[sorinc03]

Summary

Addresses the repeated accessibility-description traversal reported in #59743.

getTalkbackDescription already initializes a child node to check speaking/focusability. Passing null into the recursive call creates and initializes that child again. If its delegate synthesizes a description, that repeats the subtree recursively.

Reuse the initialized child node instead. An eight-group/one-text regression fixture goes from 511 node initializations to 9 while retaining the same description. This does not gate accessibility on TalkBack, disable Content Capture, or add a persistent cache.

Changelog

[ANDROID] [FIXED] - Avoid redundant subtree initialization while generating accessibility descriptions

Test Plan

./gradlew :packages:react-native:ReactAndroid:testDebugUnitTest \
  -Preact.internal.useHermesNightly=false \
  --tests com.facebook.react.uimanager.ReactAccessibilityDelegateTest
./gradlew :packages:react-native:ReactAndroid:testDebugUnitTest \
  -Preact.internal.useHermesNightly=false

Java 17 and the Android SDK configured locally.

• Before the production change: 22 focused tests, two expected traversal-count failures (nested counts 1/2/4/8/16/32/64/128/256, and an unlabelled wrapper initialized twice).
• After: all 22 focused tests pass. Six added cases cover traversal counts, label order, focus/hidden boundaries, EditText text priority, disabled accessibility services, and fresh text on subsequent requests.
• Full Debug JVM unit-test task passed: 863 tests across 119 suites, no failures/errors/skips. Repeated on final sources; parent review rerun completed up-to-date successfully.
• Changed Kotlin formatting and git diff --check pass.

No physical Pixel was available. The reporter's release frame-time measurements are not measurements from this patch; Content Capture frame-time improvement and on-device TalkBack behavior still need confirmation. Existing speaking-descendant scans and separate per-view requests may still repeat work; this is not a claim of globally linear traversal.

[sorinc03]

@LonelyCpp This removes the repeated child-node initialization described in your profile. The nested regression fixture goes from 511 initializations to 9 with the same description, and the full Debug JVM task passes 863 tests. I do not have a connected Pixel to measure Content Capture frame times; could you retest this change with your release-build A/B procedure and TalkBack enabled/disabled? The PR deliberately does not claim a measured device-frame improvement yet.