[0.83] fix(fabric): touch-input reliability — primary-button labeling, guaranteed touch END/CANCEL, phantom active-touch self-heal by FaithfulAudio · Pull Request #16333 · microsoft/react-native-windows

FaithfulAudio · 2026-07-24T14:57:23Z

Problem

On real touch hardware, Fabric/Composition touch input fails in ways mouse input does not (verified by differential testing: synthetic InjectTouchInput touch vs mouse_event on identical controls):

finger taps are ignored by press machinery while mouse clicks work, and
after a press whose release never reaches the island (finger slide-off, direct-manipulation steal, popup surface), presses stop firing app-wide until restart.

Root causes (all in `CompositionEventHandler.cpp`)

button=-1 for touch: onPointerPressed maps ActiveTouch.button from PointerUpdateKind (mouse-only concept); touch falls to the default -1, and the derived W3C buttons bitmask becomes 0. The dispatched pointerdown says "no button pressed", so pointer-event–driven press logic ignores finger contacts. W3C requires a touch/pen contact to be the primary button.
Leaked active touches: onPointerReleased returned before m_activeTouches.erase(...) when the up hit-tested to no component (tag == -1); onPointerCaptureLost released captures but never dispatched touch Cancel nor erased entries; onPointerPressed's stale-id guard returned, silently discarding new presses on a reused pointer id. A leaked entry is included in every subsequent DispatchTouchEvent (the loop iterates all of m_activeTouches), so the responder treats each new tap as corrupt multi-touch with a phantom finger down — presses animate but never commit.

Fix

onPointerPressed: non-mouse contacts that fell through to button = -1 get button = 0 (⇒ buttons = 1 while down) — W3C primary-button semantics for touch/pen.
onPointerReleased: dispatch the touch End and erase the entry even when tag == -1; only the target-specific OnPointerReleased requires a valid tag.
onPointerCaptureLost: dispatch touch Cancel for all active touches and clear them (proper responder reset), in addition to the existing capture release.
Self-heal in onPointerPressed: when a new primary touch contact arrives (IsPrimary() — a genuine second finger is non-primary and unaffected), cancel any leftover active touches; plus a 30s staleness backstop, and the stale-id guard now drops the leaked entry and continues instead of discarding the press.

Validation

Compiled into Microsoft.ReactNative from source (UseExperimentalNuget=false) in a production RNW 0.83.2 new-arch app (Facilitron FIT), x64 Release, on a physical Surface-class touch tablet:

Finger taps on Pressable/TouchableOpacity now fire (previously mouse-only) ✅
Slide-off presses / stolen contacts no longer wedge the app; the next tap lands on a clean responder ✅
Instrumented before/after: orphaned touchStarts and phantom double-dispatches present before, absent after ✅

Caveats for reviewers

Authored against 0.83-stable to match our production app; happy to re-cut onto main.
The self-heal uses Inner().Properties().IsPrimary() via the implementation type; if there's a preferred projection-level accessor, glad to switch.
Mirrors the TextInput fixes from the same hardening pass ([0.83] fix(textinput): center single-line text correctly at >100% display scale #16302, [0.83] fix(textinput): correct placeholder layout constraints (px vs DIP) and no-op NaN fontSize guard #16303, fix(textinput): show touch keyboard on focus and on tap (windowless island; re-tap reopens) #16330).

Microsoft Reviewers: Open in CodeFlow

…uch, guaranteed touch END/CANCEL on missed releases and capture loss, phantom active-touch self-heal

azure-pipelines · 2026-07-24T14:57:32Z

Azure Pipelines: There may be pipelines that require an authorized user to comment /azp run to run.

…ve-touch entry onPointerPressed already detects a stale m_activeTouches entry for an OS-reused pointer id and erases it so the new press can proceed. But the leaked entry's touchStart may already have reached JS - erasing silently leaves the JS gesture responder convinced that finger is still down, a wedge no later native cleanup can undo once the entry is gone. Dispatch TouchEventType::Cancel for the stale entry before erasing it, matching the other three cleanup sites in this PR.

FaithfulAudio · 2026-07-26T06:56:22Z

Pushed one more cleanup site: when onPointerPressed finds a stale m_activeTouches entry for an OS-reused pointer id, it now dispatches TouchEventType::Cancel for that entry before erasing it, instead of erasing silently.

Reason: the leaked entry's touchStart may already have reached JS. Erasing without a Cancel leaves the JS gesture responder believing that finger is still down, and once the native entry is gone no later native cleanup can reset it. The other cleanup paths in this PR already Cancel before erasing; this makes the stale-entry path consistent with them.

Validated in the same production 0.83.2 new-arch app as the rest of the PR (from-source framework build): an injected-touch soak — real PT_TOUCH contacts via InjectTouchInput — of 20 taps including capture-loss and scroll-steal sequences registered 20/20 with no stuck pressed state.

Copilot

Pull request overview

This PR hardens Fabric/Composition input handling on Windows to make touch input behave reliably like mouse input, specifically by aligning pointer-event button semantics for touch/pen and ensuring active touches can’t be stranded across missed releases/capture loss.

Changes:

Ensure non-mouse contacts dispatch with W3C primary-button semantics (button=0 ⇒ buttons=1) so touch presses are recognized by pointer-event–driven press logic.
Prevent leaked/stranded m_activeTouches by dispatching End even when hit-test misses, dispatching Cancel on capture loss, and adding “self-heal” cleanup for stale/leftover touches.
Add a change file for a patch release entry.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
vnext/Microsoft.ReactNative/Fabric/Composition/CompositionEventHandler.cpp	Adjusts touch/pointer dispatch and cleanup behavior to prevent ignored taps and app-wide press “wedges” caused by leaked active touches.
change/react-native-windows-fix-fabric-touch-input-reliability.json	Records the patch-level change for release notes/versioning.

Suppressed comments (1)

vnext/Microsoft.ReactNative/Fabric/Composition/CompositionEventHandler.cpp:1402

When tag == -1, getTargetPointerArgs() calls RootComponentView().hitTest(...), and ViewComponentView::hitTest returns -1 without writing localPt. That leaves ptLocal uninitialized, but it is still passed to UpdateActiveTouch() (screen/offset points), causing undefined coordinates on touch End for slide-off/missed-hit releases.


    UpdateActiveTouch(activeTouch->second, ptScaled, ptLocal);
    DispatchTouchEvent(TouchEventType::End, pointerId, pointerPoint, keyModifiers);

+    try {
+      auto inner = winrt::get_self<winrt::Microsoft::ReactNative::Composition::Input::implementation::PointerPoint>(
+                       pointerPoint)
+                       ->Inner();
+      if (inner && inner.Properties().IsPrimary() && !m_activeTouches.empty()) {
+        std::vector<PointerId> leftoverIds;
+        leftoverIds.reserve(m_activeTouches.size());
+        for (auto const &entry : m_activeTouches) {
+          if (entry.second.touch.identifier != pointerId) {
+            leftoverIds.push_back(entry.first);
+          }
+        }
+        for (auto leftoverId : leftoverIds) {
+          DispatchTouchEvent(TouchEventType::Cancel, leftoverId, pointerPoint, keyModifiers);
+          m_activeTouches.erase(leftoverId);
+        }
+      }
+    } catch (...) {
+    }


+  if (!m_activeTouches.empty()) {
+    std::vector<PointerId> activeIds;
+    activeIds.reserve(m_activeTouches.size());
+    for (auto const &entry : m_activeTouches) {
+      activeIds.push_back(entry.first);
+    }
+    for (auto activeId : activeIds) {
+      DispatchTouchEvent(TouchEventType::Cancel, activeId, pointerPoint, keyModifiers);
+      m_activeTouches.erase(activeId);
+    }


fix(fabric): touch-input reliability — primary-button labeling for to…

2f426fd

…uch, guaranteed touch END/CANCEL on missed releases and capture loss, phantom active-touch self-heal

FaithfulAudio requested a review from a team as a code owner July 24, 2026 14:57

FaithfulAudio mentioned this pull request Jul 26, 2026

[0.83] fix(pointer): backport #16140 (PointerRoutedAway + ScrollView contentOffset DIP) #16335

Open

This was referenced Jul 26, 2026

fix(scrollview): honor programmatic scrollTo when scrollEnabled={false} #16304

Open

fix(pointer): label a touch/pen contact as the primary button so JS press handling fires #16338

Open

vmoroz requested a review from Copilot August 3, 2026 17:14

Copilot started reviewing on behalf of vmoroz August 3, 2026 17:14 View session

Copilot AI reviewed Aug 3, 2026

View reviewed changes

Navigation Menu

Uh oh!

[0.83] fix(fabric): touch-input reliability — primary-button labeling, guaranteed touch END/CANCEL, phantom active-touch self-heal - #16333

[0.83] fix(fabric): touch-input reliability — primary-button labeling, guaranteed touch END/CANCEL, phantom active-touch self-heal#16333
FaithfulAudio wants to merge 2 commits into
microsoft:0.83-stablefrom
FacilitronWorks:fix/fabric-touch-input-reliability

FaithfulAudio commented Jul 24, 2026 •

edited by microsoft-github-policy-service Bot

Loading

Uh oh!

azure-pipelines Bot commented Jul 24, 2026

Uh oh!

FaithfulAudio commented Jul 26, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

FaithfulAudio commented Jul 24, 2026 • edited by microsoft-github-policy-service Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Problem

Root causes (all in CompositionEventHandler.cpp)

Fix

Validation

Caveats for reviewers

Microsoft Reviewers: Open in CodeFlow

Uh oh!

azure-pipelines Bot commented Jul 24, 2026

Uh oh!

FaithfulAudio commented Jul 26, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

FaithfulAudio commented Jul 24, 2026 •

edited by microsoft-github-policy-service Bot

Loading

Root causes (all in `CompositionEventHandler.cpp`)