Source code

Revision control

Copy as Markdown

Other Tools

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at */
#include "PositionedEventTargeting.h"
#include "Units.h"
#include "mozilla/EventListenerManager.h"
#include "mozilla/MouseEvents.h"
#include "mozilla/Preferences.h"
#include "mozilla/PresShell.h"
#include "mozilla/StaticPrefs_dom.h"
#include "mozilla/StaticPrefs_ui.h"
#include "mozilla/ToString.h"
#include "mozilla/ViewportUtils.h"
#include "mozilla/dom/MouseEventBinding.h"
#include "mozilla/gfx/Matrix.h"
#include "mozilla/layers/LayersTypes.h"
#include "nsContainerFrame.h"
#include "nsCoord.h"
#include "nsFrameList.h" // for DEBUG_FRAME_DUMP
#include "nsHTMLParts.h"
#include "nsLayoutUtils.h"
#include "nsGkAtoms.h"
#include "nsFontMetrics.h"
#include "nsIContentInlines.h"
#include "nsPresContext.h"
#include "nsPrintfCString.h"
#include "mozilla/dom/Element.h"
#include "nsRegion.h"
#include "nsDeviceContext.h"
#include "nsIFrame.h"
#include <algorithm>
using namespace mozilla;
using namespace mozilla::dom;
// If debugging this code you may wish to enable this logging, via
// the env var MOZ_LOG="event.retarget:4". For extra logging (getting
// frame dumps, use MOZ_LOG="event.retarget:5".
static mozilla::LazyLogModule sEvtTgtLog("event.retarget");
#define PET_LOG(...) MOZ_LOG(sEvtTgtLog, LogLevel::Debug, (__VA_ARGS__))
namespace mozilla {
* The basic goal of FindFrameTargetedByInputEvent() is to find a good
* target element that can respond to mouse events. Both mouse events and touch
* events are targeted at this element. Note that even for touch events, we
* check responsiveness to mouse events. We assume Web authors
* designing for touch events will take their own steps to account for
* inaccurate touch events.
* GetClickableAncestor() encapsulates the heuristic that determines whether an
* element is expected to respond to mouse events. An element is deemed
* "clickable" if it has registered listeners for "click", "mousedown" or
* "mouseup", or is on a whitelist of element tags (<a>, <button>, <input>,
* <select>, <textarea>, <label>), or has role="button", or is a link, or
* is a suitable XUL element.
* Any descendant (in the same document) of a clickable element is also
* deemed clickable since events will propagate to the clickable element from
* its descendant.
* If the element directly under the event position is clickable (or
* event radii are disabled), we always use that element. Otherwise we collect
* all frames intersecting a rectangle around the event position (taking CSS
* transforms into account) and choose the best candidate in GetClosest().
* Only GetClickableAncestor() candidates are considered; if none are found,
* then we revert to targeting the element under the event position.
* We ignore candidates outside the document subtree rooted by the
* document of the element directly under the event position. This ensures that
* event listeners in ancestor documents don't make it completely impossible
* to target a non-clickable element in a child document.
* When both a frame and its ancestor are in the candidate list, we ignore
* the ancestor. Otherwise a large ancestor element with a mouse event listener
* and some descendant elements that need to be individually targetable would
* disable intelligent targeting of those descendants within its bounds.
* GetClosest() computes the transformed axis-aligned bounds of each
* candidate frame, then computes the Manhattan distance from the event point
* to the bounds rect (which can be zero). The frame with the
* shortest distance is chosen. For visited links we multiply the distance
* by a specified constant weight; this can be used to make visited links
* more or less likely to be targeted than non-visited links.
// Enum that determines which type of elements to count as targets in the
// search. Clickable elements are generally ones that respond to click events,
// like form inputs and links and things with click event listeners.
// Touchable elements are a much narrower set of elements; ones with touchstart
// and touchend listeners.
enum class SearchType {
struct EventRadiusPrefs {
bool mEnabled; // other fields are valid iff this field is true
uint32_t mVisitedWeight; // in percent, i.e. default is 100
uint32_t mRadiusTopmm;
uint32_t mRadiusRightmm;
uint32_t mRadiusBottommm;
uint32_t mRadiusLeftmm;
bool mTouchOnly;
bool mReposition;
SearchType mSearchType;
explicit EventRadiusPrefs(EventClassID aEventClassID) {
if (aEventClassID == eTouchEventClass) {
mEnabled = StaticPrefs::ui_touch_radius_enabled();
mVisitedWeight = StaticPrefs::ui_touch_radius_visitedWeight();
mRadiusTopmm = StaticPrefs::ui_touch_radius_topmm();
mRadiusRightmm = StaticPrefs::ui_touch_radius_rightmm();
mRadiusBottommm = StaticPrefs::ui_touch_radius_bottommm();
mRadiusLeftmm = StaticPrefs::ui_touch_radius_leftmm();
mTouchOnly = false; // Always false, unlike mouse events.
mReposition = false; // Always false, unlike mouse events.
mSearchType = SearchType::Touchable;
} else if (aEventClassID == eMouseEventClass) {
mEnabled = StaticPrefs::ui_mouse_radius_enabled();
mVisitedWeight = StaticPrefs::ui_mouse_radius_visitedWeight();
mRadiusTopmm = StaticPrefs::ui_mouse_radius_topmm();
mRadiusRightmm = StaticPrefs::ui_mouse_radius_rightmm();
mRadiusBottommm = StaticPrefs::ui_mouse_radius_bottommm();
mRadiusLeftmm = StaticPrefs::ui_mouse_radius_leftmm();
mTouchOnly = StaticPrefs::ui_mouse_radius_inputSource_touchOnly();
mReposition = StaticPrefs::ui_mouse_radius_reposition();
mSearchType = SearchType::Clickable;
} else {
mEnabled = false;
mVisitedWeight = 0;
mRadiusTopmm = 0;
mRadiusRightmm = 0;
mRadiusBottommm = 0;
mRadiusLeftmm = 0;
mTouchOnly = false;
mReposition = false;
mSearchType = SearchType::None;
static bool HasMouseListener(nsIContent* aContent) {
if (EventListenerManager* elm = aContent->GetExistingListenerManager()) {
return elm->HasListenersFor(nsGkAtoms::onclick) ||
elm->HasListenersFor(nsGkAtoms::onmousedown) ||
return false;
static bool HasTouchListener(nsIContent* aContent) {
EventListenerManager* elm = aContent->GetExistingListenerManager();
if (!elm) {
return false;
// FIXME: Should this really use the pref rather than TouchEvent::PrefEnabled
// or such?
if (!StaticPrefs::dom_w3c_touch_events_enabled()) {
return false;
return elm->HasNonSystemGroupListenersFor(nsGkAtoms::ontouchstart) ||
static bool HasPointerListener(nsIContent* aContent) {
EventListenerManager* elm = aContent->GetExistingListenerManager();
if (!elm) {
return false;
return elm->HasListenersFor(nsGkAtoms::onpointerdown) ||
static bool IsDescendant(nsIFrame* aFrame, nsIContent* aAncestor,
nsAutoString* aLabelTargetId) {
for (nsIContent* content = aFrame->GetContent(); content;
content = content->GetFlattenedTreeParent()) {
if (aLabelTargetId && content->IsHTMLElement(nsGkAtoms::label)) {
content->AsElement()->GetAttr(nsGkAtoms::_for, *aLabelTargetId);
if (content == aAncestor) {
return true;
return false;
static nsIContent* GetTouchableAncestor(nsIFrame* aFrame,
nsAtom* aStopAt = nullptr) {
// Input events propagate up the content tree so we'll follow the content
// ancestors to look for elements accepting the touch event.
for (nsIContent* content = aFrame->GetContent(); content;
content = content->GetFlattenedTreeParent()) {
if (aStopAt && content->IsHTMLElement(aStopAt)) {
if (HasTouchListener(content)) {
return content;
return nullptr;
static nsIContent* GetClickableAncestor(
nsIFrame* aFrame, nsAtom* aStopAt = nullptr,
nsAutoString* aLabelTargetId = nullptr) {
// If the frame is `cursor:pointer` or inherits `cursor:pointer` from an
// ancestor, treat it as clickable. This is a heuristic to deal with pages
// where the click event listener is on the <body> or <html> element but it
// triggers an action on some specific element. We want the specific element
// to be considered clickable, and at least some pages that do this indicate
// the clickability by setting `cursor:pointer`, so we use that here.
