Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.
Python ai-agentsai-toolsbrowser-agentbrowser-automationbrowser-usecdpcloud-browsercomputer-useheadless-browserllmmcpplaywrightpythonremote-browserstealth-browserweb-automationweb-scraping
hot-fix/scroll
Magnus Müllere15a8a216e8e
diff --git a/browser_use/dom/serializer/serializer.py b/browser_use/dom/serializer/serializer.pyindex ccfc5dd89..7fefdc5d9 100644--- a/browser_use/dom/serializer/serializer.py+++ b/browser_use/dom/serializer/serializer.py@@ -433,7 +433,8 @@ class DOMTreeSerializer:return '\n'.join(formatted_text)# Add element with interactive_index if clickable, scrollable, or iframe- is_any_scrollable = node.original_node.is_actually_scrollable+ is_any_scrollable = node.original_node.is_actually_scrollable or node.original_node.is_scrollable+ should_show_scroll = node.original_node.should_show_scroll_infoif node.interactive_index is not None or is_any_scrollable or node.original_node.tag_name.upper() == 'IFRAME':next_depth += 1@@ -441,13 +442,13 @@ class DOMTreeSerializer:attributes_html_str = DOMTreeSerializer._build_attributes_string(node.original_node, include_attributes, '')# Build the line- if is_any_scrollable and node.interactive_index is None:- # Scrollable but not clickable+ if should_show_scroll and node.interactive_index is None:+ # Scrollable container but not clickableline = f'{depth_str}|SCROLL|<{node.original_node.tag_name}'elif node.interactive_index is not None:# Clickable (and possibly scrollable)new_prefix = '*' if node.is_new else ''- scroll_prefix = '|SCROLL+' if is_any_scrollable else '['+ scroll_prefix = '|SCROLL+' if should_show_scroll else '['line = f'{depth_str}{new_prefix}{scroll_prefix}{node.interactive_index}]<{node.original_node.tag_name}'elif node.original_node.tag_name.upper() == 'IFRAME':# Iframe element (not interactive)@@ -460,8 +461,8 @@ class DOMTreeSerializer:line += ' />'- # Add scroll information for scrollable elements- if is_any_scrollable:+ # Add scroll information only when we should show it+ if should_show_scroll:scroll_info_text = node.original_node.get_scroll_info_text()if scroll_info_text:line += f' ({scroll_info_text})'@@ -518,6 +519,12 @@ class DOMTreeSerializer:except (AttributeError, ValueError):continue+ # Include accessibility name (contains StaticText content)+ if node.ax_node and node.ax_node.name and 'ax_name' in include_attributes:+ ax_name_str = str(node.ax_node.name).strip()+ if ax_name_str:+ attributes_to_include['ax_name'] = ax_name_str+if not attributes_to_include:return ''diff --git a/browser_use/dom/views.py b/browser_use/dom/views.pyindex 864cc5f76..14d9e7a5f 100644--- a/browser_use/dom/views.py+++ b/browser_use/dom/views.py@@ -46,6 +46,8 @@ DEFAULT_INCLUDE_ATTRIBUTES = ['level','busy','live',+ # Accessibility name (contains text content for StaticText elements)+ 'ax_name',]@@ -444,21 +446,68 @@ class EnhancedDOMTreeNode:overflow_x = styles.get('overflow-x', overflow).lower()overflow_y = styles.get('overflow-y', overflow).lower()- # Allow scrolling if overflow is not explicitly hidden+ # Only allow scrolling if overflow is explicitly set to auto, scroll, or overlay+ # Do NOT consider 'visible' overflow as scrollable - this was causing the issueallows_scroll = (overflow in ['auto', 'scroll', 'overlay']or overflow_x in ['auto', 'scroll', 'overlay']or overflow_y in ['auto', 'scroll', 'overlay']- or (overflow == 'visible' and (overflow_x != 'hidden' or overflow_y != 'hidden')))return allows_scrollelse:- # No CSS info, but content overflows - likely scrollable- return True+ # No CSS info, but content overflows - be more conservative+ # Only consider it scrollable if it's a common scrollable container element+ scrollable_tags = {'div', 'main', 'section', 'article', 'aside', 'body', 'html'}+ return self.tag_name.lower() in scrollable_tagsreturn False+ @property+ def should_show_scroll_info(self) -> bool:+ """+ Simple check: show scroll info only if this element is scrollable+ and doesn't have a scrollable parent (to avoid nested scroll spam).++ Special case for iframes: Always show scroll info since Chrome might not+ always detect iframe scrollability correctly (scrollHeight: 0 issue).+ """+ # Special case: Always show scroll info for iframe elements+ # Even if not detected as scrollable, they might have scrollable content+ if self.tag_name.lower() == 'iframe':+ return True++ # Must be scrollable first for non-iframe elements+ if not (self.is_scrollable or self.is_actually_scrollable):+ return False++ # Always show for iframe content documents (body/html)+ if self.tag_name.lower() in {'body', 'html'}:+ return True++ # Don't show if parent is already scrollable (avoid nested spam)+ if self.parent_node and (self.parent_node.is_scrollable or self.parent_node.is_actually_scrollable):+ return False++ return True++ def _find_html_in_content_document(self) -> 'EnhancedDOMTreeNode | None':+ """Find HTML element in iframe content document."""+ if not self.content_document:+ return None++ # Check if content document itself is HTML+ if self.content_document.tag_name.lower() == 'html':+ return self.content_document++ # Look through children for HTML element+ if self.content_document.children_nodes:+ for child in self.content_document.children_nodes:+ if child.tag_name.lower() == 'html':+ return child++ return None+@propertydef scroll_info(self) -> dict[str, Any] | None:"""Calculate scroll information for this element if it's scrollable."""@@ -532,39 +581,39 @@ class EnhancedDOMTreeNode:def get_scroll_info_text(self) -> str:"""Get human-readable scroll information text for this element."""+ # Special case for iframes: check content document for scroll info+ if self.tag_name.lower() == 'iframe':+ # Try to get scroll info from the HTML document inside the iframe+ if self.content_document:+ # Look for HTML element in content document+ html_element = self._find_html_in_content_document()+ if html_element and html_element.scroll_info:+ info = html_element.scroll_info+ # Provide minimal but useful scroll info+ pages_below = info.get('pages_below', 0)+ pages_above = info.get('pages_above', 0)+ v_pct = int(info.get('vertical_scroll_percentage', 0))++ if pages_below > 0 or pages_above > 0:+ return f'scroll: {pages_above:.1f}↑ {pages_below:.1f}↓ {v_pct}%'++ return 'scroll'+scroll_info = self.scroll_infoif not scroll_info:return ''parts = []- # Vertical scroll info+ # Vertical scroll info (concise format)if scroll_info['scrollable_height'] > scroll_info['visible_height']:- parts.append(- f'scroll: {scroll_info["pages_above"]:.1f} pages above, '- f'{scroll_info["pages_below"]:.1f} pages below, '- f'at {scroll_info["vertical_scroll_percentage"]:.0f}% vertically'- )+ parts.append(f'{scroll_info["pages_above"]:.1f} pages above, {scroll_info["pages_below"]:.1f} pages below')- # Horizontal scroll info (if significant)+ # Horizontal scroll info (concise format)if scroll_info['scrollable_width'] > scroll_info['visible_width']:- parts.append(f'horizontal scroll: at {scroll_info["horizontal_scroll_percentage"]:.0f}%')-- # Can scroll directions- scroll_directions = []- if scroll_info['can_scroll_up']:- scroll_directions.append('↑up')- if scroll_info['can_scroll_down']:- scroll_directions.append('↓down')- if scroll_info['can_scroll_left']:- scroll_directions.append('←left')- if scroll_info['can_scroll_right']:- scroll_directions.append('→right')-- if scroll_directions:- parts.append(f'can scroll: {", ".join(scroll_directions)}')-- return ' | '.join(parts)+ parts.append(f'horizontal {scroll_info["horizontal_scroll_percentage"]:.0f}%')++ return ' '.join(parts)@propertydef element_hash(self) -> int: