAngle}, {"LargeArc", command.LargeArc}, {"SweepDirection", command.SweepDirection}, {"EndPoint", command.EndPoint}} {
 						attrs.add(pair[0], pair[1])