RELEASING.md (Gate section) states CI deploy.yml runs task check before building images. deploy.yml actually inlines its own sequence (task fmt; git diff; proto lint/check; go vet; golangci-lint; go build; go test ./... -count=1), which is a superset in some steps and a different test mode (uncached full suite vs task check's -short) in others. The divergence surfaced 2026-10-11 when the new Docker-only internal/container test (skipped under -short) broke the release gate on a runner without the alpine:3 image; that was fixed by pulling the image, but the two definitions of the gate remain. Decide: make deploy run task check (one owner, matching the doc) or keep the full-suite gate and update RELEASING.md, so the two stop drifting.
Imported from beads archie-core-t1fks · GitHub #1600 · type task · priority P3 · created 2026-10-10
RELEASING.md (Gate section) states CI deploy.yml runs `task check` before building images. deploy.yml actually inlines its own sequence (task fmt; git diff; proto lint/check; go vet; golangci-lint; go build; go test ./... -count=1), which is a superset in some steps and a different test mode (uncached full suite vs task check's -short) in others. The divergence surfaced 2026-10-11 when the new Docker-only internal/container test (skipped under -short) broke the release gate on a runner without the alpine:3 image; that was fixed by pulling the image, but the two definitions of the gate remain. Decide: make deploy run `task check` (one owner, matching the doc) or keep the full-suite gate and update RELEASING.md, so the two stop drifting.
---
_Imported from beads `archie-core-t1fks` · GitHub #1600 · type task · priority P3 · created 2026-10-10_
Triaged: no code change is needed for this. Already resolved — no code change needed.
The issue describes deploy.yml inlining its own gate sequence (task fmt; git diff; proto lint/check; go vet; golangci-lint; go build; go test ./... -count=1) and RELEASING.md claiming task check. Neither is true of the tree as it stands:
.github/workflows/deploy.yml:93-99 — the gate job pulls alpine:3 and then runs task check TEST_FLAGS=-count=1 followed by git diff --exit-code. No step list of fmt/lint/vet/build is inlined any more; the gate body lives in Taskfile.yml:279-302 (check → ui, fmt, proto:check, db:check, check:readonly → proto:lint, news:check, vet, lint, build, test, test:ui). That is the "one owner" option the issue leaned toward.
Taskfile.yml:268-270 — test runs go test {{.TEST_FLAGS | default "-short"}} ./..., so TEST_FLAGS=-count=1 replaces -short outright: uncached full suite, Docker-backed tests included, exactly as documented.
RELEASING.md:72-74 — "Gate.task check TEST_FLAGS=-count=1 (uncached, including the Docker-backed tests -short skips) and govulncheck, then a check that the tag has a [version] section in CHANGELOG.md." That matches the deploy job step for step, and internal/container/kit_test.go:17-19,51 is the testing.Short() Docker test that uses alpine:3, so the pull step the 2026-10-11 incident added is the right and only image dependency (no other test file in internal/container names an image).
.github/workflows/quality.yml (the PR gate) also calls task --concurrency 1 check rather than a private step list, so both gates share the Taskfile definition.
The repo is at 1.61.2 (2026-11-07), a month past the issue's 2026-10-10 filing, so the convergence happened after the report and the ticket is stale.
One residual, below the threshold of a change: RELEASING.md's Gate bullet doesn't mention the git diff --exit-code step both workflows run after task check. That's a missing clause about a dirtiness check, not a second definition of the gate. Follow-up if wanted: add five words to that bullet. Nothing else to build, and no workflow to route.
Triaged: no code change is needed for this. Already resolved — no code change needed.
The issue describes deploy.yml inlining its own gate sequence (task fmt; git diff; proto lint/check; go vet; golangci-lint; go build; go test ./... -count=1) and RELEASING.md claiming `task check`. Neither is true of the tree as it stands:
- `.github/workflows/deploy.yml:93-99` — the gate job pulls `alpine:3` and then runs `task check TEST_FLAGS=-count=1` followed by `git diff --exit-code`. No step list of fmt/lint/vet/build is inlined any more; the gate body lives in `Taskfile.yml:279-302` (`check` → `ui`, `fmt`, `proto:check`, `db:check`, `check:readonly` → `proto:lint`, `news:check`, `vet`, `lint`, `build`, `test`, `test:ui`). That is the "one owner" option the issue leaned toward.
- `Taskfile.yml:268-270` — `test` runs `go test {{.TEST_FLAGS | default "-short"}} ./...`, so `TEST_FLAGS=-count=1` replaces `-short` outright: uncached full suite, Docker-backed tests included, exactly as documented.
- `RELEASING.md:72-74` — "**Gate.** `task check TEST_FLAGS=-count=1` (uncached, including the Docker-backed tests `-short` skips) and `govulncheck`, then a check that the tag has a `[version]` section in `CHANGELOG.md`." That matches the deploy job step for step, and `internal/container/kit_test.go:17-19,51` is the `testing.Short()` Docker test that uses `alpine:3`, so the pull step the 2026-10-11 incident added is the right and only image dependency (no other test file in `internal/container` names an image).
- `.github/workflows/quality.yml` (the PR gate) also calls `task --concurrency 1 check` rather than a private step list, so both gates share the Taskfile definition.
The repo is at 1.61.2 (2026-11-07), a month past the issue's 2026-10-10 filing, so the convergence happened after the report and the ticket is stale.
One residual, below the threshold of a change: RELEASING.md's Gate bullet doesn't mention the `git diff --exit-code` step both workflows run after `task check`. That's a missing clause about a dirtiness check, not a second definition of the gate. Follow-up if wanted: add five words to that bullet. Nothing else to build, and no workflow to route.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
RELEASING.md (Gate section) states CI deploy.yml runs
task checkbefore building images. deploy.yml actually inlines its own sequence (task fmt; git diff; proto lint/check; go vet; golangci-lint; go build; go test ./... -count=1), which is a superset in some steps and a different test mode (uncached full suite vs task check's -short) in others. The divergence surfaced 2026-10-11 when the new Docker-only internal/container test (skipped under -short) broke the release gate on a runner without the alpine:3 image; that was fixed by pulling the image, but the two definitions of the gate remain. Decide: make deploy runtask check(one owner, matching the doc) or keep the full-suite gate and update RELEASING.md, so the two stop drifting.Imported from beads
archie-core-t1fks· GitHub #1600 · type task · priority P3 · created 2026-10-10Triaged: no code change is needed for this. Already resolved — no code change needed.
The issue describes deploy.yml inlining its own gate sequence (task fmt; git diff; proto lint/check; go vet; golangci-lint; go build; go test ./... -count=1) and RELEASING.md claiming
task check. Neither is true of the tree as it stands:.github/workflows/deploy.yml:93-99— the gate job pullsalpine:3and then runstask check TEST_FLAGS=-count=1followed bygit diff --exit-code. No step list of fmt/lint/vet/build is inlined any more; the gate body lives inTaskfile.yml:279-302(check→ui,fmt,proto:check,db:check,check:readonly→proto:lint,news:check,vet,lint,build,test,test:ui). That is the "one owner" option the issue leaned toward.Taskfile.yml:268-270—testrunsgo test {{.TEST_FLAGS | default "-short"}} ./..., soTEST_FLAGS=-count=1replaces-shortoutright: uncached full suite, Docker-backed tests included, exactly as documented.RELEASING.md:72-74— "Gate.task check TEST_FLAGS=-count=1(uncached, including the Docker-backed tests-shortskips) andgovulncheck, then a check that the tag has a[version]section inCHANGELOG.md." That matches the deploy job step for step, andinternal/container/kit_test.go:17-19,51is thetesting.Short()Docker test that usesalpine:3, so the pull step the 2026-10-11 incident added is the right and only image dependency (no other test file ininternal/containernames an image)..github/workflows/quality.yml(the PR gate) also callstask --concurrency 1 checkrather than a private step list, so both gates share the Taskfile definition.The repo is at 1.61.2 (2026-11-07), a month past the issue's 2026-10-10 filing, so the convergence happened after the report and the ticket is stale.
One residual, below the threshold of a change: RELEASING.md's Gate bullet doesn't mention the
git diff --exit-codestep both workflows run aftertask check. That's a missing clause about a dirtiness check, not a second definition of the gate. Follow-up if wanted: add five words to that bullet. Nothing else to build, and no workflow to route.