fix(web): typeset book-source HTML chapters like TXT

Source chapters are scraped fragments (`<div>  text<br><br><br><br>`) served
as text/html, so the TXT cleanup never ran: every paragraph was followed by
three empty lines and the indent setting did nothing. Like legado-E, a
text-only chapter of a `chapters` book is read back as characters when it
is loaded (htmlToText) and typeset by the TXT path. Done at read time, so
chapters the server cached earlier are cleaned too; chapters with images
or tables keep their markup and EPUBs never take this path.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/ui.md

@@ -604,6 +604,10 @@ TXT 是唯一一种**没有作者排版**的书，所以它也是唯一一处「
   （`br`/`p`/`div`/标题/列表项）换成换行、行内标签（`span`/`font`/`b`…）去掉、
   常见实体解码、只有空白的行算空行；只认真实的标签名，正文里的 `a<b`、`<书名>`
   不动，其余标签照旧转义成文字。
+  书源（`format: 'chapters'`）的 `text/html` 章节是抓下来的网页片段（`<div>…<br><br><br>`）。
+  和 legado-E 一样，只含文字和换行的章节在读取时（`readStagedSection`）用 `htmlToText`
+  读回成字符、标成纯文本，走同一套分段和缩进；含图片、表格、媒体的章节保留标记。
+  在读取时处理，所以服务端早先缓存的章节也会被清洗；EPUB 不会进入这条路径。
   `ReaderView` 在 `retypePlainText` 里对拿到的章节过这一段逻辑；如果内容带着
   历史的 `.txt-body` 标记（旧服务端、旧缓存），就按块读回文本再拆，所以旧
   响应不会被当成新响应渲染。

web/src/formats/segments.ts

@@ -71,18 +71,45 @@ const WHITESPACE_LINE = /^[ \t\u3000\u00a0]+$/gm;
  * consumer escapes it, so nothing here can turn into markup.
  */
 export function cleanTextMarkup(text: string): string {
-  return text
-    .replace(LINE_ENDINGS, '\n')
-    .replace(BLOCK_TAG, '\n')
-    .replace(INLINE_TAG, '')
-    .replace(ENTITY, (all, decimal: string | undefined, hex: string | undefined, name: string | undefined) => {
-      if (name) return NAMED_ENTITIES[name.toLowerCase()] ?? all;
-      const code = decimal ? Number(decimal) : Number.parseInt(hex!, 16);
-      return code > 0 && code <= 0x10ffff ? String.fromCodePoint(code) : all;
-    })
+  return decodeEntities(text.replace(LINE_ENDINGS, '\n').replace(BLOCK_TAG, '\n').replace(INLINE_TAG, ''))
     .replace(WHITESPACE_LINE, '');
 }
 
+function decodeEntities(text: string): string {
+  return text.replace(ENTITY, (all, decimal: string | undefined, hex: string | undefined, name: string | undefined) => {
+    if (name) return NAMED_ENTITIES[name.toLowerCase()] ?? all;
+    const code = decimal ? Number(decimal) : Number.parseInt(hex!, 16);
+    return code > 0 && code <= 0x10ffff ? String.fromCodePoint(code) : all;
+  });
+}
+
+/** Elements a chapter cannot lose by being read as text. */
+const RICH_MARKUP = /<(?:img|svg|table|video|audio|iframe|math|canvas|picture|object|embed|ruby)\b/i;
+const COMMENT = /<!--[\s\S]*?-->/g;
+const HIDDEN_BLOCK = /<(script|style|template|noscript)\b[^>]*>[\s\S]*?<\/\1\s*>/gi;
+const ANY_TAG = /<\/?[a-zA-Z][^<>]*>/g;
+
+/**
+ * Whether a book source's HTML chapter is only text with line-level markup, so it
+ * can be typeset like a TXT without losing anything (no images, tables or media).
+ */
+export function isTextOnlyMarkup(html: string): boolean {
+  return !RICH_MARKUP.test(html);
+}
+
+/**
+ * A book source's HTML chapter as characters, the way legado-E reads it
+ * (`HtmlFormatter.format`): block tags and line breaks become newlines and every
+ * other tag is removed. Unlike {@link cleanTextMarkup} the input *is* markup, so
+ * no tag survives as text; entities are decoded last, so an escaped `&lt;b&gt;`
+ * in the prose becomes the characters `<b>` rather than being stripped.
+ */
+export function htmlToText(html: string): string {
+  const stripped = html.replace(LINE_ENDINGS, '\n').replace(COMMENT, '').replace(HIDDEN_BLOCK, '')
+    .replace(BLOCK_TAG, '\n').replace(ANY_TAG, '');
+  return decodeEntities(stripped).replace(WHITESPACE_LINE, '');
+}
+
 /** Paragraphs are separated by a blank line. */
 const PARAGRAPH_BREAK = /\n{2,}/;
 

web/src/formats/windowed.ts

@@ -42,7 +42,7 @@ import type { BookDoc, Section, StagedBook, TocEntry } from './types.ts';
  */
 export interface SectionLoader {
   /** Section markup for a reflowable section, or a page's bytes. */
-  read(item: ContentItem): Promise<{ html?: string; image?: { mediaType: string; bytes: Uint8Array } }>;
+  read(item: ContentItem): Promise<{ html?: string; image?: { mediaType: string; bytes: Uint8Array }; plainText?: boolean }>;
 }
 
