skip iframes that already have tabs open

browser_use/browser/context.py

@@ -762,7 +762,7 @@ class BrowserContext:
 		# Check if current page is still valid, if not switch to another available page
 		try:
 			page = await self.get_current_page()
-			# Test if page and its iframes are still accessible
+			# Test if page is still accessible
 			await page.evaluate('1')
 		except Exception as e:
 			logger.debug(f'Current page is no longer accessible: {str(e)}')
@@ -790,6 +790,8 @@ class BrowserContext:
 			# mark the titles of the new tabs so the LLM knows to check them for additional content
 			iframe_urls = await dom_service.get_cross_origin_iframes()
 			for url in iframe_urls:
+				if url in [tab.url for tab in tabs_info]:
+					continue  # skip if the iframe if we already have it open in a tab
 				await self.create_new_tab(url)
 				tabs_info.append(
 					TabInfo(
@@ -1201,7 +1203,7 @@ class BrowserContext:
 				tab_info = TabInfo(page_id=page_id, url=page.url, title=await asyncio.wait_for(page.title(), timeout=1))
 				tabs_info.append(tab_info)
 			except asyncio.TimeoutError:
-				# page.title with hang forever on tabs that are frozen or about:blank
+				# page.title() can hang forever on tabs that are crashed/dissapeared/about:blank
 				# we dont want to try automating those tabs because they will hang the whole script
 				logger.debug('Failed to get tab info for tab: %s (ignoring)', page_id)
 				continue

browser_use/controller/service.py

@@ -1,9 +1,8 @@
 import asyncio
-import re
-import json
 import enum
 import json
 import logging
+import re
 from typing import Dict, Generic, Optional, Type, TypeVar
 
 from langchain_core.language_models.chat_models import BaseChatModel
@@ -166,7 +165,7 @@ class Controller(Generic[Context]):
 			logger.info(msg)
 			logger.debug(f'Element xpath: {element_node.xpath}')
 			return ActionResult(extracted_content=msg, include_in_memory=True)
-		
+
 		# Save PDF
 		@self.registry.action(
 			'Save the current page as a PDF file',
@@ -179,7 +178,7 @@ class Controller(Generic[Context]):
 
 			await page.emulate_media('screen')
 			await page.pdf(path=sanitized_filename, format='A4', print_background=False)
-			msg = f"Saving page with URL {page.url} as PDF to ./{sanitized_filename}"
+			msg = f'Saving page with URL {page.url} as PDF to ./{sanitized_filename}'
 			logger.info(msg)
 			return ActionResult(extracted_content=msg, include_in_memory=True)
 
@@ -211,7 +210,7 @@ class Controller(Generic[Context]):
 
 			content = markdownify.markdownify(await page.content())
 
-			# append iframe content into the page so it's accessible to the LLM too
+			# manually append iframe text into the content so it's readable by the LLM (includes cross-origin iframes)
 			for iframe in page.frames:
 				if iframe.url != page.url and not iframe.url.startswith('data:'):
 					content += f'\n\nIFRAME {iframe.url}:\n'

browser_use/dom/buildDomTree.js

@@ -1,5 +1,12 @@
-({ doHighlightElements=true, focusHighlightIndex=-1, viewportExpansion=0, debugMode=false, indexOffset=0 }={}) => {
-  // console.log('indexOffset', indexOffset)
+(
+  args = {
+    doHighlightElements: true,
+    focusHighlightIndex: -1,
+    viewportExpansion: 0,
+    debugMode: false,
+  }
+) => {
+  const { doHighlightElements, focusHighlightIndex, viewportExpansion, debugMode } = args;
   let highlightIndex = 0; // Reset highlight index
 
   // Add timing stack to handle recursion
@@ -176,7 +183,7 @@
    */
   const DOM_HASH_MAP = {};
 
-  const ID = { current: indexOffset };
+  const ID = { current: 0 };
 
   const HIGHLIGHT_CONTAINER_ID = "playwright-highlight-container";
 
@@ -768,8 +775,8 @@
       element.hasAttribute("role") ||
       element.hasAttribute("tabindex") ||
       element.hasAttribute("aria-") ||
-      element.hasAttribute("data-action") ||
-      element.getAttribute("contenteditable") == "true";
+      element.hasAttribute("data-action");
+
     return hasQuickInteractiveAttr;
   }
 
@@ -793,10 +800,6 @@
       return null;
     }
 
-    // FOR DEBUGGING: mark the node as having been processed by buildDomTree.js
-    // so that playwright can tell if it's been processed or not
-    // node.setAttribute('data-browser-use-built-dom-tree', 'true');
-
     // Special handling for root node (body)
     if (node === document.body) {
       const nodeData = {
@@ -933,9 +936,6 @@
               const domElement = buildDomTree(child, node);
               if (domElement) nodeData.children.push(domElement);
             }
-            // mark the iframe as having been processed by buildDomTree.js
-            // so that playwright can tell if got processed already or if it needs special handling
-            node.contentWindow._loadedBrowserUseBuildDomTree = true;
           }
         } catch (e) {
           console.warn("Unable to access iframe:", e);
@@ -1050,6 +1050,6 @@
   }
 
   return debugMode ?
-    { rootId, map: DOM_HASH_MAP, perfMetrics: PERF_METRICS, indexOffset } :
-    { rootId, map: DOM_HASH_MAP, indexOffset };
+    { rootId, map: DOM_HASH_MAP, perfMetrics: PERF_METRICS } :
+    { rootId, map: DOM_HASH_MAP };
 };