One thing I’m wondering about is the SBOM coverage. The docs[1] say some dependencies inside the bundled Iceberg JAR may not be listed. The SBOM can still help identify the components it does include, but consumers could mistake an unlisted component for one that is absent.
1. https://github.com/snazy/polaris/blob/license-check-gradle-cc-ip/gradle/sbom/README.md Yufei On Mon, Nov 9, 2026 at 8:29 AM Robert Stupp <[email protected]> wrote: > Hi JB, > > Thanks! Agreed on the timing. > > I'll target the Licensee and SBOM changes for main after the 1.9.x branch > is cut, and split the work into three PRs, meaning one PR for each commit > in the draft-PR. > Adding a CHANGELOG entry for removing the license-report classifier and > pointing users to the SBOMs makes sense. > > The VEX work in #6001 could still go into 1.9.0 independently, with its > release coverage checked against the artifacts we ship. > > For the product identities, I'd reuse the exact versioned PURLs used as > SBOM subjects. > For the binary distributions, that would be: > > - pkg:generic/[email protected]?variant=binary-zip > - pkg:generic/[email protected]?variant=binary-tar > > These identify the artifacts independently of whether we already publish an > SBOM for them, so #6001 wouldn't need to wait for the SBOM work. > > In OpenVEX, I'd use those as the products and keep dependency PURLs as > optional subcomponents where known. > In CycloneDX, I'd declare the corresponding artifact components and > reference their bom-refs from affects. > > A release-level grouping is fine for organization, but I'd keep the > assessment targets artifact-specific. > One assessment can cover both ZIP and TAR without duplicating the analysis, > while Spark and container images would be included only after assessing > those distributions separately. > > That gives the VEX and SBOM documents shared identifiers without > introducing another mapping layer. > The document revision and last-update tracking you mentioned would complete > the other part of that alignment. > > Thanks, > Robert > > On Mon, Nov 9, 2026 at 5:16 PM Jean-Baptiste Onofré <[email protected]> > wrote: > > > Hi Robert, > > > > Thanks for working on this! > > > > +1 about that, you read my mind :) > > > > I have a few comments: > > - CycloneDX JSON sidecars for the binary ZIP/TGZ and the Spark bundle > JARs > > are a great first step. Binding each SBOM to the artifact SHA-256 and > > signing it like the other release artifacts is what we need. I'm fine > with > > doing the container images as follow-up, but I would not wait too long ;) > > - OK for Licensee. Configuration cache support is a good reason. I also > > like that we keep our own checks between the packaged dependencies and > the > > LICENSE/NOTICE files, as that's what we verify during release votes. > > - We publish the license report on Maven Central with the > "license-report" > > classifier for polaris-server and polaris-admin. It's not on > > dist.apache.org, > > the release guide and the release workflows don't use it, and I don't use > > it as release manager :) So I'm not aware of any consumer (as it's on > Maven > > Central, I'm not 100% sure ;)). I would mention the removal in the > > CHANGELOG (just in case) and point to the SBOMs instead. I guess the > SBOMs > > include the license for each component. > > - Great idea to reference the VEX feed (it was the intent in the other > > thread). Thanks to your work, I propose to update #6001 (explicit Polaris > > release identities in both CycloneDX and OpenVEX, plus document version > and > > last update tracking). To align, should the VEX products reuse the purls > > your PR uses as SBOM subjects? Or do you prefer a release level > identifier > > in the VEX, with the distributions as subcomponents? I'm fine either > way, I > > would just like to update #6001 accordingly. > > - We are pretty close to 1.9.0 (as said this morning, I'm waiting #5927 > > ;)). As the PR changes the release workflows and the license checks run > > during the release, I would prefer to merge it after 1.9.x release > branch. > > That way, 1.9.0 is built as usual, and 1.10.0 with SBOMs. We can verify > > with a dry run. > > - The PR is pretty big. Can we split in three PR: Licensee migration, > SBOM > > publication and VEX link (depending on #6001)? > > > > Thanks again! > > > > Regards > > JB > > > > On Mon, Nov 9, 2026 at 2:50 PM Robert Stupp <[email protected]> wrote: > > > > > Hi all, > > > > > > Following the recent discussion about dependency findings and VEX [1], > > I'd > > > like to propose publishing SBOMs for Polaris release artifacts and > > updating > > > our dependency-license checks. > > > I've opened an illustrative draft PR [6021] to make the proposal > > concrete. > > > The license-check migration, SBOM publication and VEX link are separate > > > commits. > > > > > > The initial scope would be CycloneDX JSON sidecars for the binary > ZIP/TAR > > > distributions, covering server and admin, and for each Spark bundle > JAR. > > > The documents identify the actual release artifacts by SHA-256 and > would > > be > > > signed and published alongside them. > > > Spark SBOMs would also be attached to the existing Maven publications. > > > For server/admin, the implementation reuses Quarkus-generated > > inventories; > > > for Spark, it uses the official CycloneDX Gradle plugin and the > > dependency > > > inputs actually selected by Shadow. > > > Container-image SBOMs would be a separate follow-up. > > > > > > The related build change replaces the JK1 license-report plugin with > > > Licensee [3]. > > > The motivation is configuration-cache support and simpler shared policy > > > configuration. > > > Licensee handles Maven license metadata and explicit > > license/URL/dependency > > > allowances, while Polaris keeps the checks connecting packaged > > dependencies > > > to attribution markers and delivered LICENSE/NOTICE files. > > > The attribution check connects the policy results to the components > > > selected for packaging. > > > > > > The draft removes the old license-report ZIP; the packaged LICENSE and > > > NOTICE files remain. > > > Is anyone consuming that report ZIP, or relying on its particular > format? > > > > > > The SBOMs would also link to the mutable VEX feed proposed in PR > [6001], > > > allowing assessments to evolve without regenerating signed release > > > inventories. > > > That link provides discovery; the product and release scope still needs > > to > > > line up with the VEX work. > > > This proposal doesn't change the assessments themselves. > > > > > > Does this publication scope and the choice of Licensee make sense? > > > I'd appreciate feedback on the release artifacts and formats, and on > any > > > consumers we should account for before replacing the existing reports. > > > > > > Thanks, > > > Robert > > > > > > [1] > > https://lists.apache.org/thread.html/jkv11728rj1ytbql7ykjj9o9v8m4bdts > > > [6021] https://github.com/apache/polaris/pull/6021 > > > [3] https://github.com/cashapp/licensee > > > [6001] https://github.com/apache/polaris/pull/6001 > > > > > >
