Skip to content

Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.

fix(frontend): preserve inline images in EPUB previews (#4226)

Jarvisec382d79114c
diff --git a/frontend/src/utils/epubPreview.test.ts b/frontend/src/utils/epubPreview.test.ts
new file mode 100644
index 000000000..1a164201e
--- /dev/null
+++ b/frontend/src/utils/epubPreview.test.ts
@@ -0,0 +1,78 @@
+import assert from 'node:assert/strict'
+import { after, test } from 'node:test'
+import JSZip from 'jszip'
+import { DOMParser as XmlParser, XMLSerializer } from '@xmldom/xmldom'
+import { renderEpubPreview } from './epubPreview'
+
+// Adapt the XML test DOM's missing browser conveniences. EPUB parsing and ZIP
+// resolution still run through the production renderer, with no mocked output.
+const originalParser = globalThis.DOMParser
+class TestParser {
+ parseFromString(text: string) {
+ const doc = new XmlParser().parseFromString(text, 'application/xhtml+xml')
+ for (const element of Array.from(doc.getElementsByTagName('*'))) {
+ Object.defineProperty(element, 'innerHTML', {
+ get: () => Array.from(element.childNodes).map(node => new XMLSerializer().serializeToString(node)).join(''),
+ })
+ Object.assign(element, {
+ querySelectorAll(selector: string) {
+ assert.ok(selector === 'img' || selector === 'a[href]')
+ const tag = selector === 'img' ? 'img' : 'a'
+ return Array.from(element.getElementsByTagName(tag))
+ .filter(node => tag !== 'a' || node.hasAttribute('href'))
+ },
+ })
+ }
+ Object.assign(doc, { body: doc.getElementsByTagName('body')[0] })
+ return doc
+ }
+}
+globalThis.DOMParser = TestParser as unknown as typeof DOMParser
+after(() => {
+ if (originalParser) globalThis.DOMParser = originalParser
+ else Reflect.deleteProperty(globalThis, 'DOMParser')
+})
+
+const inlineImage = 'data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///ywAAAAAAQABAAACAUwAOw=='
+
+async function fixture(body: string) {
+ const zip = new JSZip()
+ zip.file('META-INF/container.xml', '<container><rootfiles><rootfile full-path="EPUB/book.opf"/></rootfiles></container>')
+ zip.file('EPUB/book.opf', '<package><manifest><item id="chapter" href="text/chapter.xhtml"/></manifest><spine><itemref idref="chapter"/></spine></package>')
+ zip.file('EPUB/text/chapter.xhtml', `<html xmlns="http://www.w3.org/1999/xhtml" xmlns:xlink="http://www.w3.org/1999/xlink"><head><title>Images</title></head><body>${body}</body></html>`)
+ zip.file('EPUB/images/picture.gif', Buffer.from(inlineImage.split(',')[1], 'base64'))
+ return zip.generateAsync({ type: 'arraybuffer' })
+}
+
+test('EPUB keeps inline HTML and SVG image sources without creating blob URLs', async () => {
+ const result = await renderEpubPreview(await fixture(
+ `<img src="${inlineImage}"/><svg xmlns="http://www.w3.org/2000/svg"><image href="${inlineImage}"/><image xlink:href="${inlineImage}"/></svg>`,
+ ))
+ assert.equal(result.html.split(inlineImage).length - 1, 3)
+ assert.deepEqual(result.objectUrls, [])
+ assert.match(result.html, /data-epub-section="1"/)
+})
+
+test('EPUB keeps relative archive images working alongside inline images and reuses blobs', async () => {
+ const result = await renderEpubPreview(await fixture(
+ `<img src="${inlineImage}"/><img src="../images/picture.gif"/><img src="../images/picture.gif"/>`,
+ ))
+ try {
+ assert.ok(result.html.includes(inlineImage))
+ assert.equal(result.objectUrls.length, 1)
+ assert.equal(result.html.split(result.objectUrls[0]!).length - 1, 2)
+ const blob = await (await fetch(result.objectUrls[0]!)).blob()
+ assert.equal(blob.type, 'image/gif')
+ assert.deepEqual(Buffer.from(await blob.arrayBuffer()), Buffer.from(inlineImage.split(',')[1], 'base64'))
+ } finally {
+ result.objectUrls.forEach(url => URL.revokeObjectURL(url))
+ }
+})
+
+test('non-image data and executable image sources are not preserved', async () => {
+ const result = await renderEpubPreview(await fixture(
+ '<img src="data:text/html;base64,PHNjcmlwdD4="/><img src="javascript:alert(1)"/><img src="missing.gif"/>',
+ ))
+ assert.doesNotMatch(result.html, /data:text\/html|javascript:/)
+ assert.deepEqual(result.objectUrls, [])
+})
diff --git a/frontend/src/utils/epubPreview.ts b/frontend/src/utils/epubPreview.ts
index c6f7f2b30..9d2c0b83d 100644
--- a/frontend/src/utils/epubPreview.ts
+++ b/frontend/src/utils/epubPreview.ts
@@ -82,6 +82,13 @@ export async function renderEpubPreview(data: ArrayBuffer): Promise<EpubPreview>
return url
}
+ // Inline images already contain their bytes; they are not ZIP entries.
+ // The caller still sanitizes the rendered HTML before inserting it.
+ const resolveImageUrl = (chapter: string, source: string): Promise<string> => {
+ if (/^data:image\//i.test(source.trim())) return Promise.resolve(source)
+ return imageUrl(resolveEpubPath(chapter, source))
+ }
+
const sections: string[] = []
for (let i = 0; i < spine.length; i++) {
const path = spine[i]
@@ -92,12 +99,12 @@ export async function renderEpubPreview(data: ArrayBuffer): Promise<EpubPreview>
const body = doc.body || doc.documentElement
for (const img of Array.from(body.querySelectorAll('img'))) {
const src = img.getAttribute('src')
- img.setAttribute('src', src ? await imageUrl(resolveEpubPath(path, src)) : '')
+ img.setAttribute('src', src ? await resolveImageUrl(path, src) : '')
}
for (const image of byLocalName(body as Element, 'image')) {
const href = image.getAttribute('href') || image.getAttribute('xlink:href')
if (href) {
- const url = await imageUrl(resolveEpubPath(path, href))
+ const url = await resolveImageUrl(path, href)
image.setAttribute('href', url)
image.removeAttribute('xlink:href')
}