ship-a-fix-roundlisted

# Ship a fix round A device-testing report is a batch, and the batch is the unit of work: the user is waiting for one build that answers every finding, not a trickle of half-verified edits. The failure this prevents is the round that looks done, fixes pushed and checks green, while a finding was quietly narrowed, a golden silently regenerated, or the build handed over predates the last fix. ## Procedure 1. **Root-cause every finding before fixing any.** Read the code path, then reproduce: on the emulator when the claim is visual or gestural, in a JVM test when a test can hold it. Write down, per finding, what is wrong and why; a finding that cannot be reproduced is reported as such, with what was tried, never silently dropped. If the emulator misbehaves, read the newest ANR trace's main-thread stack: app frames mean the app, pure renderer or system frames mean the emulator, so cold-boot it instead of debugging the app on it. ```bash adb root >/dev/null 2>&1 # refused on a production image, which is fine trace=$(adb shell ls -t /data/anr 2>/dev/null | head -1 | tr -d '\r') if [ -n "$trace" ] then adb shell cat "/data/anr/$trace" else adb bugreport anr.zip # no root: the traces are inside the report fi ``` Pick the newest rather than naming one: the directory holds every trace the device has kept, and the one you want is from the run you just watched. 2. **Fix minimally, and prove each check can fail.**