Repository navigation
feat(brew): package Dakota's Homebrew files and prefix archive - #1126
Conversation
Dakota's bluefin/brew.bst copies system_files of ublue-os/brew at 550a03e3 into the image; bluefin/brew-tarball.bst stages /usr/share/homebrew.tar.zst from ghcr.io/ublue-os/brew at a pinned per-arch manifest digest (projectbluefin/dakota bf271777). Two recipes: - brew: the same ten files at the same paths and modes from the commit archive, with systemd scriptlets for the units its preset enables. - brew-tarball: Homebrew 7.0.11, the archive copied unchanged out of the image at sha256:130848582eed by a new generated_sources.py generator that checks the manifest and every layer against their digests. The archive exists nowhere else; the provenance skill records the pattern. Assisted-by: Claude Opus via GitHub Copilot Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Danathar
left a comment
There was a problem hiding this comment.
Stand-in review (the hive's reviewers are out of quota). Approving. brew: the source archive's sha512 equals Dakota's brew.bst ref, it's Apache-2.0, and its 10 files match %files and the element's paths. brew-tarball: homebrew.tar.zst was re-derived anonymously from ghcr.io/ublue-os/brew@sha256:130848582eed… (Dakota's x86_64 ref): the layer digest checks out, and the extracted file's sha512 equals the lock. It's Homebrew 7.0.11, BSD-2-Clause, and ExclusiveArch: x86_64 is right for the portable Ruby. For the Utah follow-up: Utah's BREW_IMAGE_SHA (9e0492…, amd64 03c8d1c8…) is a different snapshot from Dakota's, so switching changes which Homebrew ships, and it must drop COPY --from=brew so RPM-owned files aren't overwritten. Each Homebrew bump will now be a ~154 MB RPM.
|
@hanthor this PR is back in your court ( Review asks:
Reply here or push a new head to this branch (same sign-off); CI and the review swarm re-run automatically and the PR returns to the normal path. Queued in your relay fix queue (6 waiting): fixes are handed out before new work as soon as you have capacity. |
There was a problem hiding this comment.
HUMAN DECISION NEEDED — @hanthor, a maintainer must decide whether Utah may ship brew-tarball, a prebuilt Homebrew folder copied byte for byte from a third-party container image that cannot be rebuilt from source.
Summary: This pull request adds Homebrew, the command-line app installer, to the system, plus a ready-made copy of its install folder that is unpacked on first boot. The only change since the last review is a routine catch-up with the main branch. That catch-up brought in a fix to the login screen package that is already merged on main and is not part of this pull request. The open question from earlier reviews is unchanged: whether shipping a prebuilt bundle that nobody here can rebuild is acceptable.
What I need from you: A maintainer decides on the prebuilt bundle; the author adds Apache-2.0 to its license line.
Re-review of head 7263897 against ff58b28. The delta is merge 7263897 of main, which brings in #1194: packages/gdm/gdm.spec:64-70 (Requires: (pam_oo7 or gnome-keyring-pam)), one packages/gdm/changelog entry, and one row in .agents/skills/build-failure-triage/SKILL.md:212. compare/main...7263897 is 0 behind, and none of those three files is in this PR's own file list. So the PR's own content is unchanged since ff58b28, and the delta adds nothing to review.
Status of earlier notes:
- Provenance decision on brew-tarball (prebuilt prefix pulled from
ghcr.io/ublue-os/brew, unpacked by a root oneshot): still open. - brew-tarball License omits Apache-2.0: still present.
packages/brew-tarball/brew-tarball.spec:25still readsBSD-2-Clause AND MIT AND (Ruby OR BSD-2-Clause), but the archive ships Homebrew 7.0.11's vendoredsorbet-runtimegem, which is Apache-2.0. Suggested fix, unchanged from the earlier review:
-# Homebrew is BSD-2-Clause; the archive also carries its vendored Ruby gems
-# (mostly MIT) and portable Ruby (Ruby OR BSD-2-Clause).
-License: BSD-2-Clause AND MIT AND (Ruby OR BSD-2-Clause)
+# Homebrew is BSD-2-Clause; the archive also carries its vendored Ruby gems
+# (mostly MIT; sorbet-runtime is Apache-2.0) and portable Ruby (Ruby OR BSD-2-Clause).
+License: BSD-2-Clause AND MIT AND Apache-2.0 AND (Ruby OR BSD-2-Clause)No new findings from: correctness, security, intent-alignment, style, docs-currency.
Confidence: 3/5 (needs attention) — 5 perspectives requires a human decision
— hive: agent=reviewer backend=claude model=claude-opus-5-5 effort=medium claude=2.1.326
brew-tarball's archive carries sorbet-runtime (Apache-2.0) among Homebrew's vendored gems, so the License tag now lists it, as the review asked. brew ships a root oneshot (brew-setup.service) that unpacks brew-tarball, a prebuilt prefix that cannot be rebuilt from source. Both go on config/bump-review-only.txt so a future upstream tag or a moved image digest is a reviewed pull request, not a bump-gate self-merge. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: James Reilly <jreilly1821@gmail.com>
|
/hive decision approve — ship brew-tarball as prebuilt bytes, digest-pinned to the same image Bluefin and Dakota use. Decided for @castrojo's open question. Shipping the prebuilt Homebrew prefix adds no new trust: Bluefin's Containerfile and Utah's own I pushed the two small fixes the review raised: Decided on behalf of @hanthor with Claude Code |
|
Approval recorded by @hanthor for head |
…ources split Port the brew-tarball block merged from projectbluefin#1126 into generated_sources_archive, re-export its names from the facade, and register it in METADATA/GENERATORS. Hive-Fix-Attempt: 5/6 Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
What
Two new direct-upstream recipes from Dakota's
elements/bluefin/brew.bstandelements/bluefin/brew-tarball.bst(projectbluefin/dakota@bf271777). Fedora has neither. Utah currently copies both out ofghcr.io/ublue-os/brew(BREW_IMAGE); switching Utah to the RPMs is a separate Utah PR.brew(0^20261027git550a03e, noarch)ublue-os/brewat550a03e3, the element'sref(GitHub commit archive, SHA-512 locked, codeload fallback).cp -r system_files/usr/ /usr; cp -r system_files/etc/ /etcdoes:brew-setup.service,brew-{update,upgrade}.{service,timer},01-homebrew.preset,/etc/profile.d/brew.sh(0755, as in git) andbrew-bash-completion.sh,/etc/security/limits.d/30-brew-limits.conf,/usr/share/fish/vendor_conf.d/ublue-brew.fish./etcfiles are%config(noreplace);%systemd_post/preun/postunfor the three units the preset enables.Requires: tar zstd(brew-setup runstar --zstd).brew-tarball(7.0.11, x86_64)/system_files/usr/share/homebrew.tar.zstfromghcr:ublue-os/brewat the per-arch manifest digestsha256:130848582eed…(x86_64 ref). Universal Blue publishes the archive only inside that image, and its builder (Homebrew'sinstall.sh: git clone + portable Ruby download) is not reproducible.tools/generated_sources.pygenerator: pins that manifest digest under the Homebrew version the archive carries (git describeof its Homebrew checkout =7.0.11), gets an anonymous ghcr pull token, checks the manifest and each layer blob against their digests, and copies the member out of the topmost layer that carries it unchanged. Outputhomebrew-7.0.11.tar.zstis byte-identical to the file in the image (cmp), 158 MB; SHA-512 locked./usr/share/homebrew.tar.zst(the element's path) and Homebrew'sLICENSE.txt;Requires: brew(the element'sruntime-depends).ExclusiveArch: x86_64: the archive holds an x86_64 portable Ruby.tests/test_generated_sources_generate.py; the pattern is recorded indocs/skills/supply-chain-provenance.md.Closes #1029
Closes #1039
Verification
python3 tools/generated_sources.py brew-tarball …against live ghcr: output identical to the image's file.rpmbuild -baof both in the pinned buildroot (ghcr.io/projectbluefin/utah-buildroot:44-20261109-5aae8bdc5f40): both build; brew-tarball's%checklistsbin/brewin the archive. RPM sizes: brew 17 KB, brew-tarball 154 MB.just check(includesjust test): pass, 1572 tests.