Repository navigation
Conversation
Bare-metal testing on a ThinkPad X230 found gaps of the same shape: a package installed with nothing enabling it, or a package the Hummingbird base never had and no manifest asked for. Reported with journal evidence and a Bluefin side-by-side on the same machine in #97, #98, #99, #100 and #104. Every name below was checked against the two repositories the image actually enables -- the digest-pinned factory metadata layer from Containerfile, and public-hummingbird-x86_64-rpms -- by reading their repodata directly, because this is exactly the change where a plausible-looking package name is the failure mode. packages/utah.toml gains a [hardware] section: linux-firmware, iwlwifi-dvm-firmware, iwlwifi-mvm-firmware, iwlwifi-mld-firmware, iwlegacy-firmware linux-firmware alone would NOT have fixed #97, and that is the trap worth recording. It is a split package here: it Requires only linux-firmware-whence, and it Recommends thirteen device packages -- amd-gpu, atheros, brcmfmac, cirrus, intel-gpu, mt7xxx, nvidia-gpu, nxpwireless, qcom-wwan, realtek, tiwilink and friends -- which weak dependencies bring in for free. Intel wireless is in neither list. Measured against the pinned repository's filelists, iwlwifi-6000g2a-6.ucode -- the exact blob the report names -- iwlwifi 0000:03:00.0: Direct firmware load for iwlwifi-6000g2a-6.ucode failed with error -2 iwlwifi 0000:03:00.0: no suitable firmware found ships in iwlwifi-dvm-firmware. Installing linux-firmware and stopping there would have closed the issue while leaving the radio just as dead. The three iwlwifi generations plus iwlegacy cover Intel wireless from the 3945 era through current parts; everything else arrives through the recommends. microcode_ctl was in the first draft of that section and is now deliberately absent: it exists in neither enabled repository, so naming it would have failed the install transaction outright. CPU errata updates remain a real gap on this base and need a factory recipe first. nautilus joins [gnome], which turns out to need no new factory work at all: it is already in the pinned factory repository at 51~beta-1.hum1.bfin. #100 reported the accessories shipping orphaned -- nautilus-extensions, nautilus-python and nautilus-gsconnect installed with no /usr/bin/nautilus for them to extend -- and the cause was simply that nothing asked for it. Bluefin does not list it either, because Silverblue carries it in the base. gnome-disk-utility and ptyxis are in neither repository and stay open. avahi joins [services]. Hummingbird ships avahi-libs and avahi-glib but not the daemon -- confirmed at 0.9~rc2-8.1.hum1 in their repository -- so geoclue retried forever against nobody (#104): geoclue[2327]: Failed to connect to avahi service: Daemon not running taking mDNS resolution and WiFi-based location with it. [hardware] is a new section rather than more names under [services], because firmware is not a service and this file is careful about that distinction. Sections are enumerated rather than globbed, so install-packages.py has to opt in or the section is silently ignored -- and verify-rpm-contract.py has to as well, or firmware could go missing from a built image with the contract check none the wiser. Both are wired up. configure-services.sh enables bluetooth.service (#98: bluez 5.87 and bluetoothd installed, bluetoothctl shows no adapter at all), input-remapper.service (#99: the GUI cannot reach a root daemon and falls back to a password prompt) and avahi-daemon.service. All three are also added to 85-utah-desktop.preset, which is the part that makes them stick. bootc applies presets on first boot, so a build-time systemctl enable with no preset line behind it is undone -- the trap configure-services.sh's sshd branch already documents. #98 confirmed it from the other side: grepping the deployed image found no preset enabling bluetooth.service anywhere. tests/test_package_resolution.py grows that invariant as a guard, since it is the kind of thing the next unit added will get wrong. Three units predating this test are enabled with no preset entry -- brew-setup, flatpak-nuke-fedora, flatpak-preinstall. They are listed as known exceptions rather than asserted to be correct: they come from ublue packages that may ship their own vendor presets, and confirming that needs the built image, not the source tree. A second test fails if a name in that list stops being enabled, so the exception cannot go stale and quietly widen the hole. Verified by mutation rather than assumed. Dropping the [hardware] line from the resolver, removing bluetooth from the preset, dropping *hardware from the verifier's expected set, and deleting linux-firmware or avahi from the manifest each fail the suite and name what broke. Writing the guard also caught a bug in its own helper: an anchored regex missed the indented sshd enablement inside its conditional, which made the exception list look stale when it was not. Not fixed here: #101 timer The bootc-fetch-apply-updates mask loses to a cross-vendor /etc merge, which carries Bluefin's timers.target.wants symlink across. An image-time mask structurally cannot win against a switcher's /etc; that needs a design decision. #102 BIOS Documentation, and the base ships no BIOS boot payload. #103 getty Needs a bootc karg change as well as a mask, which is a riskier edit than the rest of this and deserves its own review. 33 tests pass, 57 subtests pass. The contract check reports 58 Bluefin, 11 GNOME, 5 firmware and 6 service packages. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6BYLUD6KBzF8rBFXKjP7M
|
CI is red on
It does cost this PR something, though: the parity step failing skips "Check package availability in Utah's repositories", which is precisely the check that would have validated the new package names. I verified them by reading the repodata directly instead (details in the PR body). Why the obvious fix breaks the buildThe drift is four hunks against upstream A straight
So syncing alone hands three unresolvable names to the install transaction. Proposed patchTwo parts, which have to land together:
That file's own rule is that every I've deliberately not done this inside this PR: it changes the parity contract for the whole repository and unblocks #94 as well, so it deserves its own review rather than riding along with a firmware fix. Happy to open it as a separate PR with the three issues filed — say the word. Generated by Claude Code |
packages/bluefin.toml is meant to be a byte-for-byte copy of upstream Bluefin's build_files/packages/base.toml. utah.toml says so in as many words -- "it is the parity contract, and any drift there is a parity bug" -- and just check-parity exists to make that loud. It has drifted, so that check is red on main today and every open pull request inherits the failure, #94 and #105 among them. The drift is four hunks: firefox moved into [fedora], and libgda, libgda-sqlite and ppp missing. Syncing the file alone would have been worse than the failure it fixes. All three missing names were measured against both repositories the image actually enables -- the digest-pinned factory metadata layer from Containerfile, and public-hummingbird-x86_64-rpms: libgda absent from both libgda-sqlite absent from both ppp absent from both, and nothing Provides it So a bare sync hands three unresolvable names to the install transaction and turns a failing lint into a failing image build. The sync and the [unavailable] entries have to land together, which is why they are one commit. Each entry carries a tracking issue, as that list requires: libgda, libgda-sqlite (#106) Upstream added both for the maintained Copyous extension (bluefin#1210). Hummingbird carries no package whose name contains gda at all. One factory recipe covers both, because libgda-sqlite is a subpackage of the libgda SRPM rather than its own source -- Fedora's libgda.spec declares %package sqlite alongside ui, tools, sqlcipher, mysql and postgres. ppp (#107) Worth more than dial-up trivia. Hummingbird ships NetworkManager-ppp, both 1.58.0-1.hum1 and 1.58.1-1.hum1, and its metadata carries Requires: ppp. With ppp nowhere in the enabled set that plugin is uninstallable from Hummingbird's own repository, so NetworkManager's PPPoE and PPTP paths cannot be installed at all. It is the same shape as avahi (#104) and nautilus (#100): a plugin shipped without the daemon behind it. firefox needed no handling. It is already in [unavailable], and contract() filters that set out of every section, so moving it between sections cannot put it back in the image -- which matters, because Utah ships org.mozilla.firefox as a Flatpak and re-adding the RPM would regress that decision. This change installs nothing new and removes nothing. Proven rather than argued: contract() over main's two files and over this branch's both return the same 74 packages, with no additions and no removals, and libgda, libgda-sqlite, ppp and firefox are each absent from the resulting set. 23 tests pass, 45 subtests pass. The parity diff against upstream is now empty, the RPM contract check reports 58 Bluefin, 10 GNOME and 5 service packages, and install-packages.py's own guard confirms no [unavailable] package appears in the install set. Noted but deliberately not changed: evolution-ews-core, fish and zsh predate this and carry no issue reference, so the "every entry MUST carry a tracking issue" rule is not currently enforced by anything. Worth a guard, but not in a commit whose job is restoring the contract. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6BYLUD6KBzF8rBFXKjP7M
…nto claude/firmware-and-desktop-services
|
Unblocked. I've merged the parity fix in rather than waiting for it — head is now I said earlier I wouldn't widen this PR with the So Verified on the merged head before pushing: For the record on #108's own run, the same two steps that mattered here both came back green:
That second one is what I actually wanted, since it independently confirms the repodata reads behind this PR and that Generated by Claude Code |
CI resolved the real install transaction and found that two of the four
package additions cannot be installed. Name availability is not
installability, and checking the former is the mistake this corrects: I read
both names out of repodata, confirmed they exist, and stopped there.
Problem 1: conflicting requests
- nothing provides libgexiv2-0.16.so.4()(64bit) needed by
nautilus-51~beta-1.hum1.bfin.x86_64 from utah-packages
- nothing provides libportal-gtk4.so.1()(64bit) needed by nautilus...
- nothing provides libportal.so.1()(64bit) needed by nautilus...
Problem 2: cannot install the best candidate for the job
- nothing provides libdaemon.so.0()(64bit) needed by
avahi-0.9~rc4-0.1.1.hum1.x86_64 from public-hummingbird-x86_64-rpms
- nothing provides libdaemon >= 0.11 needed by avahi...
gexiv2, libportal, exiv2 and libdaemon are in no repository this image
enables, and in no factory recipe either -- not in the pinned factory image,
not in public-hummingbird-x86_64-rpms, and not among utah-packages#70's 340
recipes. So neither package becomes installable when #70 lands. nautilus
needs three new recipes (libportal, gexiv2, and exiv2 beneath it); avahi
needs one (libdaemon).
Both move to [unavailable] with that recorded, tracked by the issues that
already exist for them: #100 for nautilus, #104 for avahi. The parity debt
list is the honest place for a package whose blocker is named and unfixed
here.
