image-builder: honor repart.d partition definitions from the bootc container #99

Activity

1. cgwalters-bot

   added

   P2

   target:osbuild

   on Sep 26, 2026
2. cgwalters-bot

   added this to Workstreamon Sep 30, 2026
3. cgwalters-bot

   added a parent issue on Oct 3, 2026

   • Epic: image-builder/osbuild composefs (waits on operator text) #15
4. github-project-automation

   added this to Composefs Stableon Oct 3, 2026
5. cgwalters-forge

   locked as resolved and limited conversation to collaborators on Oct 18, 2026
6. cgwalters-bot commented on Oct 22, 2026

   

   cgwalters-bot

   on Oct 22, 2026

   Last edited by cgwalters-bot

   CollaboratorAuthor

   More actions

   ❌ dispatch failed (agent, check, apply: failure/skipped/skipped; the agent's exit state: 1).
7. cgwalters-bot commented on Oct 22, 2026

   

   cgwalters-bot

   on Oct 22, 2026

   CollaboratorAuthor

   More actions

   Conclusion: still wanted, not started, and it needs the operator's decision on the approach before an implement run. Nothing in image-builder reads repart.d yet, and the design is being argued upstream in osbuild/image-builder#2738 (opened by cgwalters on 2 Oct, six days after this item), where the maintainer proposes a different mechanism than this item's wording implies. Triage as of 21 Oct 2026.

   What exists today

   • The PR this came from, cgwalters-forge/image-builder#1, went upstream as osbuild/image-builder#2719 and merged 28 Sep. It kept the default 501 MiB ESP on purpose and deferred the size to the container, i.e. to this item.
   • bootc-dev/bootc#2314 merged 1 Oct. bootc install to-disk now runs systemd-repart when the container has repart.d/*.conf under /etc, /run, /usr/local/lib or /usr/lib, injects a root definition if none is shipped, and falls back to sfdisk otherwise (bootc-installation.7.md, "systemd-repart"; code in crates/lib/src/install/baseline.rs). So the container side is in place.
   • image-builder main (f43ddf4, 21 Oct): the only mention of repart is an alignment comment in pkg/disk/partition_table.go. The bootc base table comes from the container's /usr/lib/bootc-image-builder/disk.yaml (pkg/bib/osinfo/osinfo.go) or the built-in YAML, in basePartitionTable and genPartitionTable of pkg/distro/defs/bootc_imagetype.go. pkg/osbuild has no wrapper for osbuild's org.osbuild.systemd-repart stage (osbuild#2189, #2460), so partitioning with repart from a manifest needs that first.
   • Upstream #2738 (cgwalters and supakeen, 2 to 3 Oct). Agreed so far: it must be opt-in, because Fedora ships repart.d for root growth so mere presence cannot be the trigger; disk.yaml wins over repart.d, which wins over the built-in layout; start with the root filesystem type and the ESP. Not agreed: the mechanism. supakeen proposes letting systemd-repart create the disk image and deploying with to-filesystem onto it (mounted via systemd-dissect), and notes fstab or mount units are then a problem on BIOS, where gpt-auto does not mount the ESP. cgwalters' last word there: "we still need to define coherent semantics". The parent epic #15 also says it waits on operator text.

   Decision for cgwalters

   Which first PR, if any, to send upstream while #2738 is open. Option A, narrow mapping: read the container's repart.d, take the Type=esp size and the Type=root Format= and size, and apply them to image-builder's own partition table behind an opt-in flag, only when the container has no disk.yaml. It fits one PR and is testable with a unit test on a fake root tree plus a boot test of a UKI image shipping a 2G ESP definition, but it is a reimplementation of a slice of repart.d semantics that the maintainer may see as a half-measure. Option B, the mechanism supakeen sketched: a Go wrapper for the osbuild systemd-repart stage, then a bootc pipeline that partitions with repart and deploys with to-filesystem, which overlaps the bootc install mount work in cgwalters-forge/image-builder#7. That is two or three PRs and should wait for the semantics to be settled on #2738. Option C: hold this item, pointing at #2738, until the design converges upstream. My recommendation is to settle the mechanism on #2738 first; if you want movement now, A is the only piece that fits one PR, and a job plan for it can follow once you pick.

   Generated by dispatch · ◷