Summary

Fixes #50797. A view with a non-invertible transform (e.g. transform: [{scaleY: 0}] or {scaleX: 0}) still received touches on both Android and iOS. On Android the view appeared to "inherit" a hit region from another view in the hierarchy; on iOS the view registered taps along the collapsed axis. Both symptoms collapse to the same native behaviour: when a transform can't be inverted, the platform APIs silently fall back to stale or degenerate data during hit testing.

Root cause

Android. TouchTargetHelper.getChildPoint calls Matrix.invert(inverseMatrix) without checking its return value. Matrix.invert returns false for a non-invertible matrix and leaves its destination unchanged. inverseMatrix is a class-level field reused across every child-point conversion, so on failure we apply the previous view's inverse to the current view's touch point: the exact "inherits from another view" behavior. It is also why shrinking a view from scaleY: 0.9 to 0.0 leaves a 90 % hit region: the 0.9 frame populated the cache.

iOS. CGAffineTransformInvert is documented to return the original matrix when it can't invert. -[UIView convertPoint:fromView:] uses that "inverse" during hit testing, so the touch point gets collapsed onto the degenerate axis (e.g. y = 0 for scaleY: 0) and the inherited pointInside: still reports a hit against a visually invisible view.

Fix

• Android (TouchTargetHelper.kt): getChildPoint now returns Boolean. It returns false when Matrix.invert fails; the child iteration loop skips such children. One behaviour change, no epsilon tuning, no shared state leaks. Resolves both symptom variants via the same code path.
• iOS (RCTViewComponentView.mm, Fabric): add a small RCTLayerTransformCollapsesAxis helper that tests the determinant of the 2 × 2 XY projection of layer.transform and return NO from the existing pointInside: override when it collapses.
• Tests. TouchTargetHelperTest (new, Robolectric): initial scaleX/scaleY: 0, zero-scale parent, the "inherits from sibling" regression, the 0.9 to 0.0 transition, and the touch-path accumulator. RCTViewComponentViewTests gets four parallel iOS cases. All pass locally.
• RNTester. New Transforms entry "Zero-scale hit test (regression for scale 0 touchable is clickable, inherits dimensions from random views #50797)". Reproduces both variants and exposes a scale input so you can step through the transition interactively.

Why not #53769?

PR #53769 solved the Android side with a hasZeroScale(view) early-return plus an epsilon comparison on individual matrix entries, and handled the invert failure in getChildPoint by falling back to untransformed coordinates. That still lets the degenerate view be hit with its pre-transform bounds; the early-return then has to catch the miss. This branch fixes the root cause once: if the matrix can't be inverted we don't descend into the child. The iOS fix is the matching change on the Fabric side.

Changelog:

[GENERAL] [FIXED] - Views with a non-invertible transform (e.g. scaleX: 0 or scaleY: 0) no longer receive touches on Android or iOS.

Test Plan

Unit tests

• ./gradlew :packages:react-native:ReactAndroid:testDebugUnitTest --tests 'com.facebook.react.uimanager.TouchTargetHelperTest*': 7/7 pass.
• iOS build-for-testing of RNTesterUnitTests scheme succeeds; the four new RCTViewComponentViewTests cases live next to the existing cases in React/Tests/Mounting/.
• ./gradlew ktfmtCheck, clang-format --dry-run -Werror, prettier --check, flow, and eslint all pass.

Manual verification (Android)

RNTester → Components → Transforms → "Zero-scale hit test (regression for #50797)".

Variant 1: tap the scaleY: 0 row directly. "Last tapped:" must stay (none).

Before	After
Last tapped: zero-scale (bug)	Last tapped: (none)

Variant 2: tap the scaleY: 0.5 row (①), then tap the scaleY: 0 row (②). Without the fix, the shared inverseMatrix cache populated by tap ① leaks into the zero-scale hit test on tap ②. After the fix, tap ② is correctly ignored and "Last tapped" still reflects tap ①.

Before	After
Last tapped: zero-scale (stale cache)	Last tapped: variable (scaleY=0.5)

iOS behaves similarly. I tried the example on iPhone 17 Pro (Simulator, iOS 26.1) and the zero-scale pressable is correctly untouchable; driving synthetic taps on the simulator wasn't scripted here, so the iOS signal is the four XCTest cases plus manual verification.