// Note that descendants of `cursor:pointer` elements that override the
// inherited `pointer` to `auto` or any other value are NOT treated as
// clickable, because it seems like the content author is trying to express
// non-clickability on that sub-element.
// In the future depending on real-world cases it might make sense to expand
// this check to any non-auto cursor. Such a change would also pick up things
// like contenteditable or input fields, which can then be removed from the
// loop below, and would have better performance.
if (aFrame->StyleUI()->Cursor().keyword == StyleCursorKind::Pointer) {
return aFrame->GetContent();
// Input events propagate up the content tree so we'll follow the content
// ancestors to look for elements accepting the click.
for (nsIContent* content = aFrame->GetContent(); content;
content = content->GetFlattenedTreeParent()) {
if (aStopAt && content->IsHTMLElement(aStopAt)) {
if (HasTouchListener(content) || HasMouseListener(content) ||
HasPointerListener(content)) {
return content;
if (content->IsAnyOfHTMLElements(nsGkAtoms::button, nsGkAtoms::input,
nsGkAtoms::select, nsGkAtoms::textarea)) {
return content;
if (content->IsHTMLElement(nsGkAtoms::label)) {
if (aLabelTargetId) {
content->AsElement()->GetAttr(nsGkAtoms::_for, *aLabelTargetId);
return content;
// See nsCSSFrameConstructor::FindXULTagData. This code is not
// really intended to be used with XUL, though.
if (content->IsAnyOfXULElements(
nsGkAtoms::button, nsGkAtoms::checkbox, nsGkAtoms::radio,
nsGkAtoms::menu, nsGkAtoms::menuitem, nsGkAtoms::menulist,
nsGkAtoms::scrollbarbutton, nsGkAtoms::resizer)) {
return content;
static Element::AttrValuesArray clickableRoles[] = {
nsGkAtoms::button, nsGkAtoms::key, nullptr};
if (auto* element = Element::FromNode(*content)) {
if (element->IsLink()) {
return content;
if (element->FindAttrValueIn(kNameSpaceID_None, nsGkAtoms::role,
clickableRoles, eIgnoreCase) >= 0) {
return content;
if (content->IsEditable()) {
return content;
return nullptr;
static Scale2D AppUnitsToMMScale(RelativeTo aFrame) {
nsPresContext* presContext = aFrame.mFrame->PresContext();
const int32_t appUnitsPerInch =
const float appUnits =
static_cast<float>(appUnitsPerInch) / MM_PER_INCH_FLOAT;
// Visual coordinates are only used for quantities relative to the
// cross-process root content document's root frame. There should
// not be an enclosing resolution or transform scale above that.
if (aFrame.mViewportType != ViewportType::Layout) {
const nscoord scale = NSToCoordRound(appUnits);
return Scale2D{static_cast<float>(scale), static_cast<float>(scale)};
Scale2D localResolution{1.0f, 1.0f};
Scale2D enclosingResolution{1.0f, 1.0f};
if (auto* pc = presContext->GetInProcessRootContentDocumentPresContext()) {
PresShell* presShell = pc->PresShell();
localResolution = {presShell->GetResolution(), presShell->GetResolution()};
enclosingResolution = ViewportUtils::TryInferEnclosingResolution(presShell);
const gfx::MatrixScales parentScale =
const Scale2D resolution =
localResolution * parentScale * enclosingResolution;
const nscoord scaleX = NSToCoordRound(appUnits / resolution.xScale);
const nscoord scaleY = NSToCoordRound(appUnits / resolution.yScale);
return {static_cast<float>(scaleX), static_cast<float>(scaleY)};
* Clip aRect with the bounds of aFrame in the coordinate system of
* aRootFrame. aRootFrame is an ancestor of aFrame.
