Skip to content

[quality] test: unit coverage for files/installer/repart.d target-disk layout (tests/unit/test_repart_layout.py) - #41

Merged
castrojo merged 1 commit into
mainfrom
quality/test-installer-repart-layout
Sep 5, 2026
Merged

castrojo merged 1 commit into
mainfrom
quality/test-installer-repart-layout

Conversation

@hivecommons-hive

Copy link
Copy Markdown
Contributor

Test Improvement

Adds tests/unit/test_repart_layout.py — pure static coverage for the systemd-repart recipe the live installer follows when partitioning the target disk.

Claimed ground (for overlap checks by later work):

  • new file: tests/unit/test_repart_layout.py
  • files under test (read-only): files/installer/repart.d/10-esp.conf, 20-root-a.conf, 30-var.conf
  • cross-file reads: elements/**/*.bst (grep for the Label= stamped on the installer media data partition) and files/os/sysupdate.d/50-root.transfer ([Target] MatchPattern)
  • no production file is modified, and no Justfile recipe or workflow is added

Why

Nothing in CI parses files/installer/repart.d/ today. A bad Type=, an inverted size bound, or a CopyBlocks= source that no longer matches the label stamped on the installer media data partition surfaces only as a failed or silently mis-partitioned install — after a 180-minute image build.

Invariants covered (18 tests, 1 strict xfail)

  • each config parses and holds exactly one [Partition] section; filenames are NN-name.conf with unique ordering prefixes
  • the layout is exactly one esp, one root, one var
  • SizeMinBytes always set, never greater than SizeMaxBytes
  • GPT labels unique, so sysupdate Path=auto stays unambiguous
  • ESP is vfat, ≥100M, and capped
  • root CopyBlocks= addresses the media by /dev/disk/by-partlabel/…, and some element under elements/ actually stamps that label (bluefin-installer-data)
  • root grows its filesystem and is capped; var is xfs, FactoryReset=yes, growable and uncapped as the tail partition
  • the installed root label appears in the [Target] MatchPattern of 50-root.transfer

The strict xfail

test_every_sysupdate_root_target_is_provisioned_by_the_installer is xfail(strict=True). It records the A/B gap already documented in docs/MVP_1_0_READINESS.md and docs/skills/gap-analysis-distros.md: 50-root.transfer names root-a and root-b, but the installer provisions only root-a, so systemd-sysupdate has no inactive slot to stage into. Strict means the test turns red the moment root-b lands, prompting whoever adds it to enable the guard. This PR does not attempt the fix — that is production change, out of scope for quality.

Verification

$ python3 -m pytest tests/unit/test_repart_layout.py -q
18 passed, 1 xfailed in 0.10s

Mutation-checked: rewriting CopyBlocks= to an unstamped label, and setting root SizeMinBytes=32G above SizeMaxBytes=16G, each fail the expected test.

Runner

No Justfile change is included on purpose — open PR #30 already adds test-unit: python3 -m pytest tests/unit -q, which picks this module up automatically. Deliberately disjoint from the other open server PRs: #38 (files/bin/system-container), #30 (.github/scripts/docs-checks.py), #34 (project.conf/Justfile release-version), #40 (CPU-arch identity), #36/#37/#29 (sysupdate keyring, k3s kubeconfig mode, sshd), #22/#25/#26/#27 (PXE, bluefin-server-installer.bst).


Filed by quality agent (hold-gated mode). Human review required — do not merge without review, and do not remove the hold label.

— hive: agent=quality backend=copilot model=claude-opus-5

…k layout

Adds tests/unit/test_repart_layout.py, a pure static test module over the
systemd-repart recipe the live installer follows when partitioning the
target disk. Nothing in CI parsed these files before, so a bad Type=, an
inverted size bound, or a CopyBlocks= source that no longer matches the
label stamped on the installer media data partition could only surface as
a failed or mis-partitioned install.

Covered invariants:
  - every files/installer/repart.d/*.conf parses and holds exactly one
    [Partition] section, with an NN-name.conf ordering prefix that is unique
  - the layout is exactly one esp, one root and one var partition
  - SizeMinBytes is always set and never exceeds SizeMaxBytes
  - GPT labels are unique, so sysupdate Path=auto stays unambiguous
  - ESP is vfat, at least 100M, and capped
  - root CopyBlocks= addresses the installer media by partition label, and
    some element under elements/ actually stamps that label
  - root grows its filesystem and is capped; var is growable, xfs,
    FactoryReset=yes and uncapped as the tail partition
  - the installed root label is among the [Target] MatchPattern labels in
    files/os/sysupdate.d/50-root.transfer

One strict xfail records the A/B slot gap already documented in
docs/MVP_1_0_READINESS.md: 50-root.transfer names root-a and root-b but the
installer provisions only root-a. Strict means the test flips to a failure
the moment root-b is added, prompting the guard to be enabled.

No production files are touched.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Signed-off-by: sec-check[bot] <sec-check[bot]@users.noreply.github.com>
@hivecommons-hive hivecommons-hive Bot added the hold Indicates that a PR should not merge because someone has issued a /hold command. label Aug 29, 2026

@hanthor hanthor left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

New static tests genuinely check repart.d/*.conf invariants against real cross-file sources (elements/*.bst, sysupdate transfer), with an honest xfail(strict=True) for a documented known gap. CI green.


Generated by Claude Code

@castrojo castrojo left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Approved by @castrojo for Hive auto-merge on green CI.

@castrojo
castrojo merged commit 7398026 into main Sep 5, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

hold Indicates that a PR should not merge because someone has issued a /hold command.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants