Migrate CI from Cirrus CI to GitHub Actions#447
Conversation
fkorotkov-oai
force-pushed
the
dev/fkorotkov/migrate-ci-to-github-actions
branch
from
ad5cd25 to
262d469
Compare
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 0b796b0ae0
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| - name: Build Vetu | ||
| working-directory: _vetu | ||
| run: | | ||
| go build -o "$RUNNER_TEMP/vetu" cmd/vetu/main.go |
There was a problem hiding this comment.
When the Linux job builds Vetu from source, this setcap replaces the old apt-installed package setup but grants only cap_net_raw and cap_net_admin. Vetu's own install instructions say the binary needs cap_net_raw,cap_net_admin,cap_net_bind_service+eip; without cap_net_bind_service, the integration tests that later run vetu run can fail when Vetu binds privileged networking service ports, even though the earlier vetu pull step succeeds.
Useful? React with 👍 / 👎.
| owner: openai | ||
| name: homebrew-tools |
There was a problem hiding this comment.
This moves formula publishing to openai/homebrew-tools, but the README still tells users to run brew install cirruslabs/cli/orchard. After the next release is published only to the new tap, users following the documented install path will keep getting the stale Cirrus formula instead of current Orchard; either update the install docs to the new openai/tools tap or keep publishing the old tap as well.
Useful? React with 👍 / 👎.
| ARG VERSION=dev | ||
| ARG COMMIT=unknown | ||
|
|
||
| RUN CGO_ENABLED=0 go build \ |
There was a problem hiding this comment.
Do we need to resort to manual building here?
In this PR we still use GoReleaser in .github/workflows/release.yml, can we keep using it here too?
edi-oai
left a comment
edi-oai
left a comment
There was a problem hiding this comment.
LGTM, but it seems that the Linux integration test is stuck for some reason.
2 participants
What changed
publishenvironment for tag releases while keeping manual snapshots environment- and secret-freeFormula/orchard.rbtoopenai/homebrew-toolsthrough a versioned pull requestghcr.io/openai/orchardand source Vetu fromopenai/vetuin Linux CIInstallation
Once the formula is published to the OpenAI tap:
Release setup
Configure these secrets on the
publishenvironment before the next release:RELEASE_APP_IDRELEASE_APP_PRIVATE_KEYGORELEASER_KEYThe release app needs contents write access to
openai/orchard, plus contents and pull-request write access toopenai/homebrew-tools.Only the tag-triggered release job uses the
publishenvironment. Manualworkflow_dispatchsnapshots run without release, signing, or notarization secrets.Validation
actionlint .github/workflows/release.ymlactionlint -ignore 'label ".*" is unknown' .github/workflows/*.ymlgoreleaser check --softwith GoReleaser Pro 2.16.0goreleaser release --skip=publish --snapshot --clean, which generateddist/homebrew/Formula/orchard.rband skipped signing/notarizationgo test ./...passed outside hardware-dependent integration packagesGoReleaser reports that formula-style
brewsoutput is deprecated in favor of casks. This PR intentionally preserves Orchard's formula packaging and its install-time shell completion generation.The
macos-26GitHub-hosted ARM runner runs every package exceptinternal/tests, covering the non-VM/unit suite. The Linux Cirrus runner continues to rungo test ./...with Vetu, retaining Linux hardware integration. Tart/macOS VM cases ininternal/testsare intentionally not run on GitHub-hosted macOS because those ARM runners do not support the nested virtualization they require.