feat(性能优化): 优化阅读缓存、全文搜索及PDF导出的内存占用

assets/webui/ofdgo.js

@@ -3,6 +3,8 @@ const FONT_DATABASE = "ofdgo";
 const fontChannel = typeof BroadcastChannel === "function" ? new BroadcastChannel(FONT_DATABASE) : null;
 const COMPACT_LAYOUT = window.matchMedia("(max-width: 900px)");
 const DEFAULT_IMAGE_DPI = 300;
+const PAGE_CACHE_LIMIT = 16;
+const PAGE_CACHE_BYTES = 32 * 1024 * 1024;
 const STATUS = {
 	ready: "选择 OFD 文件",
 	opening: "正在打开文档",
@@ -1760,6 +1762,7 @@ function flowPageObserver() {
 				state.selectedPages.delete(index);
 				renderFlowPage(index, { openSeq, priority: 4 });
 			}
+			trimPageCache();
 		}, {
 			root: el.viewerPanel,
 			rootMargin: "600px 0px",
@@ -1798,6 +1801,7 @@ async function renderFlowPage(index, options = {}) {
 			}
 			updateThumbnail(index, openSeq);
 		}
+		trimPageCache();
 		return page;
 	} catch (err) {
 		if (openSeq === state.openSeq) {
@@ -1816,7 +1820,10 @@ function loadPageData(index, options = {}) {
 		return Promise.resolve(null);
 	}
 	if (state.pageCache.has(index)) {
-		return Promise.resolve(state.pageCache.get(index));
+		const page = state.pageCache.get(index);
+		state.pageCache.delete(index);
+		state.pageCache.set(index, page);
+		return Promise.resolve(page);
 	}
 	const key = `${openSeq}:${index}`;
 	const priority = options.priority ?? 3;
@@ -1869,7 +1876,9 @@ async function processPageRenderQueue() {
 							state.svgImages.set(image.name, image.url);
 						}
 					}
+					page.imageNames = page.images.map((image) => image.name);
 					delete page.images;
+					page.cacheBytes = (page.svg.length + page.text.length) * 2;
 					page.text = JSON.parse(page.text);
 					state.pageCache.set(task.index, page);
 				}
@@ -1885,6 +1894,42 @@ async function processPageRenderQueue() {
 	}
 }
 
+function trimPageCache() {
+	const protectedPages = new Set([state.pageIndex, ...state.visiblePages, ...state.visibleThumbnails, ...state.selectedPages]);
+	for (const task of state.pageInFlight.values()) {
+		protectedPages.add(task.index);
+	}
+	const references = new Map();
+	let bytes = 0;
+	for (const page of state.pageCache.values()) {
+		bytes += page.cacheBytes;
+		for (const name of page.imageNames) {
+			references.set(name, (references.get(name) || 0) + 1);
+		}
+	}
+	for (const url of state.svgImages.values()) {
+		bytes += url.length * 2;
+	}
+	for (const [index, page] of state.pageCache) {
+		if (state.pageCache.size <= PAGE_CACHE_LIMIT && bytes <= PAGE_CACHE_BYTES) {
+			break;
+		}
+		if (protectedPages.has(index)) {
+			continue;
+		}
+		state.pageCache.delete(index);
+		bytes -= page.cacheBytes;
+		for (const name of page.imageNames) {
+			const count = references.get(name) - 1;
+			references.set(name, count);
+			if (count === 0) {
+				bytes -= state.svgImages.get(name).length * 2;
+				state.svgImages.delete(name);
+			}
+		}
+	}
+}
+
 async function loadSVGFonts(fonts, openSeq) {
 	for (const font of fonts) {
 		if (openSeq !== state.openSeq) {
@@ -1992,6 +2037,7 @@ function syncSelection() {
 	for (const index of state.selectedPages) {
 		unmountPage(index);
 	}
+	trimPageCache();
 }
 
 function isDocumentSelection(range) {
@@ -2652,6 +2698,7 @@ function thumbnailObserver() {
 				state.visibleThumbnails.add(index);
 				renderThumbnail(index, openSeq);
 			}
+			trimPageCache();
 		}, {
 			root: el.pageListPanel,
 			rootMargin: "180px 0px",
@@ -2682,6 +2729,7 @@ async function renderThumbnail(index, openSeq = state.openSeq) {
 			return;
 		}
 		updateThumbnail(index, openSeq);
+		trimPageCache();
 	} catch {
 		if (openSeq === state.openSeq) {
 			markThumbnailError(index);

