T

c3bd4a86ae Fix unbound XKBLAYOUT under set -u in the keyboard-layout migration (#6539 )

* Fix unbound XKBLAYOUT under set -u in the keyboard-layout migration

/etc/vconsole.conf only guarantees KEYMAP -- XKBLAYOUT is written by some
installers but not required, and vconsole.conf on a stock install may not
define it at all. omarchy-migrate runs every migration with set -euo
pipefail, so referencing the unset variable directly aborted the migration
run instead of just skipping a layout this migration doesn't care about.

Read it with a default-empty expansion first, then apply the existing
comma-strip separately -- keeps both failure modes (unset, and set with a
trailing keymap variant) handled explicitly instead of folding them into one
expansion that only covers one of the two.

* Survive a missing vconsole.conf in the keyboard-layout migration

Defaulting XKBLAYOUT fixed the unset variable but not the other way this
line takes the migration chain down. `.` fails when /etc/vconsole.conf is
not there at all, `&&` short-circuits, and the non-zero status leaves the
command substitution and kills the assignment under omarchy-migrate's
`bash -euo pipefail` -- the same abort, one branch over. The file is
optional enough that both other readers of it, omarchy_hooks.conf and
1781485962.sh, guard with `-f` first.

Run the echo unconditionally so the substitution reports its status
instead of the source's.

Read both paths from the environment, the way the zram migration already
does, and cover the layout cases plus both crashes with a test.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* Read the keyboard layout from vconsole.conf alone

Two ways the layout could come from somewhere other than the file, both
found reviewing the fix before it:

Sourcing does not clear an exported XKBLAYOUT, so on a machine that exports
one, a vconsole.conf that sets no layout -- or none at all -- left the
caller's environment deciding what the initramfs bundles. Unset it in the
subshell so only the file can answer.

Skipping the source when the file is missing, rather than letting the
substitution swallow the failure, also stops depending on errexit being
discarded inside command substitution. inherit_errexit takes that back and
the chain aborts again; nothing in Omarchy sets it today, but the other two
readers of vconsole.conf already check `-f` first and this now matches.

Test the layout list past its one Cyrillic entry, both orders of a
comma-separated pair, an exported XKBLAYOUT, and inherit_errexit. Each of
those catches a mutation that survived before.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* Drop the test scaffolding from the keyboard-layout migration

A migration runs once and then gets deleted, so a suite entry for this one
is upkeep with a short shelf life. The path overrides existed only so that
test could aim the migration at fixtures; with the test gone they are
indirection nothing exercises, so both go back to literal paths.

The fixes stay: check the file before sourcing it, unset XKBLAYOUT so an
exported one cannot answer for a file that sets none, and strip the keymap
variant before matching.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Toni Nowak <t.nowak@ai-flow.no>
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

2026-08-07 15:43:45 +02:00