build: isolate golangci-lint cache and lock per checkout (closes #121)

script/lint used golangci-lint's per-user global state, which breaks
when several checkouts on one host lint concurrently. Two independent
failure modes, two causes:

- Cross-contamination. The analysis cache (GOLANGCI_LINT_CACHE,
  default ~/.cache/golangci-lint) is keyed by content, not by
  checkout, so a hit written by another checkout is replayed with
  that checkout's file paths. A run reports findings for files it
  never linted.

- Lock collision. golangci-lint locks os.TempDir()/golangci-lint.lock
  (pkg/commands/run.go, acquireFileLock), which is NOT in the cache
  directory, with a 5s timeout. Peers that hold it longer make the
  run abort with "parallel golangci-lint is running" - a non-result
  that reads as a lint failure. Isolating the cache does not move it.

Point GOLANGCI_LINT_CACHE and TMPDIR at .lint-cache/ under the
checkout root, using the existing $ROOT idiom. TMPDIR is what makes
the lock per-checkout, so the lock keeps serialising the runs that
actually share a cache instead of being disabled. .lint-cache/ is
git-ignored and Docker-ignored, and the cache persists across runs in
a checkout, so caching is not lost.

Reproduced both modes on the unfixed script across 12 copies of this
tree: 10 of 12 concurrent runs void with the lock error, and 11 of 12
sequential runs reported findings at ../w1/... after an identical
lint-failing file was added to every copy. After the fix, 20-way
concurrency gives 0 void and 0 foreign paths, and each copy reports
only its own relative path.

.dockerignore

@@ -1,5 +1,6 @@
 .git/
 bin/
+.lint-cache/
 *.md
 LICENSE
 .editorconfig

.gitignore

@@ -1,6 +1,7 @@
 bin/
 vendor/
 data/
+.lint-cache/
 .env
 *.exe
 /dnswatcher

Dockerfile

@@ -15,25 +15,8 @@ RUN go mod download
 
 COPY . .
 
-# Run all checks - build fails if any check fails.
+# Run all checks - build fails if any check fails
-#
+RUN make check
-# CHECK_EPOCH is a cache-busting build argument. Without it, an
-# unchanged tree leaves this layer's cache key identical and Docker
-# serves the previous verdict instead of re-running the suite, so the
-# build reports a green it did not earn. A build argument's value
-# participates in the cache key of later instructions in the stage even
-# when they do not reference it, so a fresh value busts this layer
-# either way. It is expanded into the command deliberately: that makes
-# the invalidation a property of the command string itself rather than
-# of how a given builder treats unreferenced args, and it surfaces the
-# epoch in the build log as a diagnostic.
-#
-# Placing the ARG here and nowhere earlier keeps everything above it
-# (toolchain install, go mod download) cached, so only the check and the
-# steps after it re-run. script/cibuild passes a fresh value per run; a
-# plain `docker build` without it caches as before.
-ARG CHECK_EPOCH
-RUN echo "check epoch: ${CHECK_EPOCH}" && make check
 
 # Build the binary
 RUN make build

README.md

@@ -387,16 +387,16 @@ them. We provide:
 - `script/projectname` — print the project name (used for the Docker
   image tag)
 - `script/test` — run the test suite (race detector, coverage)
-- `script/lint` — run golangci-lint
+- `script/lint` — run golangci-lint, with its cache and its lock file
+  isolated to this checkout (under the git-ignored `.lint-cache/`) so
+  concurrent checkouts on one host cannot share cache entries or
+  contend on a single lock
 - `script/fmt` — format all code (gofmt -s, goimports)
 - `script/fmt-check` — check formatting (read-only)
 - `script/check` — run test, lint, and fmt-check
 - `script/docker` — build the Docker image tagged via
   `script/projectname`
-- `script/cibuild` — CI entrypoint: `docker build .` with a fresh
+- `script/cibuild` — CI entrypoint: plain `docker build .`
-  `CHECK_EPOCH` build argument, so the Dockerfile's `make check` layer
-  is never served from the cache and a green build always means the
-  checks ran on this invocation
 - `script/precommit` — run by the git pre-commit hook; `go mod tidy`
   guard, then `script/check`
 - `script/install-precommit` — install the git pre-commit hook

TODO.md

@@ -25,16 +25,15 @@ confirm make check still passes.
 
