| Commit message (Collapse) | Author | Age | Files | Lines | |
|---|---|---|---|---|---|
| * | docs(spec): split the live upgrade and hold the kernel for a dedicated sessionHEADmain | Craig Jennings | 98 min. | 1 | -32/+80 |
| | | | | | | | | | | | | | | | I rewrote the guarded-upgrade spec around a script the maintenance panel calls: pacman -Syu --ignore on what the live-update guard would block plus the kernel set, then the AUR remainder and topgrade with the system step off. The boot oneshot now completes only the deferred GPU set. Kernels wait for a session I choose, behind a gate on the DKMS build, the initramfs, and a pre-pacman snapshot, because on velox a failed zfs-dkms rebuild after the swap can't be aborted and leaves the new kernel unbootable. All seven decisions are closed. Also closes the ratio Emacs 31 task (upgraded, verified on Wayland) and files gcalcli in the installer. | ||||
| * | docs: draft the guarded-upgrade spec and file the open velox items | Craig Jennings | 8 hours | 1 | -0/+165 |
| The topgrade freshness warning never clears because the stamp is written only when topgrade exits 0, and the live-update guard plus any failing ecosystem step makes that exit almost unreachable. The spec (DRAFT) weighs four ways to close that. Its open decisions are state versus recency, an armed boot-time oneshot, an arch-only boot run and a persistent one-shot arm flag. Four tasks filed: - the spec's decisions, review and decomposition - ratio pulling .emacs.d before its Emacs 31.1 upgrade - the stale lock-screen clock after a real sleep - the Signal tray icon waybar can't read. Electron 43 registers it under a well-known bus name and answers property reads only there. Downgrading is closed off by the database schema. The Lua-port task got its date and the sleep task carries the two Aug 23 hibernate failures. Wrap hygiene moved 42 aged resolved subtrees into the task archive. | |||||
