max / alloy

Measure bootc switch against a layered machine, which the flip made load-bearing Every Alloy machine permanently carries a layer since cad26d5, and the harness README's standing result is that bootc refuses to upgrade a machine in that state. Read quickly, that closes the bootc path entirely, and the middle line of Staleness::REMEDY is a bootc switch. So the console has been printing a remedy nobody had checked still works, and the obvious correction would have been to rewrite it onto rpm-ostree upgrade. It works. Measured on the same versions as the original run, from a c1 machine holding a layered alloy-demo, switching to c2: upgrade errors with "Deployment contains local rpm-ostree modifications" and switch queues the deployment and reboots into it. The two verbs differ, and only upgrade consults local modifications. Nothing about the screen needs changing, which is worth more than a fix would have been. What switch does is discard the layer outright rather than leave it inactive: after the reboot the package is gone from both packages and requested-packages and the binary is absent, with the layered state left on the rollback deployment where a rollback can still reach it. On a real machine that self-heals, because alloy-layer-components.service gates on ConditionPathExists=!/usr/bin/alloy read off the filesystem rather than off a stamp, which is the case its own comment says it was written for. Confirmed by re-installing on the switched machine with the network repo disabled: it resolves 0.0.3 out of the carried file:// repo, which is c2's copy and not the version that had been layered. So on this path the component tracks the image with nothing owning it, and the drop-and-re-add that alloy update has to perform by hand on the rpm-ostree upgrade path happens for free. The cost is a second reboot the user is not told about. They run the three printed commands and land on a machine with no console and no terminal while the unit lays the new ones down and reboots again. d866e125 already records that silence against the first boot after an install; this widens it to every rebuild anyone adopts, which is both more common and the case where nobody has a reason to expect it. Recorded at the remedy as well as in the harness, because the harness is not where someone edits those three strings.