 # Completed Steps
 
-- 2026-08-09: `script/cibuild` can no longer report a green it did not
+- 2026-08-09: `script/lint` now isolates golangci-lint's per-user global
-  earn. The Dockerfile declares `ARG CHECK_EPOCH` immediately above the
+  state to the checkout (#121): `GOLANGCI_LINT_CACHE` and `TMPDIR` are
-  check step and expands it into the `RUN` command, and `script/cibuild`
+  both pointed at the git-ignored, Docker-ignored `.lint-cache/`. The
-  passes a fresh `$(date +%s%N)` per invocation, so the `make check`
+  cache fixes cross-contamination; `TMPDIR` is what moves the lock,
-  layer is always re-executed while the pinned toolchain install and
+  which lives at `$TMPDIR/golangci-lint.lock` and not in the cache
-  `go mod download` stay cached. Verified by experiment: before the fix
+  directory. Reproduced both failure modes on the unfixed script (10 of
-  a second run on an unchanged tree returned in 283 ms with the check
+  12 concurrent runs void with `parallel golangci-lint is running`; 11
-  layer `CACHED`; after it the check runs every time, and a deliberately
+  of 12 reporting another checkout's paths) and both are gone at 20-way
-  planted always-failing test made the build fail with exactly that
+  concurrency after the fix
-  test's message
 - 2026-08-07: golangci-lint bumped to v2.12.2 (commit-pinned installs
   in `Dockerfile` and `script/bootstrap`); `.golangci.yml` set to the
   org-standard v2-schema config used across the org's repos

script/cibuild

@@ -1,18 +1,13 @@
 #!/bin/sh
-# script/cibuild: run the CI build. The Dockerfile runs make check, and
+# script/cibuild: run the CI build. The Dockerfile runs make check, so
-# the CHECK_EPOCH build argument below is fresh on every invocation, so
+# a successful build implies all checks pass.
-# the check layer is never served from the Docker layer cache: a
-# successful build means the checks were executed and passed on this
-# run, not on some earlier one. Only the check step and the steps after
-# it are invalidated; the toolchain install and go mod download stay
-# cached.
 set -eu
 
 ROOT="$(cd "$(dirname "$0")/.." && pwd -P)"
 
 main() {
     cd "$ROOT"
-    docker build --build-arg CHECK_EPOCH="$(date +%s%N)" .
+    docker build .
 }
 
 main "$@"

script/lint

@@ -1,11 +1,40 @@
 #!/bin/sh
 # script/lint: run the linter.
+#
+# golangci-lint keeps two pieces of per-user global state, and both of
+# them break when several checkouts on one host lint concurrently:
+#
+#   1. Its analysis cache (GOLANGCI_LINT_CACHE, default
+#      ~/.cache/golangci-lint). Entries are keyed by content, not by
+#      checkout, so a hit written by another checkout is replayed
+#      verbatim - including that checkout's file paths. The run then
+#      reports findings for files it never linted.
+#
+#   2. Its "one runner at a time" lock, which does NOT live in the
+#      cache directory: golangci-lint locks
+#      $(os.TempDir())/golangci-lint.lock, i.e.
+#      "$TMPDIR"/golangci-lint.lock (pkg/commands/run.go,
+#      acquireFileLock). It waits 5s, then aborts with "parallel
+#      golangci-lint is running" - a non-result that looks like a lint
+#      failure. Setting GOLANGCI_LINT_CACHE alone does not move it.
+#
+# So both are pinned under the checkout root. The cache is never shared,
+# and TMPDIR makes the lock file per-checkout, which keeps the lock
+# doing its actual job (serialising runs that share one cache) at the
+# right scope. .lint-cache/ is git-ignored and Docker-ignored, and
+# caching still works: it persists across runs in this checkout.
 set -eu
 
 ROOT="$(cd "$(dirname "$0")/.." && pwd -P)"
 
 main() {
     cd "$ROOT"
+
+    GOLANGCI_LINT_CACHE="$ROOT/.lint-cache/cache"
+    TMPDIR="$ROOT/.lint-cache/tmp"
+    export GOLANGCI_LINT_CACHE TMPDIR
+    mkdir -p "$GOLANGCI_LINT_CACHE" "$TMPDIR"
+
     golangci-lint run --config .golangci.yml ./...
 }