Skip to content

provenance: define canonical PURL encoding for digest qualifier in materials/subjects #6525

@crazy-max

Description

@crazy-max

In #5641, bumping github.com/package-url/packageurl-go from v0.1.1 to v0.1.3 changes how digest qualifiers are serialized in PURLs used by provenance/attestation output:

  • Previous: pkg:docker/busybox?digest=sha256:...&platform=linux%2Famd64
  • New: pkg:docker/busybox?digest=sha256%3A...&platform=linux%2Famd64

BuildKit emits these PURLs in provenance-facing data such as resolved dependencies via util/purl.RefToPURL:

if s.Local {
uri, err = purl.RefToPURL(packageurl.TypeOCI, s.Ref, s.Platform)
} else {
uri, err = purl.RefToPURL(packageurl.TypeDocker, s.Ref, s.Platform)
}

And in-toto attestation default subjects:

pl, err := purl.RefToPURL(packageurl.TypeDocker, name, &p.Platform)

Even if %3A and : are semantically equivalent after decoding, this is a canonical string change in provenance payloads and can impact policy engines or custom checks that match URI strings or downstream tooling expecting existing URI formatting.

Upstream discussion that motivated this: package-url/packageurl-go#54. Also current PURL standard text says : shall not be percent-encoded in component content: https://github.com/package-url/purl-spec/blob/main/docs/standard/specification.md

So we currently have a practical interoperability/canonicalization choice to make in BuildKit.

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions