brooktrails/gllm
Watch
2
0
Fork
You've already forked gllm
 0
Code Issues 3 Pull requests 2 Projects Releases Packages Wiki Activity Actions

chore(go): move the toolchain to Go 1.27 #74

Merged
rcsheets merged 1 commit from chore/go-1.27 into main 2026-09-17 08:40:12 +00:00
Conversation 1 Commits 1 Files changed 6 +17 -6
rcsheets commented 2026-09-16 08:14:59 +00:00
Owner
Copy link

Moves the toolchain floor to Go 1.27: go.mod, the three image builds, and the AGENTS.md repo fact.

Changes

  • go.mod: go 1.26 -> go 1.27.
  • Dockerfile, Dockerfile.cuda: golang:1.26 -> golang:1.27.
  • Dockerfile.cuda-prerelease: golang:1.27-trixie, not the bare tag. It installs NVIDIA's trixie packages, and where golang:1.26 was trixie, the bare golang:1.27 tag resolves to a different base -- so the suffix now carries weight the comment used to get for free.
  • ci.yaml: a step printing go version and go env GOTOOLCHAIN GOPROXY.

Verified locally (go1.27.1)

gofmt, go vet ./..., go test ./..., go build ./..., and the cuda-tagged build all pass.

What this PR's CI run is actually testing

The pool's runner image bundles the toolchain (the workflow sets no container: and no setup step), so no run has ever recorded which Go built it -- hence the new step. That matters because go.mod's directive is only a floor: a runner older than it downloads a toolchain through GOPROXY, which fails closed if the proxy will not serve golang.org/toolchain.

The runner image (forgejo-runner-operator, RUNNER_GO_VERSION) still pins golang:1.26-alpine, and those images are built by hand, so the deployed forgejo-runner-go:12-latest is very likely 1.26. I could not confirm it: Harbor needs credentials, and Athens refuses requests from outside the cluster (403 on everything, /healthz included), so whether it serves toolchain modules in-cluster is unknown from here.

So this run resolves it either way:

  • passes -- the runner is already 1.27, or it downloaded 1.27 through Athens. Nothing else to do.
  • fails on the toolchain -- bump RUNNER_GO_VERSION to 1.27-alpine in forgejo-runner-operator (plus the README row), rebuild and push forgejo-runner-go:12-latest, restart the pool, then re-run this PR.

🤖 Generated with Claude Code

Moves the toolchain floor to **Go 1.27**: `go.mod`, the three image builds, and the AGENTS.md repo fact. ## Changes - `go.mod`: `go 1.26` -> `go 1.27`. - `Dockerfile`, `Dockerfile.cuda`: `golang:1.26` -> `golang:1.27`. - `Dockerfile.cuda-prerelease`: `golang:1.27-trixie`, **not** the bare tag. It installs NVIDIA's trixie packages, and where `golang:1.26` was trixie, the bare `golang:1.27` tag resolves to a different base -- so the suffix now carries weight the comment used to get for free. - `ci.yaml`: a step printing `go version` and `go env GOTOOLCHAIN GOPROXY`. ## Verified locally (go1.27.1) `gofmt`, `go vet ./...`, `go test ./...`, `go build ./...`, and the `cuda`-tagged build all pass. ## What this PR's CI run is actually testing The pool's runner image bundles the toolchain (the workflow sets no `container:` and no setup step), so **no run has ever recorded which Go built it** -- hence the new step. That matters because `go.mod`'s directive is only a floor: a runner older than it downloads a toolchain through `GOPROXY`, which fails closed if the proxy will not serve `golang.org/toolchain`. The runner image (`forgejo-runner-operator`, `RUNNER_GO_VERSION`) still pins `golang:1.26-alpine`, and those images are built by hand, so the deployed `forgejo-runner-go:12-latest` is very likely 1.26. I could not confirm it: Harbor needs credentials, and Athens refuses requests from outside the cluster (403 on everything, `/healthz` included), so whether it serves toolchain modules in-cluster is unknown from here. So this run resolves it either way: - **passes** -- the runner is already 1.27, or it downloaded 1.27 through Athens. Nothing else to do. - **fails** on the toolchain -- bump `RUNNER_GO_VERSION` to `1.27-alpine` in `forgejo-runner-operator` (plus the README row), rebuild and push `forgejo-runner-go:12-latest`, restart the pool, then re-run this PR. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
chore(go): move the toolchain to Go 1.27
All checks were successful
ci / test_and_build (pull_request) Successful in 1m51s
Details
2a09a8085d 
go.mod's floor, the three image builds, and the AGENTS.md repo fact all move
from 1.26 to 1.27. Verified locally on go1.27.1: gofmt, vet, the full test
suite, the pure-Go build, and the cuda-tagged build.

Dockerfile.cuda-prerelease pins golang:1.27-trixie rather than the bare tag.
It installs NVIDIA's trixie packages, and where golang:1.26 was trixie, the
bare golang:1.27 tag now resolves to a different base, so the suffix carries
weight the comment used to get for free.

CI prints the toolchain it ran with, which nothing recorded before: the pool's
runner image bundles Go (no container:, no setup step), so a run's Go version
was invisible. It matters here because go.mod's directive is only a floor -- a
runner older than it downloads a toolchain through GOPROXY, and that fails
closed if the proxy does not serve golang.org/toolchain.

The runner image (forgejo-runner-operator, RUNNER_GO_VERSION) still pins
golang:1.26-alpine, so this PR's own CI run is the test of whether that
download path works. If it does not, the runner image needs the same bump
before this can land.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
rcsheets commented 2026-09-16 08:27:38 +00:00
Author
Owner
Copy link

Blocked on forgejo-runner-operator#53, which builds forgejo-runner-go on Go 1.27.

The CI run here answered the question this PR was opened to settle -- the pool's runner is 1.26 and cannot download a newer toolchain:

go: go.mod requires go >= 1.27 (running go 1.26.5; GOTOOLCHAIN=local)

GOTOOLCHAIN=local comes from the official golang image, so there is no fetch path regardless of what Athens serves. The new "Go toolchain" step is what makes that visible.

Order: merge #53, rebuild and push forgejo-runner-go from main (make container-build-runner-go, then push), get the pool onto the new image (it pins the moving 12-latest with no imagePullPolicy, so IfNotPresent can keep a cached copy -- pinning the immutable 12-<VERSION> tag is the deterministic fix), then re-run this PR.

Blocked on [forgejo-runner-operator#53](https://git.brooktrails.org/brooktrails/forgejo-runner-operator/pulls/53), which builds `forgejo-runner-go` on Go 1.27. The CI run here answered the question this PR was opened to settle -- the pool's runner is 1.26 and cannot download a newer toolchain: ``` go: go.mod requires go >= 1.27 (running go 1.26.5; GOTOOLCHAIN=local) ``` `GOTOOLCHAIN=local` comes from the official `golang` image, so there is no fetch path regardless of what Athens serves. The new "Go toolchain" step is what makes that visible. Order: merge #53, rebuild and push `forgejo-runner-go` from `main` (`make container-build-runner-go`, then push), get the pool onto the new image (it pins the moving `12-latest` with no `imagePullPolicy`, so `IfNotPresent` can keep a cached copy -- pinning the immutable `12-<VERSION>` tag is the deterministic fix), then re-run this PR.
rcsheets deleted branch chore/go-1.27 2026-09-17 08:40:13 +00:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
brooktrails/gllm!74
No description provided.