Skip to content

Centralize the VM/ISO end-to-end test machinery shared across utah, dakota-iso, iso, and bluefin #45

Closed

Description

@hanthor

Four repos now carry near-copies of the same VM/ISO test machinery, and this
week's Utah work fixed bugs in one copy that the others still have. The
duplication map, from working in all four:

The copies

Repo What it has State
projectbluefin/iso src/luks-unlock.py (origin of the unlock automation), justfile LUKS recipes, socat-driven QEMU monitor Original; requires socat; screen-hash heuristics
projectbluefin/dakota-iso Its own scripts/luks-install-qemu.sh, BLS patching, offline-store design, plain-verify-qemu serial assertions Fixed the BLS/offline problems independently
projectbluefin/utah (PR #41) iso/scripts/luks-e2e.sh, vendored+improved luks-unlock.py, boot-installed.sh Currently the most hardened copy
projectbluefin/bluefin build_files/base/20-tests.sh in-image assertions Different layer, same goal

Bugs fixed in one copy that the others likely share

  • False-positive pass at a UEFI shell. The screen-hash heuristic reads an
    EFI shell as a Plymouth prompt; Utah's copy now asserts on
    Reached target Graphical Interface over serial instead. The iso repo's
    copy still uses the heuristic as its primary signal.
  • Shutdown race. Powering down the live VM and force-quitting it seconds
    later boots a disk the guest is still flushing — intermittent
    Dependency failed for boot.mount in the installed system. Utah's copy now
    waits for the QEMU process to exit; dakota-iso and iso sleep and quit.
  • NVRAM reset. Booting the installed disk with fresh OVMF vars discards
    the installer's boot entry; with no \EFI\BOOT\BOOTX64.EFI fallback the
    firmware drops to its shell. Utah's copy carries the live VM's vars over.
  • GTK-over-SSH launches. The session bus denies registration
    (ApplicationRegisterFailed), and the attempt leaves a half-made user
    flatpak repo that breaks every later flatpak call for that account.
    Utah's copy uses a session autostart entry instead.
  • socat dependency. Utah's luks-unlock.py drives the QEMU monitor over
    a plain UNIX socket; the iso copy requires socat, which is not packaged
    everywhere the tests need to run.

What would centralize well

  1. The QEMU E2E harness — boot ISO → assert live desktop → install →
    boot → unlock → login → screenshots. Parameterize: payload ref, installer
    app id, recipe shape (composefs vs ostree/bootc), filesystem, expected
    session, terminal app.
  2. The verification-record generator — screenshots plus a regenerated
    docs/verification/README.md, so every image PR carries its proof.
  3. boot-installed.sh — boot the tested artifact with VNC + noVNC + SSH
    for hands-on use, through a qcow2 overlay so the artifact stays pristine.
  4. Longer term: the ISO assembly scripts themselves — Utah's and Dakota's
    diverged from the same design and re-fixed each other's bugs.

Suggested sequencing

Extract after #41 lands, using Utah's copy as the seed —
it is the one with all five fixes. Consumers: utah, dakota-iso, iso; bluefin
can adopt the in-image assertion layer later. A projectbluefin/testing (or
shared-workflows) repo consumed as a submodule, a versioned tarball, or a
reusable workflow — maintainer's choice.

🤖 Generated with Claude Code

https://claude.ai/code/session_015VF8ttMH6WGQHeQ9hdMSY9

Activity

  1. mrbobbytables commented on Sep 22, 2026

    @mrbobbytables
    Contributor

    Utah's copy of the VM/ISO test machinery is already hardened and landed in projectbluefin/utah via PR #41 (feat: Build the Utah ISO, install it to an encrypted disk, and prove the desktop starts), which resolved all five bugs noted above for this repository (iso/scripts/luks-e2e.sh, iso/scripts/luks-unlock.py, boot-installed.sh).

    Centralizing this test machinery across utah, dakota-iso, iso, and bluefin is blocked on an unanswered maintainer decision regarding the target architecture:

    1. Destination repository (e.g. creating projectbluefin/testing, projectbluefin/shared-workflows, or adopting projectbluefin/actions).
    2. Distribution and consumption mechanism (Git submodule, reusable GitHub Actions workflow, versioned tarball, or container image).

    Until maintainers decide on the repository location and consumption model, there is no actionable change shippable directly within projectbluefin/utah.

    — hive: backend=copilot model=gemini-3.8-flash

    🐝 Hive Agent: contributor | SHA: unknown

  2. hivecommons-hive commented on Oct 10, 2026

    @hivecommons-hive
    Contributor

    Fixed by #41 (merged 281735c). Closing. Reply /reopen if the problem persists.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions