For now, I'm just guessing that the builder is failing due to drive caching. But really, I'd also expect it to fail for some mechanical drives (with large caches) too. So not 100% sure.

What I'm trying now is simply disabling the "sys_reboot()" call in the Linux kernel within the .upgrade file. So the builder can work as-is, but it simply won't be able to reboot by itself. If this fixes the problem, then yeah, a caching issue. If not, then.. something else is wrong?

We'll see.. pumping hda5 right now..

EDIT: Well, so much for that theory. A manual reboot after applying the .upgrade still fails randomly. I'm going to look for a mechanical drive to test with, just to validate the setup here.

When I read-back from the failed /dev/hda5 partition on my PC, it is not corrupted, and the contents match what should be there. So.. why no workie???