Describes the bug
An <img> tag whose declared width exceeds the actual width of the column EnrichedText renders it into fails to paint on Android. This is fully deterministic — not a timing race — and reproduces on the very first cold launch, every time.
I originally filed this issue theorizing an unsynchronized bounds race in AsyncDrawable (ImageDecoder.setTargetSize reading a still-zero Rect before the text-layout pass assigns real bounds). That was a plausible read of the source, but after extensive testing (cold cache, warm cache, a local low-latency image server, and the exact resolve-then-remount sequence a real consuming app uses, run concurrently across 16 images) I was never able to reproduce it. Everything I've actually observed — including the original production bug that led me to file this — is fully explained by the mechanism below instead, with no timing element at all. Retracting that theory.
The failure takes two different shapes depending on context:
- The
<img> is the only content in the EnrichedText — the failure is completely silent. No exception, no broken-image placeholder, nothing logged anywhere.
- There's any other text in the same
EnrichedText (true of essentially all real admin-authored content — headings, paragraphs, etc. around an image) — the same failure now logs a concrete diagnostic:
ReactNativeJNI: EnrichedTextView returned an invalid measurement. Min: [0,0] Max: [384,inf] Actual: [400,404.333]
Max is the real available column width; Actual is what the measure function reported back, driven by the embedded image's declared size.
This split traces to MeasurementStore.measure(): single-line content reports StaticLayout.getLineWidth(0), but multi-line content reports StaticLayout.width instead — and that can still exceed the nominal wrap width when one line contains an unsplittable inline image wider than the rest.
To Reproduce
Minimal repro app: https://github.com/ekrapfl/rne-html-repro-android-img-paint
npm install && npm run android (native dev build, Fabric).- Three cases render: an image that fits (paints normally), the same image declared far wider than the screen with no other content (renders at full size, overflowing the screen edge uncropped), and that same oversized image with a paragraph before it (reserves no visible space, shows nothing).
adb logcat | grep -i "invalid measurement" around the third case.
Expected behavior
An oversized declared width shouldn't cause total, silent render failure. At minimum, a diagnostic should surface consistently (right now it only does when there's other text in the same view); ideally the library would clamp/scale the image to the available column itself, the way a browser does with img { max-width: 100% }, rather than leaving every consumer to compute and inject a safe pixel width themselves.
Device tested
- Pixel 10 Pro XL emulator (
sdk_gphone16k_arm64), Android API 37
react-native-enriched-html 1.1.1, React Native 0.86.3, Expo ~57.0.23, Fabric (New Architecture)
Additional context
One honest caveat: the visual failure reproduces reliably in the minimal repro above, on a fresh cold install, every time. The exact JNI log line was confirmed repeatedly against a full production app under a matching layout structure, but hasn't fired in the standalone repro despite matching nesting depth, padding, and styling — something about the fuller app context matters there that I haven't isolated. The visual symptom (reserves no space, never paints) is the same failure either way.
Describes the bug
An
<img>tag whose declaredwidthexceeds the actual width of the columnEnrichedTextrenders it into fails to paint on Android. This is fully deterministic — not a timing race — and reproduces on the very first cold launch, every time.I originally filed this issue theorizing an unsynchronized bounds race in
AsyncDrawable(ImageDecoder.setTargetSizereading a still-zeroRectbefore the text-layout pass assigns real bounds). That was a plausible read of the source, but after extensive testing (cold cache, warm cache, a local low-latency image server, and the exact resolve-then-remount sequence a real consuming app uses, run concurrently across 16 images) I was never able to reproduce it. Everything I've actually observed — including the original production bug that led me to file this — is fully explained by the mechanism below instead, with no timing element at all. Retracting that theory.The failure takes two different shapes depending on context:
<img>is the only content in theEnrichedText— the failure is completely silent. No exception, no broken-image placeholder, nothing logged anywhere.EnrichedText(true of essentially all real admin-authored content — headings, paragraphs, etc. around an image) — the same failure now logs a concrete diagnostic:Maxis the real available column width;Actualis what the measure function reported back, driven by the embedded image's declared size.This split traces to
MeasurementStore.measure(): single-line content reportsStaticLayout.getLineWidth(0), but multi-line content reportsStaticLayout.widthinstead — and that can still exceed the nominal wrap width when one line contains an unsplittable inline image wider than the rest.To Reproduce
Minimal repro app: https://github.com/ekrapfl/rne-html-repro-android-img-paint
npm install && npm run android(native dev build, Fabric).adb logcat | grep -i "invalid measurement"around the third case.Expected behavior
An oversized declared width shouldn't cause total, silent render failure. At minimum, a diagnostic should surface consistently (right now it only does when there's other text in the same view); ideally the library would clamp/scale the image to the available column itself, the way a browser does with
img { max-width: 100% }, rather than leaving every consumer to compute and inject a safe pixel width themselves.Device tested
sdk_gphone16k_arm64), Android API 37react-native-enriched-html1.1.1, React Native 0.86.3, Expo ~57.0.23, Fabric (New Architecture)Additional context
One honest caveat: the visual failure reproduces reliably in the minimal repro above, on a fresh cold install, every time. The exact JNI log line was confirmed repeatedly against a full production app under a matching layout structure, but hasn't fired in the standalone repro despite matching nesting depth, padding, and styling — something about the fuller app context matters there that I haven't isolated. The visual symptom (reserves no space, never paints) is the same failure either way.