[hryhoriiK97]

Summary:

On Android, every Text update goes through updateExtraData in the Text view manager (PreparedLayoutTextViewManager, and ReactTextViewManager for the older text view). There, the text's clickable spans are stored in a view tag and the accessibility delegate is reset, so that TalkBack can reach the links.

Current behavior. The comment says this is for a text view that "contains any clickable spans", but PreparedLayoutTextViewManager checks is Spanned, which is true for every Text, and ReactTextViewManager has no check at all. So every update of every Text creates a new ReactTextViewAccessibilityDelegate, even when TalkBack is off and the text has no links. Creating one is an ExploreByTouchHelper: it looks up the AccessibilityManager, makes the view focusable and then restores it, and allocates a Handler and a map. When the text has no links, nothing changes for the delegate, since it reads the links only when it's created.

With this change. Both managers call a new ReactTextViewAccessibilityDelegate.updateAccessibilityLinks. When the view already has a ReactTextViewAccessibilityDelegate and the text has no links, before or after the update, it keeps that delegate. In every other case (the first update, links added, changed or removed, or another delegate on the view) it resets the delegate as before. A plain Text still gets the delegate on its first update; PreparedLayoutTextView needs it to expose its text to TalkBack.

Setting a delegate through ViewCompat also makes a view whose importantForAccessibility is auto important. The kept path does the same, so a Text set back to auto (by the prop, or by view recycling) still ends up important, as before. The one difference: the accessibility actions a delegate caches from the accessibilityActions prop now stay between text updates, as they already do for every other view, instead of being dropped with the old delegate.

Changelog:

[ANDROID] [CHANGED] - Text updates keep the accessibility delegate when the text has no links

Test Plan:

• The pinned ktfmt (--dry-run --set-exit-if-changed) passes on all changed files.
• Added ReactTextViewAccessibilityDelegateTest (4 Robolectric tests: the first update installs the text delegate; updates without links keep the same delegate; adding and removing links resets it; a view set back to auto is important after an update). With the old code only "updates without links keep the delegate" fails. PreparedLayoutTextViewAccessibilityTextTest and ReactTextViewPropertyTest still pass.
• Device: Samsung Galaxy A16 (SM-A165F, Android 16), RNTester built with Gradle in Release (assembleRelease, arm64-v8a) and compiled with cmd package compile -m speed. The scenarios are a local Playground screen, not part of this PR: 200 numbers that change every 100 ms, and 60 lines of mixed-script text (emoji, CJK, Arabic, Hebrew, Polish) that change every 100 ms.
• A/B setup: a single APK with a local switch (an adb shell setprop, not part of this PR) that runs either the old reset on every update or this change. Launches alternate A B B A A B B A, 4 per arm and scenario; 15 s Perfetto traces with the app's trace sections. The changes are medians.

	before	this PR	change
updateExtraData per call, numbers	36.6–39.1 µs	28.4–30.0 µs	−21%
Mount per commit, numbers	12.65–13.20 ms	10.74–11.35 ms	−13%
Main thread CPU in 15 s, numbers	3,614–3,654 ms	3,484–3,625 ms	−3%
updateExtraData per call, mixed scripts	48.3–55.4 µs	38.5–41.1 µs	−28%
Mount per commit, mixed scripts	5.49–6.06 ms	4.79–5.07 ms	−18%
Main thread CPU in 15 s, mixed scripts	2,858–2,950 ms	2,804–2,849 ms	−4%

These times include the cost of the trace sections themselves, about half of updateExtraData's samples in a sampled profile, so without tracing the relative gain is larger.

• Accessibility: uiautomator dump of TextExample, AccessibilityExample and AccessibilityAndroidExample, 4 scroll positions each, with the old and the new code: the same 166 nodes, with the same text, content description and focusable, clickable and important-for-accessibility flags (positions differ only by the scroll). I couldn't run TalkBack on the device I used.

🤖 Generated with Claude Code

[github-actions]

Visual regression

No visual differences at 834bd43677d7.

ios: no capture

[YevheniiKotyrlo]

Device measurement — with and without this change

I'm carrying this change in an app on React Native 0.86.3 and measured it on two phones.

Without it — each PreparedLayoutTextViewManager.updateExtraData call takes a median of 16.4 µs on a Galaxy S23 and 2.3 µs on a RedMagic NX809J.

With it — the median is 8.9 µs on the S23 (−46%) and 1.6 µs on the RedMagic (−30%).

updateExtraData per call, median of each run	without	with
Galaxy S23, Android 14	16.3 µs, 16.5 µs	8.9 µs, 8.9 µs
RedMagic NX809J, Android 16	2.3 µs, 2.3 µs	1.6 µs, 1.6 µs

Release builds of the same app with and without this PR's three files, react-native 0.86.3 compiled from source, each run flinging a list that mounts rows of Text; the durations are the method's trace slices, recorded by perfetto with the app's trace sections on, so both sides include the section's own cost.

Every run

Device	Side	Calls	Median	Interquartile range	Mean	p90
Galaxy S23	without	799	16.25 µs	12.86–22.29 µs	19.36 µs	31.35 µs
Galaxy S23	without	799	16.51 µs	13.28–21.15 µs	18.61 µs	26.20 µs
Galaxy S23	with	1130	8.85 µs	6.72–14.38 µs	11.52 µs	20.26 µs
Galaxy S23	with	799	8.86 µs	6.09–11.46 µs	10.02 µs	14.69 µs
RedMagic NX809J	without	753	2.29 µs	1.67–2.50 µs	2.57 µs	2.92 µs
RedMagic NX809J	without	642	2.29 µs	1.62–2.50 µs	2.73 µs	3.28 µs
RedMagic NX809J	with	753	1.62 µs	1.30–2.19 µs	2.20 µs	3.39 µs
RedMagic NX809J	with	642	1.62 µs	1.36–2.03 µs	2.20 µs	3.18 µs

The runs alternated without, with, with, without on each phone. Two runs on the same side differ by at most 0.26 µs in median.

Query

SELECT slice.dur FROM slice
JOIN thread_track ON slice.track_id = thread_track.id
WHERE thread_track.utid = <the app's main thread>
  AND slice.name = 'PreparedLayoutTextViewManager.updateExtraData'

[hryhoriiK97]

@javache could you take a look at this one? There's a device measurement in the comment above: updateExtraData is 30–46% faster per call on two phones.

[meta-codesync]

@javache has imported this pull request. If you are a Meta employee, you can view this in D130917594.

[react-native-bot]

This pull request was successfully merged by @hryhoriiK97 in 502f112

^{When will my fix make it into a release? | How to file a pick request?}