diff options
Diffstat (limited to 'CLAUDE.md')
| -rw-r--r-- | CLAUDE.md | 11 |
1 files changed, 10 insertions, 1 deletions
@@ -125,6 +125,15 @@ Full palette reference: `assets/color-themes/dupre/dupre-palette.org` - Desktop file overrides go in `~/.dotfiles/hyprland/.local/share/applications/` - MPD is configured but mpv handles audio file associations - Firewall is ufw (configured in `archsetup`, default-deny incoming, explicit allow list). Tailscale traffic **does** traverse ufw on ratio — a probe from a tailnet IP is still blocked unless a rule covers the port. Don't assume tailnet-only services bypass the firewall; they need an explicit ufw rule like any other. -- This machine is **ratio**; **velox** is a laptop. Both run Hyprland (Wayland). archsetup still supports dwm/X11, but no current machine uses it. +- Never assume which machine this is — always run `uname -n` to find the hostname (the `hostname` binary is absent, so `uname -n` is the source of truth; `uname -r` is the kernel release, not the host). The fleet is **ratio** (workstation) and **velox** (laptop), both Hyprland (Wayland). archsetup still supports dwm/X11, but no current machine uses it. - Remote repository on cjennings.net - .ai/ is gitignored; living project context is in .ai/notes.org + +## Codified Insights + +- **VM tests run committed code, not your working tree.** `scripts/testing/run-test.sh` provisions the VM from `git bundle create <file> HEAD` (it simulates `git clone`), so an uncommitted edit to `archsetup` or the pytest suite silently runs the old code. Commit (even a throwaway WIP commit) before `make test FS_PROFILE=...`, or the change isn't exercised. (`gotcha` — 2026-06-25) +- **Iterate the pytest sweep against a kept VM, not a reinstall.** `make test-keep FS_PROFILE=...` leaves the VM up after the install and writes `testinfra_ssh_config` + `root_key` into `test-results/<timestamp>/`. Point pytest at that ssh-config to re-run only the Testinfra checks in ~30s instead of a ~70-minute full reinstall. Use it when iterating test assertions, not installer logic. (`pattern` — 2026-06-25) +- **VM UEFI NVRAM lives outside the qcow2 and must be per-profile.** OVMF boot entries live in the `OVMF_VARS` file, not the disk image, so reverting the `clean-install` snapshot does NOT restore them. The base ESPs have no removable `\EFI\BOOT\BOOTX64.EFI` fallback, so a base boots only via its NVRAM entry — lose or overwrite it and the VM dies in UEFI ("No bootable option") and SSH-times-out before archsetup runs. `init_vm_paths` now suffixes `OVMF_VARS` per `FS_PROFILE` (matching the disk image); never share one NVRAM file across btrfs/zfs. (`gotcha` — 2026-06-28) +- **sed/awk function extraction breaks on column-0 `}` inside heredocs.** The `tests/` harness and any `/^name() {/,/^}/` extraction stop at the first line beginning with `}` — but a JSON heredoc body (e.g. the docker `daemon.json` in `developer_workstation`) has a column-0 `}` that is NOT the function's close. Find the real closing brace before slicing, or the bounds are silently wrong. (`gotcha` — 2026-06-28) +- **AUR builds need ≥8 GiB VM RAM.** `makepkg` runs `-j$VM_CPUS`, and parallel `cc1plus` (~700 MB each on heavy C++ AUR packages) OOM-killed under the old 4 GiB `VM_RAM` default; the install still passed (yay retries) but the kills showed as attributed issues. Default is now 8192 MB. If you raise `VM_CPUS`, raise `VM_RAM` with it. (`threshold` — 2026-06-28) +- **Guard live upgrades with a PreTransaction hook, not a wrapper.** `hypr-live-update-guard` is a pacman `PreTransaction` hook (`AbortOnFail` + `NeedsTargets`) so it fires no matter how the upgrade launches (pacman, yay, topgrade) and aborts before any package is swapped — the safe point, since nothing is replaced yet. A shell wrapper around `pacman` would be bypassed by the other front-ends. (`pattern` — 2026-06-28) |