# 8. Native Void Linux support and XBPS delivery Status: Accepted scope and hosting direction; implementation and acceptance proof pending. Fenris will support Void Linux natively with runit and full application feature parity, while retaining its existing Debian/RPM and systemd support. This extends the platform boundary in [ADR 0003](0003-service-lifecycle-and-sanctioned-toggle.md) and the delivery scope in [ADR 0007](0007-package-delivery-amends-0004.md): requiring Void users to replace their init system would not meet the native-support goal. Delivery will include a Fenris-maintained, signed XBPS repository that users configure once for subsequent installation and updates through XBPS, plus versioned release artifacts and notes on Gitea. All downloads must be served directly by Gitea itself; a separate static HTTP repository, even alongside Gitea, does not satisfy this requirement. Use a dedicated public Gitea repository, provisionally `xavierk/Fenris-xbps`, with a permanent `stable` branch. Its raw-file URL serves the XBPS index, versioned packages, and package signatures as ordinary Git blobs without LFS. Keeping binaries in a separate repository avoids increasing application source-clone size. This accepts growth in distribution-repository Git history in exchange for publishing index and artifacts together through one branch update, without the generic registry's delete-and-upload index replacement gap. The proposed repository has not yet been created. Serialize publication, commit the signed index and its new artifacts together, and retain older versioned artifacts in the current tree so clients with cached older indexes can still download them. Native XBPS installation and update tests against the actual endpoint are required before release validation. Verify binary delivery limits and index freshness: the inspected Gitea raw endpoint advertised six-hour HTTP caching, so immediate update visibility has not been established. Immediate availability is required: after successful XBPS publication, an explicit repository refresh against the permanent URL must discover the newly published version without a cache-expiry wait or a URL change. This does not promise automatic installation on client machines. Acceptance must exercise a client that fetched the previous index before publication and verify that refresh retrieves the new index and its signed package afterward. Resolve and document actual client and intermediary cache behavior; if the selected Gitea route cannot meet this requirement, hold XBPS publication and revisit its delivery mechanics rather than silently accepting delayed availability. Hosting evidence: [Gitea generic registry](https://docs.gitea.com/usage/packages/generic), [raw download routing](https://github.com/go-gitea/gitea/blob/main/routers/web/web.go), [download handler](https://github.com/go-gitea/gitea/blob/main/routers/web/repo/download.go), and [Void repository signing](https://docs.voidlinux.org/xbps/repositories/signing.html). Source inspection and an existing raw-file GET establish feasibility, not end-to-end XBPS validation. The first supported Void target is x86_64 with glibc, matching the inspected development machine. Release validation must cover installation, real collection, pause/resume, reboot persistence, upgrade, and removal on that machine, preserving existing observation history. Debian/RPM compatibility remains part of the acceptance scope. Additional architectures and musl support are outside this first release. Package formats have independent publication gates: publish each validated format, and hold only formats that have not passed release validation. A failure or pending validation in XBPS must not prevent a validated Debian or RPM package from shipping, and vice versa. Release notes must identify available formats and those still withheld; publication must not imply validation of a missing format. After successful native acceptance testing, leave the released XBPS package installed on this machine and monitoring the selected NVMe drive. Preserve the observation history collected during testing. Coordinate the reboot test with the user so it can occur at a suitable interruption point.