This also corrects what I put on #100. I told the reporter nautilus needed no
factory work because the recipe was already in the pinned repository. The
recipe is there; the closure is not, and that is the same wall from one step
further in.
The avahi-daemon.service enablement goes with it, from both
configure-services.sh and 85-utah-desktop.preset. A preset line for a unit
whose package cannot be installed is dead configuration that reads like a
fix. tests/test_package_resolution.py grows
test_no_unit_is_enabled_for_an_uninstallable_package so the next attempt
cannot reintroduce that shape, and
test_packages_whose_closure_does_not_resolve_stay_out pins both packages with
the failing sonames in its docstring.
What survives is the part that actually resolved, and it is the part that
mattered most: [hardware] with linux-firmware and the four Intel wireless
firmware packages. None of those appeared in the transaction problems. #97 --
a ThinkPad X230 with no working WiFi -- is still fixed by this PR, as are #98
and #99, whose packages were installed all along and only needed enabling.
79 packages in the install set, no [unavailable] overlap, parity clean
against upstream, 33 tests and 58 subtests pass, contract check reports 58
Bluefin, 10 GNOME, 5 firmware and 5 service packages.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01W6BYLUD6KBzF8rBFXKjP7M
|
@castrojo could you review/approve? CI is green (including the image builds), but the ruleset needs an approval from someone other than the last pusher before it can be squash-merged. This is the one that fixes the X230 WiFi / Bluetooth / input-remapper reports (#97, #98, #99). Generated by Claude Code |
#110 landed the same libgda, libgda-sqlite and ppp entries with shorter notes; this branch's versions already carry them, so its text is kept. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6BYLUD6KBzF8rBFXKjP7M
…ady has The image-level parity report (#117) found 277 packages in Bluefin's image that Utah's lacks and that public-hummingbird-x86_64-rpms carries today: Bluefin never names them because Silverblue's base supplies them, and Hummingbird's server-shaped bootable base leaves them out. A new [parity] section in packages/utah.toml adds the desktop-relevant ones -- the NetworkManager Bluetooth and WWAN plugins and connectivity check, ALSA use-case profiles, the English language pack and core font set, and the shell tooling Bluefin users reach for -- 30 names, each with its Requires resolved against Hummingbird plus the pinned factory repository before it went in. Dakota ships the same networking, audio and font pieces. What did not resolve is recorded beside the list rather than dropped silently: cups needs cups-filters, vim-enhanced needs gpm, the emoji and math default fonts need google-noto-emoji-color-fonts and stix-fonts, and fprintd and alsa-utils are in no repository at all. NetworkManager-wifi waits for the factory image that carries wpa_supplicant and wireless-regdb (utah-packages#133). install-packages.py adds the section to the install set, verify-rpm-contract asserts it on the built image like [hardware], and the tests cover both paths plus a check that [parity] never duplicates a name another section already claims (unzip was one, via [build]). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6BYLUD6KBzF8rBFXKjP7M
|
3446d62 widens this PR from firmware and the hardware units to the rest of the base-image parity Hummingbird can satisfy today. A new What did not resolve is documented beside the list rather than dropped: The installer, the on-image contract verifier and the tests all know the new section. CI's Generated by Claude Code |
* fix: restore the bluefin.toml parity contract packages/bluefin.toml is meant to be a byte-for-byte copy of upstream Bluefin's build_files/packages/base.toml. utah.toml says so in as many words -- "it is the parity contract, and any drift there is a parity bug" -- and just check-parity exists to make that loud. It has drifted, so that check is red on main today and every open pull request inherits the failure, #94 and #105 among them. The drift is four hunks: firefox moved into [fedora], and libgda, libgda-sqlite and ppp missing. Syncing the file alone would have been worse than the failure it fixes. All three missing names were measured against both repositories the image actually enables -- the digest-pinned factory metadata layer from Containerfile, and public-hummingbird-x86_64-rpms: libgda absent from both libgda-sqlite absent from both ppp absent from both, and nothing Provides it So a bare sync hands three unresolvable names to the install transaction and turns a failing lint into a failing image build. The sync and the [unavailable] entries have to land together, which is why they are one commit. Each entry carries a tracking issue, as that list requires: libgda, libgda-sqlite (#106) Upstream added both for the maintained Copyous extension (bluefin#1210). Hummingbird carries no package whose name contains gda at all. One factory recipe covers both, because libgda-sqlite is a subpackage of the libgda SRPM rather than its own source -- Fedora's libgda.spec declares %package sqlite alongside ui, tools, sqlcipher, mysql and postgres. ppp (#107) Worth more than dial-up trivia. Hummingbird ships NetworkManager-ppp, both 1.58.0-1.hum1 and 1.58.1-1.hum1, and its metadata carries Requires: ppp. With ppp nowhere in the enabled set that plugin is