Skip to content

v0.87.0

Choose a tag to compare

@github-actions github-actions released this 13 Aug 21:50
· 579 commits to main since this release
4ea29bc

Added

  • GitLab's Dependency List and License Compliance tab are populated. Both read a CycloneDX
    SBOM rather than a report of their own, and Draugr's SBOM already carries SPDX identifiers in
    each component's licenses — the only form GitLab reads. The template collects
    draugr-out/sbom-*.cdx.json, so enabling config.sbom in the descriptor is all it takes, and
    one artifact fills both: every package with its version and its licence. Ultimate tier.

  • DRAUGR_GATE_DEFAULT_BRANCH in the GitLab template, because gating the default branch and
    populating GitLab's merge-request widgets turn out to be mutually exclusive. GitLab baselines
    those widgets against the default branch's last successful pipeline, so a project whose default
    branch legitimately fails the gate never establishes one — the reports upload and there is
    nothing to compare them against. On by default, since a gate that does not fail is not a gate;
    set it to "false" to keep the default branch green and reporting, and leave gating to the
    merge-request path. The guide gives the trade-off as a table.

Fixed

  • The GitLab template's merge-request diff compares the right two trees. It named the base
    descriptor while standing in the head checkout, and a component's url: . resolves against the
    working directory — so both scans read the head, every diff reported no change, and the
    differential gate could not trip. It failed green, which is the only way a gate fails that nobody
    notices. The base is scanned from inside its own worktree now, as the GitHub Action already did.

  • The GitLab template's merge-request diff has something to diff. --report replaces the
    default json,sarif rather than adding to it, so a job asking only for GitLab's own formats
    wrote no results.sarif — and the diff then failed on a missing file, which points at the diff
    rather than at the report list that caused it. The template names sarif explicitly, and a test
    now checks that whatever the template diffs is a format it asked the scan to write.