static nsRect ClipToFrame(RelativeTo aRootFrame, const nsIFrame* aFrame,
nsRect& aRect) {
nsRect bound = nsLayoutUtils::TransformFrameRectToAncestor(
aFrame, nsRect(nsPoint(0, 0), aFrame->GetSize()), aRootFrame);
nsRect result = bound.Intersect(aRect);
return result;
static nsRect GetTargetRect(RelativeTo aRootFrame,
const nsPoint& aPointRelativeToRootFrame,
const nsIFrame* aRestrictToDescendants,
const EventRadiusPrefs& aPrefs, uint32_t aFlags) {
const Scale2D scale = AppUnitsToMMScale(aRootFrame);
nsMargin m(aPrefs.mRadiusTopmm * scale.yScale,
aPrefs.mRadiusRightmm * scale.xScale,
aPrefs.mRadiusBottommm * scale.yScale,
aPrefs.mRadiusLeftmm * scale.xScale);
nsRect r(aPointRelativeToRootFrame, nsSize(0, 0));
// Don't clip this rect to the root scroll frame if the flag to ignore the
// root scroll frame is set. Note that the GetClosest code will still
// enforce that the target found is a descendant of aRestrictToDescendants.
r = ClipToFrame(aRootFrame, aRestrictToDescendants, r);
return r;
static float ComputeDistanceFromRect(const nsPoint& aPoint,
const nsRect& aRect) {
nscoord dx =
std::max(0, std::max(aRect.x - aPoint.x, aPoint.x - aRect.XMost()));
nscoord dy =
std::max(0, std::max(aRect.y - aPoint.y, aPoint.y - aRect.YMost()));
return float(NS_hypot(dx, dy));
static float ComputeDistanceFromRegion(const nsPoint& aPoint,
const nsRegion& aRegion) {
"can't compute distance between point and empty region");
float minDist = -1;
for (auto iter = aRegion.RectIter(); !iter.Done(); iter.Next()) {
float dist = ComputeDistanceFromRect(aPoint, iter.Get());
if (dist < minDist || minDist < 0) {
minDist = dist;
return minDist;
// Subtract aRegion from aExposedRegion as long as that doesn't make the
// exposed region get too complex or removes a big chunk of the exposed region.
static void SubtractFromExposedRegion(nsRegion* aExposedRegion,
const nsRegion& aRegion) {
if (aRegion.IsEmpty()) {
nsRegion tmp;
tmp.Sub(*aExposedRegion, aRegion);
// Don't let *aExposedRegion get too complex, but don't let it fluff out to
// its bounds either. Do let aExposedRegion get more complex if by doing so
// we reduce its area by at least half.
if (tmp.GetNumRects() <= 15 || tmp.Area() <= aExposedRegion->Area() / 2) {
*aExposedRegion = tmp;
static nsIFrame* GetClosest(RelativeTo aRoot,
const nsPoint& aPointRelativeToRootFrame,
const nsRect& aTargetRect,
const EventRadiusPrefs& aPrefs,
const nsIFrame* aRestrictToDescendants,
nsIContent* aClickableAncestor,
nsTArray<nsIFrame*>& aCandidates) {
nsIFrame* bestTarget = nullptr;
// Lower is better; distance is in appunits
float bestDistance = 1e6f;
nsRegion exposedRegion(aTargetRect);
for (uint32_t i = 0; i < aCandidates.Length(); ++i) {
nsIFrame* f = aCandidates[i];
bool preservesAxisAlignedRectangles = false;
nsRect borderBox = nsLayoutUtils::TransformFrameRectToAncestor(
f, nsRect(nsPoint(0, 0), f->GetSize()), aRoot,
PET_LOG("Checking candidate %p with border box %s\n", f,
nsRegion region;
region.And(exposedRegion, borderBox);
if (region.IsEmpty()) {
PET_LOG(" candidate %p had empty hit region\n", f);
if (preservesAxisAlignedRectangles) {
// Subtract from the exposed region if we have a transform that won't make
// the bounds include a bunch of area that we don't actually cover.
SubtractFromExposedRegion(&exposedRegion, region);
nsAutoString labelTargetId;
if (aClickableAncestor &&
!IsDescendant(f, aClickableAncestor, &labelTargetId)) {
PET_LOG(" candidate %p is not a descendant of required ancestor\n", f);
if (aPrefs.mSearchType == SearchType::Clickable) {
nsIContent* clickableContent =
GetClickableAncestor(f, nsGkAtoms::body, &labelTargetId);
if (!aClickableAncestor && !clickableContent) {
PET_LOG(" candidate %p was not clickable\n", f);
} else if (aPrefs.mSearchType == SearchType::Touchable) {
nsIContent* touchableContent = GetTouchableAncestor(f, nsGkAtoms::body);
if (!touchableContent) {
PET_LOG(" candidate %p was not touchable\n", f);
// If our current closest frame is a descendant of 'f', skip 'f' (prefer
// the nested frame).
if (bestTarget && nsLayoutUtils::IsProperAncestorFrameCrossDoc(
f, bestTarget, aRoot.mFrame)) {
PET_LOG(" candidate %p was ancestor for bestTarget %p\n", f, bestTarget);
if (!aClickableAncestor && !nsLayoutUtils::IsAncestorFrameCrossDoc(
aRestrictToDescendants, f, aRoot.mFrame)) {
PET_LOG(" candidate %p was not descendant of restrictroot %p\n", f,
// distance is in appunits
float distance =
ComputeDistanceFromRegion(aPointRelativeToRootFrame, region);
nsIContent* content = f->GetContent();
if (content && content->IsElement() &&
ElementState(ElementState::VISITED))) {
distance *= aPrefs.mVisitedWeight / 100.0f;
if (distance < bestDistance) {
PET_LOG(" candidate %p is the new best\n", f);
bestDistance = distance;
bestTarget = f;
return bestTarget;
// Walk from aTarget up to aRoot, and return the first frame found with an
// explicit z-index set on it. If no such frame is found, aRoot is returned.
static const nsIFrame* FindZIndexAncestor(const nsIFrame* aTarget,
const nsIFrame* aRoot) {
const nsIFrame* candidate = aTarget;
while (candidate && candidate != aRoot) {
if (candidate->ZIndex().valueOr(0) > 0) {
PET_LOG("Restricting search to z-index root %p\n", candidate);
return candidate;
candidate = candidate->GetParent();
return aRoot;
nsIFrame* FindFrameTargetedByInputEvent(
WidgetGUIEvent* aEvent, RelativeTo aRootFrame,
const nsPoint& aPointRelativeToRootFrame, uint32_t aFlags) {
using FrameForPointOption = nsLayoutUtils::FrameForPointOption;
EnumSet<FrameForPointOption> options;
options += FrameForPointOption::IgnoreRootScrollFrame;
nsIFrame* target = nsLayoutUtils::GetFrameForPoint(
aRootFrame, aPointRelativeToRootFrame, options);
"Found initial target %p for event class %s message %s point %s "
"relative to root frame %s\n",
target, ToChar(aEvent->mClass), ToChar(aEvent->mMessage),
EventRadiusPrefs prefs(aEvent->mClass);
if (!prefs.mEnabled || EventRetargetSuppression::IsActive()) {
PET_LOG("Retargeting disabled\n");
return target;
// Do not modify targeting for actual mouse hardware; only for mouse
// events generated by touch-screen hardware.
if (aEvent->mClass == eMouseEventClass && prefs.mTouchOnly &&
aEvent->AsMouseEvent()->mInputSource !=
MouseEvent_Binding::MOZ_SOURCE_TOUCH) {
PET_LOG("Mouse input event is not from a touch source\n");
return target;
// If the exact target is non-null, only consider candidate targets in the
// same document as the exact target. Otherwise, if an ancestor document has
// a mouse event handler for example, targets that are !GetClickableAncestor
// can never be targeted --- something nsSubDocumentFrame in an ancestor
// document would be targeted instead.
const nsIFrame* restrictToDescendants = [&]() -> const nsIFrame* {
if (target && target->PresContext() != aRootFrame.mFrame->PresContext()) {
return target->PresShell()->GetRootFrame();
return aRootFrame.mFrame;
// Ignore retarget if target is editable.
nsIContent* targetContent = target ? target->GetContent() : nullptr;
if (targetContent && targetContent->IsEditable()) {
PET_LOG("Target %p is editable\n", target);
return target;
// If the target element inside an element with a z-index, restrict the
// search to other elements inside that z-index. This is a heuristic
// intended to help with a class of scenarios involving web modals or
// web popup type things. In particular it helps alleviate bug 1666792.
restrictToDescendants = FindZIndexAncestor(target, restrictToDescendants);
nsRect targetRect = GetTargetRect(aRootFrame, aPointRelativeToRootFrame,
restrictToDescendants, prefs, aFlags);
PET_LOG("Expanded point to target rect %s\n", ToString(targetRect).c_str());
AutoTArray<nsIFrame*, 8> candidates;
nsresult rv = nsLayoutUtils::GetFramesForArea(aRootFrame, targetRect,
candidates, options);
if (NS_FAILED(rv)) {
return target;
nsIContent* clickableAncestor = nullptr;
if (target) {
clickableAncestor = GetClickableAncestor(target, nsGkAtoms::body);
if (clickableAncestor) {
PET_LOG("Target %p is clickable\n", target);
// If the target that was directly hit has a clickable ancestor, that
// means it too is clickable. And since it is the same as or a
// descendant of clickableAncestor, it should become the root for the
// GetClosest search.
clickableAncestor = target->GetContent();
nsIFrame* closest =
GetClosest(aRootFrame, aPointRelativeToRootFrame, targetRect, prefs,
restrictToDescendants, clickableAncestor, candidates);
if (closest) {
target = closest;
PET_LOG("Final target is %p\n", target);
// At verbose logging level, dump the frame tree to help with debugging.
// Note that dumping the frame tree at the top of the function may flood
// logcat on Android devices and cause the PET_LOGs to get dropped.
if (MOZ_LOG_TEST(sEvtTgtLog, LogLevel::Verbose)) {
if (target) {
} else {
if (!target || !prefs.mReposition) {
// No repositioning required for this event
return target;
// Take the point relative to the root frame, make it relative to the target,
// clamp it to the bounds, and then make it relative to the root frame again.
nsPoint point = aPointRelativeToRootFrame;
if (nsLayoutUtils::TRANSFORM_SUCCEEDED !=
nsLayoutUtils::TransformPoint(aRootFrame, RelativeTo{target}, point)) {
return target;
point = target->GetRectRelativeToSelf().ClampPoint(point);
if (nsLayoutUtils::TRANSFORM_SUCCEEDED !=
nsLayoutUtils::TransformPoint(RelativeTo{target}, aRootFrame, point)) {
return target;
// Now we basically undo the operations in GetEventCoordinatesRelativeTo, to
// get back the (now-clamped) coordinates in the event's widget's space.
nsView* view = aRootFrame.mFrame->GetView();
if (!view) {
return target;
LayoutDeviceIntPoint widgetPoint = nsLayoutUtils::TranslateViewToWidget(
aRootFrame.mFrame->PresContext(), view, point, aRootFrame.mViewportType,
if (widgetPoint.x != NS_UNCONSTRAINEDSIZE) {
// If that succeeded, we update the point in the event
aEvent->mRefPoint = widgetPoint;
return target;
uint32_t EventRetargetSuppression::sSuppressionCount = 0;
EventRetargetSuppression::EventRetargetSuppression() { sSuppressionCount++; }
EventRetargetSuppression::~EventRetargetSuppression() { sSuppressionCount--; }
bool EventRetargetSuppression::IsActive() { return sSuppressionCount > 0; }
} // namespace mozilla