internal/webui/webui.go

@@ -278,15 +278,8 @@ func (s *Session) SetFonts(fonts []FontFile) error {
 func (s *Session) PageText(index int) (*ofdgo.PageText, error) {
 	text := s.textCache[index]
 	if text == nil {
-		page := s.pageCache[index]
 		var err error
-		if page == nil {
+		text, err = s.readPageText(index)
-			page, err = s.Reader.PageContentByIndex(index)
-			if err != nil {
-				return nil, err
-			}
-		}
-		text, err = s.Renderer.PageText(page)
 		if err != nil {
 			return nil, err
 		}
@@ -301,11 +294,8 @@ func (s *Session) PageText(index int) (*ofdgo.PageText, error) {
 func (s *Session) PageTextString(index int) (string, error) {
 	text := s.textCache[index]
 	if text == nil {
-		page, err := s.Reader.PageContentByIndex(index)
+		var err error
-		if err != nil {
+		text, err = s.readPageText(index)
-			return "", err
-		}
-		text, err = s.Renderer.PageText(page)
 		if err != nil {
 			return "", err
 		}
@@ -313,6 +303,16 @@ func (s *Session) PageTextString(index int) (string, error) {
 	return text.String(), nil
 }
 
+// readPageText 复用已加载页面，否则只解析文字所需图元
+// 入参: index 页面索引
+// 返回: *ofdgo.PageText 页面文字, error 错误信息
+func (s *Session) readPageText(index int) (*ofdgo.PageText, error) {
+	if page := s.pageCache[index]; page != nil {
+		return s.Renderer.PageText(page)
+	}
+	return s.Renderer.PageTextByIndex(index)
+}
+
 // SearchPage 搜索指定页面，复用当前会话的文字索引
 // 入参: index 页面索引, query 搜索文字
 // 返回: []ofdgo.TextMatch 匹配结果, error 错误信息
@@ -667,21 +667,10 @@ func (s *Session) ExportPDF(writer io.Writer, indices ...int) error {
 	if s == nil || s.Reader == nil || s.Renderer == nil || s.doc == nil {
 		return fmt.Errorf("ofd document is not opened")
 	}
-	if len(indices) > 0 {
+	return s.Renderer.RenderToMultiPagePDF(writer, indices...)
-		return s.Renderer.RenderToMultiPagePDF(writer, indices...)
-	}
-	pages := make([]*ofdgo.PageContent, len(s.doc.Pages.Page))
-	for i := range pages {
-		page, err := s.pageContent(i)
-		if err != nil {
-			return err
-		}
-		pages[i] = page
-	}
-	return s.Renderer.RenderPagesToPDF(pages, writer)
 }
 
-// pageContent 获取页面内容
+// pageContent 读取页面，复用最近一次解析结果
 // 入参: index 页面索引
 // 返回: *ofdgo.PageContent 页面内容, error 错误信息
 func (s *Session) pageContent(index int) (*ofdgo.PageContent, error) {
@@ -691,6 +680,7 @@ func (s *Session) pageContent(index int) (*ofdgo.PageContent, error) {
 	if page, ok := s.pageCache[index]; ok {
 		return page, nil
 	}
+	clear(s.pageCache)
 	page, err := s.Reader.PageContentByIndex(index)
 	if err != nil {
 		return nil, err