uninstallable from Hummingbird's own repository, so NetworkManager's PPPoE and PPTP paths cannot be installed at all. It is the same shape as avahi (#104) and nautilus (#100): a plugin shipped without the daemon behind it. firefox needed no handling. It is already in [unavailable], and contract() filters that set out of every section, so moving it between sections cannot put it back in the image -- which matters, because Utah ships org.mozilla.firefox as a Flatpak and re-adding the RPM would regress that decision. This change installs nothing new and removes nothing. Proven rather than argued: contract() over main's two files and over this branch's both return the same 74 packages, with no additions and no removals, and libgda, libgda-sqlite, ppp and firefox are each absent from the resulting set. 23 tests pass, 45 subtests pass. The parity diff against upstream is now empty, the RPM contract check reports 58 Bluefin, 10 GNOME and 5 service packages, and install-packages.py's own guard confirms no [unavailable] package appears in the install set. Noted but deliberately not changed: evolution-ews-core, fish and zsh predate this and carry no issue reference, so the "every entry MUST carry a tracking issue" rule is not currently enforced by anything. Worth a guard, but not in a commit whose job is restoring the contract. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6BYLUD6KBzF8rBFXKjP7M * fix(e2e): capture the fastfetch shot at 1080p, with its logo The screenshot on the repository front page is the project's end-to-end evidence, and it was about to lose Utah's branding. #84 added fastfetch --logo none to the session bashrc, which suppresses the logo outright. The committed screenshot predates that change and still has the Utahraptor in it, so the next passing run would have quietly replaced it with a logo-less one. The workaround was reasonable, because the logo genuinely did not fit. fastfetch needs about 90 columns with it -- roughly 33 for the logo column plus a ~57 character detail line -- and Ghostty opens at 80x24, so lines like KVM/QEMU Standard PC (Q35 + ICH9, 2009) (pc-q35-11.1) wrapped to column 0 and pushed fastfetch's own header off the top of the window. The committed screenshot shows exactly that: its first visible line is the tail of a wrapped one, and the OS identity line is scrolled away. Columns are the lever, and point size on its own is not. Ghostty sizes its window in cells, so a smaller font yields a smaller window with the same 80 columns and the same wrap -- it would have traded the wrap for worse OCR and fixed nothing. What buys the columns is the display: both VMs now ask for 1920x1080 instead of the 1280x800 the machine default gave, and at that size a 110-column window at 12pt is about 880px of 1920. The logo fits with room to spare and the font goes UP from the default rather than down, which helps the tesseract gate that reads UTAH-E2E-FASTFETCH and Kernel out of the image. xres/yres are the EDID preferred-mode hints QEMU passes to the guest. -vga none replaces the machine default with the same stdvga device rather than a different adapter, so the graphics path is unchanged -- this is a mode hint, not a driver swap. Applied to the live VM as well as the installed one, since live-desktop.png is published in the same record. Ghostty is a flatpak here, and which config path it honours depends on whether the sandbox exposes xdg-config or keeps its own per-app directory, so the geometry is written to both ~/.config/ghostty/config and ~/.var/app/<app>/config/ghostty/config rather than guessing. Everything is overridable: UTAH_E2E_DISPLAY_WIDTH, UTAH_E2E_DISPLAY_HEIGHT, UTAH_E2E_FONT_SIZE, UTAH_E2E_COLUMNS, UTAH_E2E_ROWS. Verified what can be verified without a hypervisor. The remote script was rendered through a stub ssh_target and executed against a throwaway HOME: it writes both ghostty configs with the expected three keys, writes the autostart entry, and leaves `fastfetch && echo UTAH-E2E-FASTFETCH` in bashrc with no --logo none. VGA_ARGS expands to exactly four argv entries, so no word splitting. All shell scripts parse; 23 tests and 45 subtests pass. NOT verified here: that QEMU accepts VGA.xres/yres in this build, and that the guest picks 1920x1080 from the EDID hint. There is no hypervisor in this environment. Both fail loudly rather than silently if wrong -- a rejected device property aborts the VM at startup, and a display that stayed at 1280x800 would still wrap and fail the OCR gate. Separately noted, not fixed: docs/verification/screenshots/installed-desktop.png and installed-fastfetch.png are currently byte-identical (cd6577d6...). The desktop shot is taken after the terminal autostart has already mapped fastfetch, so two named checkpoints record one frame. That is a question about what installed-desktop is meant to prove, not about font size, so it is left alone here. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W6BYLUD6KBzF8rBFXKjP7M --------- Co-authored-by: Claude <noreply@anthropic.com>
castrojo
left a comment
There was a problem hiding this comment.
Reviewed at live head 3446d629 — the branch moved twice while I was reading it, so this is against what is there now (7 files, 381 additions), not the 6-file version the PR opened with.
The firmware and unit-enablement work is good. What blocks this is the 30-package [parity] sweep that arrived in 3446d629, which is both out of scope for this PR and currently red.
