Skip to content

ci: pin the noble-dependent runners, and attest the SBOM without the wrapper - #2249

Open
DeusData wants to merge 1 commit into
mainfrom
distill/2109-ci-pins-and-sbom
Open

DeusData wants to merge 1 commit into
mainfrom
distill/2109-ci-pins-and-sbom

Conversation

@DeusData

Copy link
Copy Markdown
Owner

Distilled from #2109 by @rarepops, carried with Co-authored-by. #2109 stays open — this takes two slices, not the PR.

1. Three runners pinned to ubuntu-24.04

_lint.yml lint and lint-mem, and _test.yml test-diag, each install a toolchain from an apt repo that names the release:

deb http://apt.llvm.org/noble/ llvm-toolchain-noble-20 main
deb http://apt.llvm.org/noble/ llvm-toolchain-noble-21 main

noble is 24.04. These jobs already depend on a specific release while asking for whatever ubuntu-latest means that morning. When GitHub advances ubuntu-latest, all three break at once on a day nobody chose, and it surfaces as an apt error far from its cause.

2. actions/attest-sbomactions/attest

attest-sbom is deprecated; what it does now is print a deprecation warning and forward to actions/attest with NODE_OPTIONS set, passing sbom-path through unchanged.

Verified rather than assumed, because this is on the release path:

  • 59d89421af93a897026c735860bf21b6eb4f7b26 is exactly tag v4.1.0 of actions/attest (checked via the git ref API)
  • sbom-path is a real input on that action at that SHA (checked against its action.yml)

The step keeps no continue-on-error, so a failed SBOM attestation still fails the release — and the comment now says so explicitly. An unattested SBOM is worse than an absent one, because the artifact ships looking complete.

Deliberately not taken from #2109

slice why not
actionlint job in _lint.yml _lint.yml is consumed by pr.yml, dry-run.yml and release.yml; pr.yml gates its whole test stage on lint succeeding and release.yml:105 gates build on needs.lint.result == 'success'. A job there is merge- and release-blocking with no branch-protection edit to make it visible — and this one curls a tarball from github.com releases with no cache or fallback. Worth having, not on that path, and not before it has been run against current main.
test-package-wrappers node matrix Reruns the Go and PyPI suites a second time per OS to vary a Node version only the npm leg reads.
test-frontend, .node-version, dependabot graph-ui The UI is being rebuilt from the ground up; gating on the current one buys coverage of a surface that is going away. .node-version would also desync the Windows VM, which pins v22.23.1 in provision-windows.sh:96.

Note

main currently carries one unrelated failing test (fix in #2248), so expect that red here until it lands. PyYAML is not installed locally so I could not lint the YAML offline — this PR's own CI consumes both edited reusable workflows, so a syntax error fails immediately.

…wrapper

Distilled from #2109 by Rares Popa. Two slices only; the rest of that PR is
either frontend work the UI rebuild makes redundant, or a gating change that
needs re-siting first.

1. Three jobs move from ubuntu-latest to ubuntu-24.04

   _lint.yml `lint` and `lint-mem`, and _test.yml `test-diag`, each install a
   toolchain from an apt.llvm.org repo whose name contains the release:

       deb http://apt.llvm.org/noble/ llvm-toolchain-noble-20 main
       deb http://apt.llvm.org/noble/ llvm-toolchain-noble-21 main

   noble IS 24.04. So these jobs already depend on a specific release while
   asking for whatever ubuntu-latest happens to mean that morning. The day
   GitHub advances ubuntu-latest, all three break at once, on a schedule
   nobody here chose, and the failure surfaces as an apt error far from its
   cause. Pinning the runner to the release the repo line already names costs
   nothing and removes the trapdoor.

2. actions/attest-sbom -> actions/attest

   attest-sbom is deprecated. Its remaining behaviour is to print
   "::warning::actions/attest-sbom has been deprecated" and forward to
   actions/attest with NODE_OPTIONS set, passing sbom-path through unchanged.
   Calling attest directly yields the same attestation minus the wrapper.

   Verified rather than assumed, because this sits on the release path:
   59d89421af93a897026c735860bf21b6eb4f7b26 is exactly tag v4.1.0 of
   actions/attest, and sbom-path is a real input on that action at that SHA.

   The step keeps no continue-on-error, so a failed SBOM attestation still
   fails the release. That is the intended behaviour and the comment now says
   so: an unattested SBOM is worse than an absent one, because the artifact
   ships looking complete.

Deliberately NOT taken from #2109, each for its own reason:

  - the actionlint job, as sited. _lint.yml is consumed by pr.yml, dry-run.yml
    and release.yml, pr.yml gates its whole test stage on `if lint succeeded`,
    and release.yml gates build on needs.lint.result == 'success'. A job there
    is merge-blocking and release-blocking with no branch-protection edit to
    make it visible, and this one curls a tarball from github.com releases with
    no cache and no fallback. actionlint is worth having; not on that path, and
    not before it has been run against current main.
  - the test-package-wrappers node matrix. It reruns the Go and PyPI suites a
    second time per OS to vary a Node version only the npm leg reads.
  - everything frontend: the test-frontend job, .node-version, and the
    dependabot graph-ui block. The UI is being rebuilt from the ground up, so
    gating on the current one buys coverage of a surface that is going away.

Co-authored-by: Rares Popa <2606875+rarepops@users.noreply.github.com>
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
@DeusData
DeusData force-pushed the distill/2109-ci-pins-and-sbom branch from 051702c to 238f9e6 Compare September 20, 2026 15:36
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant