Skip to content

armgit / cindy

Public Imported from GitHub
0

Consider it done. The open-source AI agent that works out of the box · 想到,就能做到。开源、开箱即用的 AI Agent。

fix(desktop): prevent Markdown previews from hanging the main process (#5207)

Praiseb2f98143234f
diff --git a/apps/desktop/src/main/__tests__/notificationPreview.test.ts b/apps/desktop/src/main/__tests__/notificationPreview.test.ts
index 0229825aba..5e40b62de1 100644
--- a/apps/desktop/src/main/__tests__/notificationPreview.test.ts
+++ b/apps/desktop/src/main/__tests__/notificationPreview.test.ts
@@ -9,8 +9,15 @@ describe('notification plain text', () => {
['`a_b * 2`\n\n```ts\nconst a_b = 2 * 3;\n```', 'a_b * 2 const a_b = 2 * 3;'],
['> 已完成\n\n- 第一项\n- 第二项\n\n~~旧内容~~', '已完成 第一项 第二项 旧内容'],
['![private filename](/private/file.png)', ''],
+ ['![private filename][image]\n\n[image]: https://example.com/image.png', ''],
+ ['<img src="https://example.com/image.png" alt="private filename">', ''],
+ ['&amp; &#x1f600; &lt;tag&gt;', '& 😀 <tag>'],
['**日本語** / _한국어_ / [繁體中文](url)', '日本語 / 한국어 / 繁體中文'],
])('converts %s', (markdown, expected) => expect(notificationPreview(markdown)).toBe(expected));
+ it('does not stall on bracketed repeated code spans', () => {
+ const markdown = '[' + '`````a`````'.repeat(15) + ']';
+ expect(notificationPreview(markdown)).toBe('[' + ('a' + '`'.repeat(10)).repeat(14) + 'a]');
+ });
it('shortens after parsing, without splitting Unicode characters', () => {
expect(notificationPreview('**😀中文abc**', 4)).toBe('😀中文');
expect(notificationPreview('abc😀', 4)).toBe('abc');
diff --git a/apps/desktop/src/main/localDb/__tests__/mapperMarkdownPreview.test.ts b/apps/desktop/src/main/localDb/__tests__/mapperMarkdownPreview.test.ts
index 7cd284c6dd..7b6d44d302 100644
--- a/apps/desktop/src/main/localDb/__tests__/mapperMarkdownPreview.test.ts
+++ b/apps/desktop/src/main/localDb/__tests__/mapperMarkdownPreview.test.ts
@@ -14,6 +14,13 @@ describe('sidebar Markdown preview', () => {
expect(extractMessagePreview(JSON.stringify({ text: markdown }), 'user')).toBe(plain);
});
+ it('handles bracketed code spans that made the old link lexer backtrack', () => {
+ const markdown = '[' + '`````a`````'.repeat(15) + ']';
+ const expected = ('[' + ('a' + '`'.repeat(10)).repeat(14) + 'a]').slice(0, 140);
+ expect(finalizePlainPreview(markdown, 'user')).toBe(expected);
+ expect(extractMessagePreview(JSON.stringify({ text: markdown }), 'user')).toBe(expected);
+ });
+
it('normalizes existing raw cached previews on read', () => {
const session = sessionToCamel({
createdAt: 1,
@@ -59,6 +66,8 @@ describe('sidebar Markdown preview', () => {
['<img src="javascript:alert(1)" alt="不支持的图片">', null],
['<img src="https://example.com/image.png">', null],
['<script><img src="https://example.com/image.png" alt="隐藏内容"></script>', null],
+ ['![reference alt][image]\n\n[image]: https://example.com/image.png', 'reference alt'],
+ ['&amp; &#x1f600; &lt;tag&gt;', '& 😀 <tag>'],
['---', null],
])('extracts readable content from %s', (text, expected) => {
expect(finalizePlainPreview(text, 'assistant')).toBe(expected);
diff --git a/apps/desktop/src/main/notificationPreview.ts b/apps/desktop/src/main/notificationPreview.ts
index 32559f3b55..836f23e812 100644
--- a/apps/desktop/src/main/notificationPreview.ts
+++ b/apps/desktop/src/main/notificationPreview.ts
@@ -1,24 +1,8 @@
-import { Lexer, type Tokens, type Token } from 'marked';
+import { markdownPreviewText } from '../shared/markdownPreviewText';
-/** Parse Markdown before shortening it: punctuation in ordinary text/code is content. */
+/** Parse before shortening, without leaking image filenames into notifications. */
export function notificationPlainText(markdown: string): string {
- const render = (tokens: Token[]): string => tokens.map((token): string => {
- switch (token.type) {
- case 'space': case 'br': return ' ';
- case 'html': case 'image': case 'def': case 'hr': return '';
- case 'code': case 'codespan': return token.text;
