[torlando-tech]

Problem

Closes #1209. Sending a large image at Original quality crashed with Canvas: trying to draw too large(200540160 bytes) bitmap.

The composer preview decoded the selected image bytes at full resolution (MessagingScreen BitmapFactory.decodeByteArray). The ORIGINAL preset caps at 8192px, so the reporter's 8160×6144 photo passed through the pipeline unscaled; decoding it for the 80.dp preview allocated an 8160×6144×4 = 200,540,160-byte bitmap, which the hardware canvas refuses to draw above its 100MB limit (RecordingCanvas.MAX_BITMAP_SIZE).

Fix

Decode the preview bounds-first with power-of-two sampling via a new ImageUtils.decodeSampledBitmap(bytes, maxDimensionPx) (1024px cap) — the same strategy the message-bubble path already uses. Every live draw path now samples (preview, bubble, Micron images, Coil GIFs/fullscreen).

The send pipeline is unchanged: ORIGINAL still ships the full-resolution image. Only the preview is sampled, so image quality on the wire is preserved.

Metadata stripping still holds at ORIGINAL

The EXIF/GPS-stripping guarantee comes from decode → re-encode, not from the WebP container: loadBitmap decodes to raw pixels (baking EXIF orientation in) and Bitmap.compress writes a fresh file, so source APP1 metadata (GPS, device model) never survives — even though the preset path emits JPEG rather than WebP. A new regression test pins this: stamped GPS + rotate-90 input → ORIGINAL output has no GPS, claims no rotation, and carries the rotation in pixels.

Tests

ImageUtilsRobolectricTest (Robolectric, real compressImageWithPreset pipeline on a synthetic 8160×6144 JPEG):

• ORIGINAL preset preview fits hardware canvas limit — the red repro (200,540,160 bytes) now ≤ 100MB
• ORIGINAL preset preserves source dimensions on the wire — output stays 8160×6144 (guards against shrinking the send)
• ORIGINAL preset strips EXIF metadata from the sent image — privacy invariant: GPS never rides along at ORIGINAL quality
• HIGH preset preview fits hardware canvas limit — control
• decodeSampledBitmap samples large images below twice the target dimension

ImageUtilsRobolectricTest 38/38, MessagingScreenTest green, :app:detekt clean.

Fixes COLUMBA-EC

[greptile-apps]

Confidence Score: 5/5

[Low impact] The PR appears safe to merge; no actionable new issue was found.

Summary

The PR samples composer image previews without changing the image sent.

• The composer samples static images before showing their preview.

No actionable new issues were found. Tests were not run during this review.

_{Reviews (3) · Last reviewed commit: "test(ui): pin that preset re-encoding st..." · Reviewed by Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[torlando-tech]

@greptile review