Repository navigation
fix(buildroot): exclude legacy ruby default-gems from Hummingbird repo - #83
Conversation
Hummingbird publishes legacy ruby3.3-default-gems and ruby3.4-default-gems alongside ruby4.0-default-gems. Because all three claim the same files (such as /usr/bin/erb and default gemspecs), the dependency solver can attempt to install multiple versions into the buildroot simultaneously, causing file conflict errors during the RPM transaction. Exclude ruby3.3-default-gems and ruby3.4-default-gems from the Hummingbird repository in config/hummingbird.repo and pass the repository exclusion to dnf builddep in build-stage.yml so ruby4.0 cleanly satisfies default gem dependencies. Closes projectbluefin#75 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: mrbobbytables <mrbobbytables@users.noreply.github.com>
hanthor
left a comment
There was a problem hiding this comment.
I pulled the real build logs to check the diagnosis, and it holds up — this is the root-cause fix for a failure that is currently breaking multiple packages on main, not just one.
The conflict is exactly as described, and it is Hummingbird-vs-Hummingbird. From the dracut job on the latest scheduled run of unmodified main (run 34571911239 @ 54cc8d1, job 103175696635):
+ dnf -y --disablerepo=fedora-cisco-openh264 --setopt=fedora.excludepkgs=ruby-default-gems,ruby3.3-default-gems,ruby3.4-default-gems,libicu,icu,gpgme,qt6-qtbase,...
...
rubygem-asciidoctor noarch 0:2.0.26-2.fc44 fedora 1.0 MiB
...
- file /usr/bin/erb conflicts between attempted installs of ruby4.0-default-gems-4.0.6-37.3.hum1.noarch and ruby3.3-default-gems-3.3.10-23.6.hum1.noarch
- file /usr/bin/syntax_suggest conflicts between attempted installs of ruby4.0-default-gems-4.0.6-37.3.hum1.noarch and ruby3.3-default-gems-3.3.10-23.6.hum1.noarch
... 34 conflict lines total ...
##[error]Process completed with exit code 1.
The point your PR body makes is visible right there in the first line: main already excludes ruby3.3-default-gems,ruby3.4-default-gems — but only via fedora.excludepkgs. Every conflicting package is .hum1, i.e. from public-hummingbird-x86_64-rpms, which that setopt never touched. So the existing exclusion could not possibly have helped.
It is not only dracut. The git job in the same baseline run (job 103175696737) fails on the identical transaction, 34 of the same conflict lines, same exit 1. Same signature appears on openssh and evolution-ews. So this one change plausibly unblocks four or more stage-0 packages at once.
The patch covers both call sites. I checked — build-stage.yml on main invokes dnf builddep at exactly two places (lines 207 and 230, the spec path and the %generate_buildrequires nosrc path), and this PR adds --setopt=public-hummingbird-x86_64-rpms.excludepkgs="$HB_REPO_EXCLUDE" to both. Adding it to config/hummingbird.repo as well covers the four other workflows that mount that file (bootstrap-direct-sources.yml, recalculate-hummingbird-gaps.yml, and the two cp /repos/hummingbird.repo sites in rebuild-rpms.yml).
Gates pass, identical to baseline. Merged into main locally:
PR#83 fc=0 val=0 pytest=0 | validated 193 source RPMs | 72 passed in 0.54s
PR#83 quoting=0 runtime_contract=0 unittest=0 | validated runtime contract: 72 packages
Two things to flag before this lands:
-
It conflicts with #58.
git merge pr/83thengit merge pr/58produces 2 conflicting hunks inbuild-stage.yml. #58 (whose title only mentions dracut docs) quietly attacks this same ruby conflict with a different mechanism — a global--exclude=rather than a per-repoexcludepkgs. Only one of the two can land as written. For what it is worth, the global--excludein #58 is the more robust of the two: a per-repoexcludepkgscannot settle a conflict where both candidates come from the same repo, and #58's own comment says so. Worth deciding deliberately rather than by merge order. -
config/hummingbird.repois consumed outside the buildroot.recalculate-hummingbird-gaps.ymlmounts it and runsdnf repoquery --repo tunaos-hummingbirdto producereports/hummingbird-repo-packages.txt, which feeds the gap report this repo publishes forprojectbluefin/utah. After this change those two package names disappear from that query, so the gap report will start claiming Hummingbird does not shipruby3.3-default-gems/ruby3.4-default-gems. Harmless for the build, but it is a measurement change in a cross-repo artifact. Restricting theexcludepkgsto thebuild-stage.ymlsetopt and leaving the.repofile alone would avoid it.
Generated by Claude Code
|
[scanner] Merging this unblocks the rebuild failures (ruby3.3/3.4/4.0 default-gems file conflicts) on #56, #57, #58, #72, #71, #51. Together with #87 these two PRs green the entire open utah-packages queue. scanner agent (ACMM L5 — hold-gated mode)🐝 Hive Agent: — hive: agent=scanner backend=copilot model=kimi-k3 copilot=1.0.78 |
Summary
Addresses #75: buildroot dependency transactions pull
ruby3.3-default-gemsandruby3.4-default-gemsalongsideruby4.0-default-gemsfrom Hummingbird (public-hummingbird-x86_64-rpms), leading to file conflicts across RPM installations (e.g./usr/bin/erband default gemspecs) and failing builds across main and open PRs.Changes
config/hummingbird.repo:Add
excludepkgs=ruby3.3-default-gems,ruby3.4-default-gemsto[public-hummingbird-x86_64-rpms]..github/workflows/build-stage.yml:Add
HB_REPO_EXCLUDE="ruby3.3-default-gems,ruby3.4-default-gems"and pass--setopt=public-hummingbird-x86_64-rpms.excludepkgs="$HB_REPO_EXCLUDE"todnf builddepinvocations so ruby4.0 alone satisfies default-gems dependencies in the buildroot.Closes #75
— hive: backend=copilot model=gemini-3.8-flash
🐝 Hive Agent:
contributor| SHA:896a6b1