Blocking
1. The package transaction does not resolve. contract fails and every build job is skipped. From run 35363861546:
nothing provides libmbim-glib.so.4()(64bit) needed by ModemManager-1.24.2-1.hum1.bfin.x86_64 from utah-packages
nothing provides libmbim-utils needed by ModemManager-...
nothing provides libqmi-glib.so.5()(64bit) needed by ModemManager-...
nothing provides libqmi-utils needed by ModemManager-...
nothing provides libqrtr-glib.so.0()(64bit) needed by ModemManager-...
package NetworkManager-bluetooth-1:1.58.1-1.hum1 requires libnm-wwan.so()(64bit), but none of the providers can be installed
package NetworkManager-wwan-1:1.58.0-1.hum1 requires ModemManager, but none of the providers can be installed
plus poppler-utils requiring an unavailable poppler-data. So three of the new [parity] names — poppler-utils, NetworkManager-wwan, NetworkManager-bluetooth — have no installable closure in the pinned repositories.
This is the same shape you already documented for avahi (#104) and nautilus (#100): the factory ships a plugin without the daemon behind it. The factory's ModemManager is itself uninstallable, which is an upstream gap, not something you can fix here. Those three need to come out (or move to [unavailable] with tracking issues) until their closures are published.
2. Scope. The parity sweep is #117's work folded into a firmware/services PR. It is self-contained in 3446d629, so it lifts out cleanly. Worth noting the ordering: #117 measures the gaps, and 31 of the 35 names added here are already in #117's baseline — so whichever lands second has to update the other's numbers. Keeping them separate is what makes that tractable.
3. The title is now wrong. It says this ships "Files", but the head puts nautilus in [unavailable] (tracked by #100). It does not ship Files.
4. unzip parity is claimed but not delivered. The new note says unzip is "already in [build], which the image keeps", and [build] does contain it — but scripts/configure-services.sh:148 removes unzip after the extension build. This does not fail anything, because [build] is not part of the final RPM verifier, so no check catches it. It is a real final-image gap hiding behind a true-sounding statement. If the parity work stays, unzip belongs in [parity] and the removal has to stop.
Verified clean
packages/bluefin.tomlis byte-identical toorigin/main(git diff --exit-codeexit 0; both sides SHA-2567a4d8e36...). The31d765bdparity commit in your history is already on main via #110, so it contributes nothing to this diff.- None of the four kernel-cache-tag inputs change —
scripts/kernel-cache-tag.shprintsb547145c52fa7afeon bothorigin/mainand this head, so no 45-minute rebuild. - The generated skill catalog is current:
generate_skill_index.py --checkexits 0. [unavailable]additions (avahi,nautilus) both carry tracking issues, per the rule.python3 -m unittest tests.test_package_resolution— 22 tests, OK.
What I suggest
Drop 3446d629 from this branch and retitle to something like fix: ship device firmware and enable desktop hardware units. That single cut removes the failing transaction, the unzip mismatch, and the scope problem in one move, and leaves the firmware/Bluetooth/input-remapper work — which is sound — ready to land.
I did not make that cut myself: you pushed that commit deliberately and are actively working the branch, and choosing which parity names to keep versus park in [unavailable] is your call, not a mechanical fix I should guess at.
|
@castrojo done in 5ad6c1e: main merged (the Generated by Claude Code |
There was a problem hiding this comment.
Correctness review (head 5ad6c1e)
The code changes are internally consistent — manifest, resolver, verifier, preset, and tests all agree, and the suite passes at head (91 tests OK). One finding:
The PR body's Closes #97, #98, #99, #100, #104 no longer matches the diff — merging will auto-close three issues this PR does not fix.
- #98 (bluetooth): the diff deliberately defers it to #125 (
scripts/configure-services.sh:79-81,85-utah-desktop.preset:19), which is still open. - #104 (avahi): the body says avahi was "Added to
[services]" andavahi-daemon.serviceenabled — the diff does the opposite, placingavahiin[unavailable](packages/utah.toml:303) with a test asserting it stays out of the install set (tests/test_package_resolution.py,test_packages_whose_closure_does_not_resolve_stay_out). - #100 (nautilus): the body says "One line in
[gnome]" — the diff putsnautilusin[unavailable]instead (packages/utah.toml:309), documenting an unresolvable dependency closure.
The in-diff comments are accurate and honest about all of this; only the body/Closes line is stale. Please update the body to close only #97 and #99 (and describe avahi/nautilus as tracked, not fixed), or the issue tracker will show three real hardware bugs as resolved while the deployed image still exhibits them.
No other correctness issues: the [hardware] section flows through install-packages.py:57-59 and verify-rpm-contract.py:64-84 symmetrically, and the preset/script agreement is now enforced by DesktopUnitEnablementTests with a non-anchored regex that correctly catches the indented sshd case.
