On a btrfs root the swapfile was written with dd from /dev/zero, so a
swapfile the size of the RAM cost tens of GiB of writes and stalled
every other writer on the filesystem. A caller that bounds the run, such
as an async Ansible task, killed dd halfway, and the next run found the
wrong size and started over. btrfs accepts a preallocated NOCOW
swapfile and the swap directory already carries +C, so every filesystem
now allocates with fallocate.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Turns the loose main.py into the 'swap-forge' distribution with a src layout, shipping both the swap-forge and the older swafo command.
A failing swap command no longer produces a traceback: CalledProcessError is caught and reported as "'mkswap' failed with exit code 1" at exit 1, a missing binary as exit 127. Required binaries are declared per filesystem path so an ext4 host is never asked for btrfs tooling.
The fstab entry is still written last, after swapon, so a failed activation leaves /etc/fstab untouched. The e2e suite proves that byte for byte across a create, a same-size rerun and a resize cycle, with real fallocate, chmod and file sizes.
Adds unit, integration and container-based e2e tests, ruff and markdown/mermaid linting, CodeQL and Dependabot, and pins the core metadata to 2.4 so twine accepts the artifacts.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
