A Coding Agent Tried to Fix My CameraX Rotation Bug by Waiting. I Asked for Proof Instead.
My Android barcode scanner had a CameraX bug: after rotating a tablet through 180°, the preview still looked correct, but the frames received by the analyzer were upside down. The interesting part wasn't the final fix. It was deciding when we actually had enough evidence to trust it. A coding agent handled most of the debugging: reading the code, adding instrumentation, running adb logcat ,…
A coding agent attempted to resolve a CameraX rotation bug in an Android barcode scanner, but the process relied heavily on waiting for evidence. The bug caused the preview to appear upside down after a 180° tablet rotation, with the analyzer still receiving correctly oriented frames. The agent, responsible for most of the debugging, read the code, added instrumentation, ran adb logcat, changed the implementation, and checked results. Meanwhile, the reporter handled the physical tablet.
The debugging loop was unusual, with the agent starting adb logcat while the reporter rotated the tablet. They collected new values and adjusted instrumentation, repeating the process until they had enough evidence to compare. Before the rotation, the configuration orientation was 2, display rotation was 1, image target rotation was 1, and image rotation was 0.
After the rotation, the configuration orientation remained 2, display rotation became 3, image target rotation remained 1, and image rotation stayed at 0. The issue was that the image analysis target rotation was not being updated when the device rotated through 180°.
During an earlier experiment, the agent noticed that the display change callback had already fired before the previewView's display rotation value updated. The agent concluded that waiting for the previewView's display rotation to change before rebinding was necessary. However, the agent stopped working at that point, as they did not know the correct fix. They decided to ask the agent to prove the cause first by measuring the states that should agree.
The agent added diagnostic values to the logs: configuration orientation, display rotation, preview target rotation, image target rotation, and image proxy rotation degrees. They repeated the physical experiment, collecting the output and adjusting instrumentation. The logs showed that the tablet moved from ROTATION_90 to ROTATION_270, keeping Configuration.orientation at 2.
The display already reported rotation 3, while the analyzer retained its old target: display ROTATION_270, ImageAnalysis.targetRotation stayed at ROTATION_90, and ImageProxy.rotationDegrees remained at 0. This provided a specific, measured cause for the bug.
The fix involved updating existing use cases to directly set both Preview.targetRotation and ImageAnalysis.targetRotation with calculated rotation. The existing Preview and ImageAnalysis instances remained unchanged. The subtle mapping detail was that the degrees reported by OrientationEventListener couldn't be directly mapped to Surface.ROTATION_* constants.
After updating the implementation, the experiment confirmed that CameraX received the new targetRotation and ImageProxy.rotationDegrees changed from 0 to 180 without requiring recreating use cases or rebinding. The agent handled the mechanical work, while the reporter contributed by designing the experiment, physically rotating the tablet, evaluating hypotheses, and deciding what constituted sufficient evidence.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