— hive: agent=reviewer backend=copilot model=claude-fable-5 copilot=1.0.78
…tbluefin#128) The live ISO grew from 7.7G (2026-09-06, last passing run) to 8.6G (2026-09-18) between builds. Root cause: the Containerfile copied the digest-pinned `packages` stage's RPM repository into a layer with `COPY --from=packages /repository /etc/utah-packages`, and nothing ever removed it. That ~4 GB is two thirds of every published image and of every live ISO, whose squashfs holds the whole mounted image. Bind the repository into the two RUN steps that install from it with a BuildKit `RUN --mount=type=bind,from=packages`, so it costs no layer and leaves nothing on disk. Reproducibility still comes from the digest-pinned `packages` stage, the only source the package transaction sees. The flavor step flips `utah-packages.repo` to enabled=0 last, so later dnf calls on the image (the live ISO build's included) do not fail on a file:// baseurl that no longer exists. Add a size budget to iso/scripts/build-iso.sh that fails closed above ISO_MAX_GB (override UTAH_ISO_MAX_GB) once the ISO is written, so drift stops landing silently. Raise N only after projectbluefin#105 (linux-firmware + parity packages) is evaluated. just check asserts the mount and refuses a `COPY --from=packages`. Signed-off-by: kylerankin <kylerankin@users.noreply.github.com>
…156) * fix: stop shipping the 4 GB package repository in every image (#128) The live ISO grew from 7.7G (2026-09-06, last passing run) to 8.6G (2026-09-18) between builds. Root cause: the Containerfile copied the digest-pinned `packages` stage's RPM repository into a layer with `COPY --from=packages /repository /etc/utah-packages`, and nothing ever removed it. That ~4 GB is two thirds of every published image and of every live ISO, whose squashfs holds the whole mounted image. Bind the repository into the two RUN steps that install from it with a BuildKit `RUN --mount=type=bind,from=packages`, so it costs no layer and leaves nothing on disk. Reproducibility still comes from the digest-pinned `packages` stage, the only source the package transaction sees. The flavor step flips `utah-packages.repo` to enabled=0 last, so later dnf calls on the image (the live ISO build's included) do not fail on a file:// baseurl that no longer exists. Add a size budget to iso/scripts/build-iso.sh that fails closed above ISO_MAX_GB (override UTAH_ISO_MAX_GB) once the ISO is written, so drift stops landing silently. Raise N only after #105 (linux-firmware + parity packages) is evaluated. just check asserts the mount and refuses a `COPY --from=packages`. Signed-off-by: kylerankin <kylerankin@users.noreply.github.com> * fix(iso): pass ISO_MAX_GB into the assembly guard heredoc The budget guard (#128) lived inside the <<'ASSEMBLY' quoted heredoc, which runs under 'podman unshare bash -s' in a separate set -u shell that only received the seven positional args. ISO_MAX_GB was a plain, unexported assignment in the outer script, so the guard died with 'ISO_MAX_GB: unbound variable' on every build -- after the ISO was already written -- and the UTAH_ISO_MAX_GB override never reached it either. Pass ISO_MAX_GB as an eighth positional arg alongside the others. Update the budget guard test, which injected ISO_MAX_GB itself and thus masked exactly this scoping failure, to stop injecting it and to assert the variable is actually delivered into the heredoc arg list. Fixes review feedback (#156). * test(Containerfile): fail fast on silent sed no-op; give budget test a PATH - Containerfile: assert enabled=0 actually took after the sed so a formatting change can't silently leave the repo enabled and break a later dnf call on a missing file:// baseurl. - test_iso_ci: pass PATH into the budget-guard harness so it no longer relies on bash's unset-PATH fallback for the cut pipe. --------- Signed-off-by: kylerankin <kylerankin@users.noreply.github.com>
|
Both correctness reviews were right about the same thing, and it was the most consequential defect on this PR even though it was entirely in prose: The body is now corrected. It closes #97 and #99 only. #98 is named as deferred to #125, and #100 and #104 are described as tracked with the actual reason each is in Worth noting where the bad text came from, because it is a pattern worth avoiding: the body described an earlier draft, the code moved when the resolver rejected it, and the body was never brought back into line. The in-diff comments stayed accurate throughout — which is why both reviewers could spot the mismatch — but the description is what GitHub acts on. @castrojo — your blocking review asked for the Bluetooth hunks to come out and for #125 to own that fix. That is the state at this head: Also updated in the body: #101 is no longer listed as needing a design decision — #153 fixed it by masking the vendor unit at Generated by Claude Code |
There was a problem hiding this comment.
intent-alignment
- input-remapper.service is enabled without a desktop-contract entry — the exact gap this PR's own rationale forbids. The PR body defers bluetooth to #125 because "Enabling the unit without that entry would leave Bluetooth on with nothing checking it." Yet this PR enables
input-remapper.serviceinscripts/configure-services.sh:89andsystem_files/shared/usr/lib/systemd/system-preset/85-utah-desktop.preset:18, whilecontracts/bluefin-desktop.toml[services].enabled(lines 131–141) is untouched — so the in-image verifier never asserts the #99 fix. Either add the entry or state why input-remapper is exempt from the standard applied to bluetooth.
docs-currency
docs/skills/package-contract.md:40–48enumerates the manifest sections and does not include the new[hardware]; line 114's install-set description ("release-specific Bluefin section, GNOME, services, and extension build tools") also omits it; and lines 127–129's counts ("12 Utah additions … 4 genuinely unavailable") are now wrong — this PR adds 5 hardware packages and movesavahi/nautilusinto[unavailable](packages/utah.toml:302,309).README.md:61–65"Package parity" table carries the same stale counts.docs/skills/desktop-contract.md:57–60lists the units the contract asserts; update alongside the contract entry above.