- case 'list': return (token as Tokens.List).items.map((item) => render(item.tokens)).join(' ');
- case 'table': return [(token as Tokens.Table).header, ...(token as Tokens.Table).rows]
- .map((row) => row.map((cell) => render(cell.tokens)).join(' ')).join(' ');
- default: {
- const value = token as Token & { tokens?: Token[]; text?: string };
- const text = value.tokens ? render(value.tokens) : (value.text ?? '');
- return ['paragraph', 'heading', 'blockquote', 'list_item'].includes(token.type)
- ? ` ${text} ` : text;
- }
- }
- }).join('');
- return render(Lexer.lex(markdown)).replace(/\s+/gu, ' ').trim();
+ return markdownPreviewText(markdown, { includeImageAlt: false });
}
export function notificationPreview(markdown: string, limit = 240): string {
diff --git a/apps/desktop/src/shared/markdownPreviewText.ts b/apps/desktop/src/shared/markdownPreviewText.ts
index 66d4e79edd..75a9660cb8 100644
--- a/apps/desktop/src/shared/markdownPreviewText.ts
+++ b/apps/desktop/src/shared/markdownPreviewText.ts
@@ -1,45 +1,62 @@
-import { Lexer, type Token, type Tokens } from 'marked';
+import type { Nodes } from 'mdast';
+import remarkGfm from 'remark-gfm';
+import remarkParse from 'remark-parse';
+import { unified } from 'unified';
import { htmlImgToImageNode } from './htmlImage';
-/** Extract display text without generating HTML or counting link destinations. */
-export function markdownPreviewText(markdown: string): string {
- return previewTokens(Lexer.lex(markdown, { gfm: true }))
- .replace(/\s+/g, ' ')
- .trim();
-}
+const previewParser = unified().use(remarkParse).use(remarkGfm);
-function previewTokens(tokens: Token[]): string {
- return tokens
- .map((token): string => {
- switch (token.type) {
- case 'space':
- case 'br':
- case 'hr':
- case 'def':
- return ' ';
- case 'html':
- return `${htmlImgToImageNode({ type: 'html', value: token.text })?.alt ?? ''} `;
- case 'list':
- return `${token.items.map((item: Tokens.ListItem) => previewTokens(item.tokens)).join(' ')} `;
- case 'table':
- return (
- [token.header, ...token.rows]
- .map((row: { tokens: Token[] }[]) =>
- row.map((cell) => previewTokens(cell.tokens)).join(' '),
- )
- .join(' ') + ' '
- );
- case 'heading':
- case 'paragraph':
- case 'blockquote':
- return `${previewTokens(token.tokens ?? [])} `;
- case 'code':
- return `${token.text} `;
- default:
- if ('tokens' in token && token.tokens) return previewTokens(token.tokens);
- return 'text' in token ? token.text : '';
- }
- })
- .join('');
+/** Extract readable text without HTML output or link destinations. */
+export function markdownPreviewText(
+ markdown: string,
+ options: { includeImageAlt?: boolean } = {},
+): string {
+ // Marked's link regex can backtrack for minutes on short bracketed code spans.
+ // These previews run on Main after DB replies, so use the existing remark parser.
+ const stack: Array<Nodes | string> = [previewParser.parse(markdown)];
+ const parts: string[] = [];
+ const includeImageAlt = options.includeImageAlt !== false;
+ while (stack.length > 0) {
+ const node = stack.pop()!;
+ if (typeof node === 'string') {
+ parts.push(node);
+ continue;
+ }
+ switch (node.type) {
+ case 'text':
+ case 'inlineCode':
+ parts.push(node.value);
+ break;
+ case 'code':
+ parts.push(' ', node.value, ' ');
+ break;
+ case 'image':
+ case 'imageReference':
+ if (includeImageAlt) parts.push(node.alt ?? '');
+ break;
+ case 'html':
+ if (includeImageAlt) parts.push(htmlImgToImageNode(node)?.alt ?? '', ' ');
+ break;
+ case 'definition':
+ break;
+ case 'break':
+ case 'thematicBreak':
+ parts.push(' ');
+ break;
+ default:
+ if ('children' in node) {
+ const block = node.type !== 'emphasis' && node.type !== 'strong'
+ && node.type !== 'delete' && node.type !== 'link' && node.type !== 'linkReference';
+ if (block) {
+ parts.push(' ');
+ stack.push(' ');
+ }
+ for (let index = node.children.length - 1; index >= 0; index--) {
+ stack.push(node.children[index]);
+ }
+ }
+ }
+ }
+ return parts.join('').replace(/\s+/gu, ' ').trim();
}