Cli: Fixes #16894: Fix quadratic conversion time for large tables (#16901)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

packages/app-cli/tests/HtmlToMd.ts

@@ -2,6 +2,7 @@ import shim from '@joplin/lib/shim';
 import * as os from 'os';
 import { readFile } from 'fs/promises';
 import { filename } from '@joplin/lib/path-utils';
+import { dirname } from 'path';
 import HtmlToMd, { ParseOptions } from '@joplin/lib/HtmlToMd';
 import MdToHtml from '@joplin/renderer/MdToHtml';
 
@@ -130,6 +131,31 @@ describe('HtmlToMd', () => {
 		expect(collapsedResult).toContain('\n\n');
 	});
 
+	it('should not re-read the rows of a table for every row', async () => {
+		// Under domino, each access to `table.rows` walks the whole table, so reading it once per row made
+		// converting large tables quadratic: https://github.com/laurent22/joplin/issues/16894
+		const resolveFrom = (moduleName: string, fromPath: string) => require.resolve(moduleName, { paths: [dirname(fromPath)] });
+		const domino: { createDocument: (html: string)=> Document } = require(resolveFrom('@mixmark-io/domino', resolveFrom('@joplin/turndown', require.resolve('@joplin/lib/HtmlToMd'))));
+
+		const rowCount = 100;
+		const rows = Array.from({ length: rowCount }, (_, i) => `<tr><td>${i}</td><td>b</td><td>c</td></tr>`).join('');
+		const root = domino.createDocument(`<div><table><thead><tr><th>A</th><th>B</th><th>C</th></tr></thead><tbody>${rows}</tbody></table></div>`).querySelector('div');
+		// Turndown converts a clone of the element it's given, which wouldn't have the counter below
+		Object.defineProperty(root, 'cloneNode', { value: () => root });
+
+		const table = root.querySelector('table');
+		const getRows = Object.getOwnPropertyDescriptor(Object.getPrototypeOf(table), 'rows').get;
+		let rowsReadCount = 0;
+		Object.defineProperty(table, 'rows', { get: () => {
+			rowsReadCount++;
+			return getRows.call(table);
+		} });
+
+		const markdown = new HtmlToMd().parse(root);
+		expect(markdown.split('\n')).toHaveLength(rowCount + 2);
+		expect(rowsReadCount).toBeLessThan(rowCount);
+	});
+
 	// Covers how newlines are carried between children when joining them, which was rewritten to fix
 	// https://github.com/laurent22/joplin/issues/16895
 	it.each([

packages/tools/cspell/dictionary4.txt

@@ -375,4 +375,5 @@ twodo
 Gemma
 untruncated
 instr
+mixmark
 Hrvatski

packages/turndown-plugin-gfm/src/tables.js

@@ -29,8 +29,11 @@ function getColumnAlignment(table, columnIndex) {
 
   var align = '';
 
-  for (var i = 0; i < table.rows.length; ++i) {
-    var row = table.rows[i];
+  // Read `rows` once: under domino each access re-walks the whole table, which made converting large tables
+  // quadratic: https://github.com/laurent22/joplin/issues/16894
+  var rows = table.rows;
+  for (var i = 0; i < rows.length; ++i) {
+    var row = rows[i];
     if (columnIndex < row.childNodes.length) {
       var cellAlignment = getAlignment(row.childNodes[columnIndex]);
       ++votes[cellAlignment];
@@ -371,8 +374,9 @@ function handleColSpan(content, node, emptyChar) {
 
 function tableColCount(node) {
   let maxColCount = 0;
-  for (let i = 0; i < node.rows.length; i++) {
-    const row = node.rows[i]
+  const rows = node.rows;
+  for (let i = 0; i < rows.length; i++) {
+    const row = rows[i]
     const colCount = row.childNodes.length
     if (colCount > maxColCount) maxColCount = colCount
   }