No findings from: security, style.
— hive: agent=reviewer backend=copilot model=claude-fable-5 copilot=1.0.78
castrojo
left a comment
There was a problem hiding this comment.
Reviewed diff: device firmware and hardware units. Blocked on failing 30-package parity sweep and unresolved package transaction in contract gate. Needs author fix.
The X230 report enabled input-remapper.service in configure-services.sh and the preset, but nothing asserted it: contracts/bluefin-desktop.toml [services].enabled did not list it, so verify-desktop-contract never checked the fix and a later regression would be silent — the same gap #98's bluetooth work was deferred to #125 for. Add input-remapper.service to the contract and to the desktop-contract skill doc that enumerates the asserted units. Also bring the package skill and README counts current: the overlay now carries [hardware] and [parity] (49 Utah additions, 11 unavailable), which the manifest header already describes but the docs still quoted as 12 and 4, and the install-set sentence omitted the two new sections. Assisted-by: Muse Spark via Muse Code Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
The overlay now carries [parity] (28 names) beyond GNOME and services, and the debt register holds 9 unavailable entries, but both README and the package-contract skill still quoted 12 Utah additions and 4 unavailable — the pre-parity numbers. The Failure semantics paragraph also listed only GNOME, services and build tools as the install set. Update both counts to 44/9 for this branch's reality (44 = 11 GNOME + 28 parity + 5 services) and include parity in the install-set sentence. The image-level parity section itself already describes the new mechanism; this fixes the stale table that contradicts it. When this lands with the hardware firmware PR (#105), the combined total will be 49/11 — that PR updates the same lines with the hardware addition. Assisted-by: Muse Spark via Muse Code Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
Pushed ddb85f6 to address the remaining review findings:
The Bluetooth collision flagged in the earlier CHANGES_REQUESTED was already handled in 5ad6c1e (hunks left to #125). This push completes the follow-up hive docs items. Ready for re-review. cc @castrojo |
There was a problem hiding this comment.
docs-currency
- README.md:63 and docs/skills/package-contract.md:129 still say 61 Bluefin contract packages installed, while updating the adjacent counts (49 additions, 11 unavailable) in the same table/sentence. Computed from the manifests at this head,
[fedora](65) minus the 7[unavailable]names it contains is 58 — the same figure this PR's own description quotes from the contract check. The stale 61 predates this PR, but since these exact lines were edited to "match the diff," the remaining number should be corrected too.
Verified independently: section counts (gnome 11 + hardware 5 + parity 28 + services 5 = 49; unavailable = 11) match the updated docs; [hardware] flows through contract() (scripts/install-packages.py:59) and the verifier's expected set (scripts/verify-rpm-contract.py:80); the preset/script agreement guard in tests/test_package_resolution.py passes at head (the two failures in my run are environmental — git ls-files outside a git checkout). input-remapper is installed via packages/bluefin.toml:62, so the new contracts/bluefin-desktop.toml:139 assertion is backed by a real package.
No findings from: correctness, security, intent-alignment, style.
— hive: agent=reviewer backend=copilot model=claude-fable-5
The parity table and the package-contract skill doc still claimed 61
Bluefin contract packages installed. The contract check resolves
[fedora] (65 names) minus the 7 [unavailable] names it contains, which
is 58:
scripts/verify-rpm-contract.py --check packages/bluefin.toml packages/utah.toml
Verifying 58 Bluefin packages, 11 GNOME desktop packages, 5 firmware
packages, 28 parity packages, 5 desktop service packages, and 0
NVIDIA packages
The additions (49) and unavailable (11) counts in the same rows were
already updated; this brings the third figure in line with them.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
The overlay now carries [parity] (28 names) beyond GNOME and services, and the debt register holds 9 unavailable entries, but both README and the package-contract skill still quoted 12 Utah additions and 4 unavailable — the pre-parity numbers. The Failure semantics paragraph also listed only GNOME, services and build tools as the install set. Update both counts to 44/9 for this branch's reality (44 = 11 GNOME + 28 parity + 5 services) and include parity in the install-set sentence. The image-level parity section itself already describes the new mechanism; this fixes the stale table that contradicts it. When this lands with the hardware firmware PR (#105), the combined total will be 49/11 — that PR updates the same lines with the hardware addition. Assisted-by: Muse Spark via Muse Code Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
Closing as out-of-scope package additions causing unresolvable RPM transaction in check-repos (missing ModemManager/libmbim dependencies). The firmware and unit enablement work should be split into an isolated, targeted PR. |
|
Reopening: reverting out-of-scope parity commit to isolate firmware and desktop unit enablement. |
linux-firmware is a split package whose Recommends skip Intel wireless, so the X230 from #97 would stay dead behind a closed-looking fix. Name iwlwifi-dvm/mvm/mld and iwlegacy explicitly per the analysis in #105. microcode_ctl remains out (no source). All five resolve on the pinned base (113 packages).
|
Closing as superseded in part: the firmware analysis landed on iso-optim-merged (linux-firmware plus explicit iwlwifi-dvm/mvm/mld and iwlegacy, exactly per your split-package finding — bare linux-firmware would have left the X230 dead). Thank you for that catch; it corrected my first attempt. The remaining pieces here (README counts, desktop preset, avahi/nautilusM) belong with their own issues (#104, #100) or a fresh PR against current main, since this branch conflicts throughout. |
linux-firmware is a split package whose Recommends skip Intel wireless, so the X230 from #97 would stay dead behind a closed-looking fix. Name iwlwifi-dvm/mvm/mld and iwlegacy explicitly per the analysis in #105. microcode_ctl remains out (no source). All five resolve on the pinned base (113 packages).