ofdgo_reader.go

@@ -16,6 +16,7 @@ package ofdgo
 
 import (
 	"archive/zip"
+	"bytes"
 	"encoding/xml"
 	"fmt"
 	"io"
@@ -254,13 +255,26 @@ func resolveResourcePath(resPath, baseLoc, filePath string) string {
 // 入参: page 页面对象
 // 返回: *PageContent 页面内容, error 错误信息
 func (r *Reader) PageContent(page Page) (*PageContent, error) {
+	return r.readPageContent(page, false)
+}
+
+// readPageContent 读取页面，可跳过文字提取不使用的独立路径和图片
+// 入参: page 页面对象, textOnly 是否仅用于文字提取
+// 返回: *PageContent 页面内容, error 错误信息
+func (r *Reader) readPageContent(page Page, textOnly bool) (*PageContent, error) {
 	fullPath := r.ResPath(page.BaseLoc)
 	data, err := r.readFile(fullPath)
 	if err != nil {
 		return nil, err
 	}
 	var content PageContent
-	if err := xml.Unmarshal(data, &content); err != nil {
+	if textOnly {
+		decoder := &textPageTokens{Decoder: xml.NewDecoder(bytes.NewReader(data))}
+		err = xml.NewTokenDecoder(decoder).Decode(&content)
+	} else {
+		err = xml.Unmarshal(data, &content)
+	}
+	if err != nil {
 		return nil, fmt.Errorf("failed to unmarshal page content: %w", err)
 	}
 	content.ID = page.ID

ofdgo_renderer_output.go

@@ -146,11 +146,7 @@ func (r *Renderer) RenderToMultiPageText(writer io.Writer, indices ...int) error
 	}
 	written := false
 	for i, index := range indices {
-		page, err := r.Reader.PageContentByIndex(index)
+		text, err := r.PageTextByIndex(index)
-		if err != nil {
-			return fmt.Errorf("failed to read page %d: %w", index+1, err)
-		}
-		text, err := r.PageText(page)
 		if err != nil {
 			return fmt.Errorf("failed to extract page %d text: %w", index+1, err)
 		}
@@ -420,6 +416,7 @@ func (r *Renderer) renderPDFPages(pages []pdfPage, writer io.Writer, progress fu
 			buf.Truncate(start)
 			return fmt.Errorf("failed to render page %d: %w", i+1, err)
 		}
+		pages[i].Content = nil
 		if progress != nil {
 			if err := progress(i+1, len(pages)); err != nil {
 				buf.Truncate(start)

ofdgo_search.go

@@ -15,6 +15,8 @@
 package ofdgo
 
 import (
+	"encoding/xml"
+	"fmt"
 	"math"
 	"strings"
 	"unicode"
@@ -23,6 +25,39 @@ import (
 	"github.com/tdewolff/canvas"
 )
 
+// textPageTokens 保留文字及其复合图元和裁剪，跳过独立路径与图片
+type textPageTokens struct {
+	*xml.Decoder
+	parents []string
+}
+
+// Token 读取文字页面所需的XML节点
+// 返回: xml.Token XML节点, error 错误信息
+func (d *textPageTokens) Token() (xml.Token, error) {
+	for {
+		token, err := d.Decoder.Token()
+		if err != nil {
+			return nil, err
+		}
+		switch node := token.(type) {
+		case xml.StartElement:
+			if len(d.parents) > 0 && (node.Name.Local == "PathObject" || node.Name.Local == "ImageObject") {
+				switch d.parents[len(d.parents)-1] {
+				case "Layer", "PageBlock", "CompositeGraphicUnit", "CompositeObject":
+					if err := d.Decoder.Skip(); err != nil {
+						return nil, err
+					}
+					continue
+				}
+			}
+			d.parents = append(d.parents, node.Name.Local)
+		case xml.EndElement:
+			d.parents = d.parents[:len(d.parents)-1]
+		}
+		return token, nil
+	}
+}
+
 // PageText 页面文字，按绘制顺序保留文本对象，不包含图像和签名外观
 type PageText struct {
 	Runs []TextRun `json:"runs"`
@@ -74,6 +109,24 @@ func (r *Renderer) PageText(page *PageContent) (*PageText, error) {
 	return renderer.pageText, nil
 }
 
+// PageTextByIndex 按页面索引提取文字，避免解析独立路径和图片图元
+// 入参: index 页面索引，从0开始
+// 返回: *PageText 页面文字, error 错误信息
+func (r *Renderer) PageTextByIndex(index int) (*PageText, error) {
+	doc, err := r.Reader.Doc()
+	if err != nil {
+		return nil, err
+	}
+	if index < 0 || index >= len(doc.Pages.Page) {
+		return nil, fmt.Errorf("page index %d out of range", index)
+	}
+	page, err := r.Reader.readPageContent(doc.Pages.Page[index], true)
+	if err != nil {
+		return nil, err
+	}
+	return r.PageText(page)
+}
+
 // String 按文本对象顺序合并原文，以换行分隔非空对象
 // 返回: string 页面原文
 func (p *PageText) String() string {