Make a missing CHECK_EPOCH fail the build (closes #91)
All checks were successful
check / check (push) Successful in 3m2s

PR #89 stopped script/cibuild replaying cached check layers, but left a
gap: a bare `docker build .` with no --build-arg still faked. An unset
ARG is an empty string, an empty string is a stable cache key, and the
check layers replay from it. That gap mattered because REPO_POLICIES.md
names `docker build .` verbatim as a command that must be green, so the
documented command was the one that lied.

Both check stages now carry `RUN [ -n "$CHECK_EPOCH" ] || exit 1`
immediately under their own ARG. Failed steps are never cached, so this
fails on every invocation rather than once - a bare build now stops with
a named error instead of reporting a green it did not earn. Each stage
needs its own guard because ARG scope is per-stage; a gate-carrying stage
without one is a silent hole if ordering ever changes.

The check RUNs now reference the value (`echo "check epoch: ${CHECK_EPOCH}"
&& make <target>`), so the cache miss is contractual rather than resting
on BuildKit's current treatment of unreferenced ARGs, and the epoch is
visible in the build log.

The epoch becomes "$(date +%s%N)$$" so concurrent invocations in the same
second cannot collide. busybox silently drops %N and exits 0, so $$ is
what makes it correct there. The bare-assignment form is retained
deliberately: inlining the substitution into --build-arg would, under
set -eu, yield an empty and therefore constant epoch without aborting.

script/docker gets the same treatment - it is not the gate, but two
entrypoints disagreeing about whether the tree is green is its own
hazard, and local builds are almost always warm.

Verified by negative control rather than inspection: a bare build fails
twice consecutively here and succeeds twice on the parent commit, so the
change is demonstrably not a no-op. The builder-stage guard was fired
directly with a targeted probe build, since the lint stage otherwise
fails first and would leave it unexercised.
This commit was merged in pull request #92.
This commit is contained in:
2026-08-09 10:09:27 +02:00
parent c3bb3b5580
commit 50816b7415
5 changed files with 102 additions and 40 deletions

36
TODO.md
View File

@@ -19,6 +19,42 @@ or delete the branch.
# Completed Steps
- 2026-08-09: Adopted the remaining upstream `CHECK_EPOCH` hardening
(issue #91), closing the gap #85 knowingly left open. Four changes,
all four decided as adopt upstream in `sneak/prompts` #26. (1) Each
check stage now asserts `[ -n "$CHECK_EPOCH" ] || exit 1` before
running anything, so a build that supplies no `--build-arg` fails
instead of lying. This is the item that mattered: an unset `ARG` is
an empty string and an empty string is a stable cache key, so the
second and every later bare `docker build .` on an unchanged tree
replayed all three check layers and still exited 0 — and `docker
build .` is the command `REPO_POLICIES.md` names verbatim as a thing
that must be green, so the documented command was precisely the one
that lied. Failed steps are never cached, which is what makes the
guard fire on every invocation rather than once. (2) The epoch is now
expanded into each check command rather than left as a bare
declaration, so the cache miss no longer depends on BuildKit's
unreferenced-`ARG` handling staying as it is, and the value appears
in the build log. (3) `script/cibuild` uses
`epoch="$(date +%s%N)$$"`, unique per invocation rather than per
second; `%N` alone is insufficient because busybox drops it silently
and exits 0, and `$$` is what makes the guarantee hold regardless.
The bare-assignment form is kept deliberately — inlined in an
argument, a failing substitution does not abort under `set -eu` and
would yield an empty constant epoch, restoring the exact false green
being fixed. (4) `script/docker` passes the same fresh arg, so the
two entrypoints cannot disagree about whether the tree is green;
local builds are almost always warm, which made it the likelier
fooling in practice. The `ARG` placement from #85 is unchanged, below
`apk add`, `COPY go.mod go.sum` and `go mod download`, so dependency
layers still cache and the build is not cold. Verified by negative
control rather than inspection — a bare `docker build .` run twice
back to back, plus back-to-back pairs of both scripts and a host-side
`make check`; the measurements are recorded once, in the PR
verification comment, rather than restated here. `.golangci.yml`, the
lint-stage `FROM` line and its digest, `script/lint`,
`REPO_POLICIES.md` and `.gitea/workflows/check.yml` are all
untouched.
- 2026-08-09: Stopped `script/cibuild` from reporting a green it did
not earn (issue #85). A bare `docker build .` let Docker serve the
check layers from the layer cache whenever the tree had not changed: