The hands-free boot testing loop is the unsung hero of this writeup. Two-minute test cycles with an automatic fallback slot turn what would be a terrifying flash-and-pray workflow into something almost civilized. I would love to see the same logger trick adopted by more device porters, since dead-before-adb boots are exactly where debugging goes to die.
Exactly, the A/B slot trick is what makes a borked boot cheap instead of scary. Do you persist the pre-flash logs automatically, or is capturing the failure state still a manual step?
All real time - so claude code keeps one A or B working to pull logs as needed. The real trick is the loop driven dev- on a crash - pull the logs, analyze on host system, then build and push back to the device
That loop makes sense, because the fallback slot turns a crash into data instead of a dead end. When a crash log points to a different subsystem than expected, do you restart the loop from the kernel build, or do you first test a smaller device tree change?
The hands-free boot testing loop is the unsung hero of this writeup. Two-minute test cycles with an automatic fallback slot turn what would be a terrifying flash-and-pray workflow into something almost civilized. I would love to see the same logger trick adopted by more device porters, since dead-before-adb boots are exactly where debugging goes to die.
Claude has full visibliity and can use a/b partition tricks when a boot got borked.
Exactly, the A/B slot trick is what makes a borked boot cheap instead of scary. Do you persist the pre-flash logs automatically, or is capturing the failure state still a manual step?
All real time - so claude code keeps one A or B working to pull logs as needed. The real trick is the loop driven dev- on a crash - pull the logs, analyze on host system, then build and push back to the device
That loop makes sense, because the fallback slot turns a crash into data instead of a dead end. When a crash log points to a different subsystem than expected, do you restart the loop from the kernel build, or do you first test a smaller device tree change?
I feed it back to Claude code and do a review for the next steps.
getting real close on an android 17 based lineage 24.x ROM with this approach
Kernel 4.9 isn't actually dead. What's dead is only the willingness and effort from individual developers.
Some 4.9 kernels even use the BPF backport from kernel version 5.15, and are already running Android 17 with LineageOS 24.
Take a look at this: xdaforums.com/t/development-rom-on...
This could potentially be applied to Pixel devices as well.