 /**
@@ -173,6 +173,9 @@ export function createStagedDoc(options: StagedDocOptions): StagedDoc {
       const loaded = await options.loader.read(item);
       if (loaded.html !== undefined) section.html = loaded.html;
       if (loaded.image !== undefined) section.image = loaded.image;
+      // The loader read the body and found characters (a book source's text-only
+      // HTML chapter, converted): typeset it like a TXT, and do not parse it.
+      if (loaded.plainText) { section.plainText = true; section.bodyIsMarkup = false; }
       return section;
     },
     setWindow(content: BookContent, spine: number): number {

web/src/ui/reader-screen.tsx

@@ -36,7 +36,7 @@ import { ReadingTools, type SearchSection } from './reading-tools.tsx';
 import { decodeAnchor, encodeAnchor, searchableText, type TextAnchor } from './text-anchor.ts';
 import { AnnotationPopover, type AnnotationTarget } from './annotation-popover.tsx';
 import { newId } from '../core/id.ts';
-import { textToChapterHtml } from '../formats/segments.ts';
+import { htmlToText, isTextOnlyMarkup, textToChapterHtml } from '../formats/segments.ts';
 import { parseLocator } from './locator.ts';
 
 export interface ReaderScreenOptions {
@@ -489,7 +489,7 @@ export class ReaderScreen {
   private async readStagedSection(
     book: Book,
     item: { href: string; kind: string; mediaType: string; format?: string; resourceRef?: string },
-  ): Promise<{ html?: string; image?: { mediaType: string; bytes: Uint8Array } }> {
+  ): Promise<{ html?: string; image?: { mediaType: string; bytes: Uint8Array }; plainText?: boolean }> {
     const ref = renditionRef(item);
     // Keyed by the rendition, not just the section: a client that read the plain
     // text before this change and the markup after it must not serve the cached
@@ -505,7 +505,16 @@ export class ReaderScreen {
     if (item.kind === 'page') {
       return { image: { mediaType: item.mediaType, bytes: new Uint8Array(await blob.arrayBuffer()) } };
     }
-    return { html: await blob.text() };
+    const body = await blob.text();
+    // A book source's chapter is a scraped page fragment (`<div>…<br><br><br>`), not
+    // an authored document: like legado-E, read a text-only one back as characters so
+    // it gets the TXT paragraphs and the reader's indent. Applied when *read*, so
+    // chapters the server cached before this change are cleaned too. Chapters with
+    // images or tables keep their markup; EPUBs never reach this branch.
+    if (book.format === 'chapters' && item.kind === 'chapter' && /html/i.test(item.mediaType) && isTextOnlyMarkup(body)) {
+      return { html: htmlToText(body), plainText: true };
+    }
+    return { html: body };
   }
 
   dispose(): void {

web/test/segments.test.ts

@@ -2,6 +2,8 @@
 import { describe, expect, it } from 'vitest';
 import {
   cleanTextMarkup,
+  htmlToText,
+  isTextOnlyMarkup,
   endsAtParagraphBreak,
   escapeHtml,
   paragraphElement,
@@ -296,3 +298,22 @@ describe('web-page markup in a TXT', () => {
     expect(endsAtParagraphBreak('一段')).toBe(false);
   });
 });
+
+/** A book source's chapter is a scraped HTML fragment; legado-E reads it as text. */
+describe('book-source HTML chapters', () => {
+  const fragment = '<div>    天斗星作为第一颗行政星。<br /><br /><br /><br />    星球表面本来是没有生物的。<br /><br /></div>';
+
+  it('reads a text-only fragment back into the paragraphs it meant', () => {
+    expect(splitTextParagraphs(htmlToText(fragment)).map((p) => p.text)).toEqual(['天斗星作为第一颗行政星。', '星球表面本来是没有生物的。']);
+  });
+
+  it('strips every tag but keeps escaped angle brackets as text', () => {
+    expect(htmlToText('<p>a&lt;b&gt;<custom-tag x>c</custom-tag><!-- ad --><script>bad()</script></p>')).toBe('\na<b>c\n');
+  });
+
+  it('keeps chapters with images or tables as markup', () => {
+    expect(isTextOnlyMarkup(fragment)).toBe(true);
+    expect(isTextOnlyMarkup('<p>图</p><img src="x">')).toBe(false);
+    expect(isTextOnlyMarkup('<table><tr><td>1</td></tr></table>')).toBe(false);
+  });
+});