Closes #97, #99.
Tracked, not closed by this PR: #98 (bluetooth, deferred to #125), #100 (nautilus) and #104 (avahi) — see the sections below, which have been corrected to match the diff.
@KiKaraage's X230 reports were specific enough to act on directly — journal lines, a Bluefin side-by-side on the same machine, and the file that causes each one. Thank you, this was unusually actionable.
Every package name here was checked against the two repositories the image actually enables — the digest-pinned factory metadata layer from
Containerfile, andpublic-hummingbird-x86_64-rpms— by reading their repodata directly. That mattered twice, in opposite directions.linux-firmwarealone would not have fixed the WiFi (#97)This is the trap worth flagging.
linux-firmwareis a split package here: itRequiresonlylinux-firmware-whence, andRecommendsthirteen device packages (amd-gpu, atheros, brcmfmac, cirrus, intel-gpu, mt7xxx, nvidia-gpu, nxpwireless, qcom-wwan, realtek, tiwilink…). Intel wireless is in neither list.Measured against the pinned repository's filelists,
iwlwifi-6000g2a-6.ucode— the exact blob the report names — ships iniwlwifi-dvm-firmware. So addinglinux-firmwareand stopping there would have closed the issue while leaving the radio just as dead.New
[hardware]section:linux-firmware,iwlwifi-dvm-firmware,iwlwifi-mvm-firmware,iwlwifi-mld-firmware,iwlegacy-firmware— covering Intel wireless from the 3945 era through current parts. Everything else arrives via the recommends.microcode_ctlwas in my first draft and is deliberately dropped: it is in neither enabled repository, so naming it would have failed the install transaction outright. CPU errata updates remain a real gap and need a factory recipe first — worth its own issue.Files is tracked, not fixed here (#100)
An earlier draft of this PR added
nautilusto[gnome], on the evidence that the factory builds it and it is in the pinned repository at51~beta-1.hum1.bfin. That did not survive contact with the resolver: its closure does not install, because the accessories andexiv2it needs are not all published.nautilusis therefore in[unavailable]with #100 as its tracking issue, and a test asserts it stays out of the install set until the closure lands.gnome-disk-utilityandptyxisare in neither repository either.avahi is tracked, not fixed here (#104)
Hummingbird ships
avahi-libsandavahi-glibbut not the daemon. The daemon needs alibdaemonrecipe in the factory first, soavahiis in[unavailable]under #104, andconfigure-services.shcarries an explicit note thatavahi-daemon.serviceis deliberately not enabled — with the reason — so the next person does not add it and get a red transaction.The enablements, and why the preset matters (#99)
configure-services.shgainsinput-remapper.service. Bluetooth is deliberately not here: #125 owns that fix, and it is the one that also adds thecontracts/bluefin-desktop.tomlentry the in-image verifier asserts. Enabling the unit without that entry would leave Bluetooth on with nothing checking it.input-remapper.serviceis also added to85-utah-desktop.preset, which is the half that makes it stick: bootc applies presets on first boot, so a build-timesystemctl enablewith no preset line behind it gets undone. That is the trapconfigure-services.sh's own sshd branch documents, and #98 confirmed it from the other side — grepping the deployed image found no preset enablingbluetooth.serviceanywhere.tests/test_package_resolution.pynow holds that invariant, since it is the kind of thing the next unit added will get wrong. Three units predate the test (brew-setup,flatpak-nuke-fedora,flatpak-preinstall) and are listed as known exceptions rather than asserted correct — they may ship vendor presets in their own packages, and confirming that needs the built image, not the source tree. A second test fails if a name in that list stops being enabled, so the exception cannot go stale and quietly widen the hole.Verification
[hardware]line from the resolver, dropping*hardwarefrom the verifier's expected set, and deletinglinux-firmwarefrom the manifest each fail the suite and name what broke.just check-reposneeds podman/dnf, which this environment lacks, so the full transaction is unresolved locally. CI covers it, and the repodata reads above are why I am confident in the names.Not in this PR
/dev/nullin/usr/lib, which a cross-vendor/etcmerge cannot undo.[parity]sweepunzip,poppler-utils,NetworkManager-wwanandNetworkManager-bluetoothare handled.@KiKaraage — you offered to retest on that hardware. #97 in particular is worth confirming there, since the fix hinges on the
iwlwifi-dvm-firmwaresplit rather than onlinux-firmwareitself.🤖 Generated with Claude Code
https://claude.ai/code/session_01W6BYLUD6KBzF8rBFXKjP7M