The upload steps named the web host by its internal address. It now comes
from the TARBALLZ_TARGET repository variable, so the public workflow carries
no infrastructure details.
Claude-Session: https://claude.ai/code/session_01GGe7kUhooMFuzY6GaDb2w9
The Maintainer line uses the usual AUR form (asger at weircon dot dk)
instead of a raw address that spam crawlers harvest. The AUR workflow
can now also be started by hand with a version and pkgrel, so a fix to
the PKGBUILD alone can be published without a new upstream release.
Claude-Session: https://claude.ai/code/session_012LCTmU2ewo8tnM1dByYAoG
`init` planned every file on every run and refused if any of them
existed, so attaching a new monitor could not be configured without
--force — which reallocated global workspace numbers by i3's current
output order and renumbered the monitors already in use.
- allocate_global_starts() keeps the block each known output already
holds and gives new outputs the lowest free block; blocks of inactive
outputs stay reserved.
- read_allocations()/parse_allocation() read those blocks back out of
the generated files (inverse of render_workspaces_for_output).
- --sync writes only the missing files, --prune drops configs for
outputs that are gone and frees their numbers.
- Files without the auto-generated header are never written or
deleted, --force included.
Co-Authored-By: Claude <noreply@anthropic.com>
Asger Weirsøe