go clean — Optimization¶

go clean mostly trades speed for disk. Optimization here means cleaning surgically and not undermining the caches you rely on. Numbers are illustrative.

Exercise 1: Stop cleaning caches you persist¶

Before — CI restores GOCACHE/GOMODCACHE for speed, then runs go clean -cache each job: every build is cold.

After — remove the per-job clean entirely:

# restore cache; do NOT clean it
- uses: actions/cache@v4
  with: { path: ~/.cache/go-build, key: gocache-${{ hashFiles('**/go.sum') }} }
- run: go build ./...

Metric	clean each job	persist, no clean
Build time per job	cold (~90s)	warm (~8s)

The single biggest "optimization" is not cleaning.

Exercise 2: Threshold-based cleaning on persistent runners¶

Before — a self-hosted runner is wiped fully on a fixed schedule, throwing away a still-useful warm cache.

After — clean only when disk pressure warrants it:

used=$(df --output=pcent "$(go env GOMODCACHE)" | tail -1 | tr -dc 0-9)
if [ "$used" -gt 85 ]; then go clean -modcache; fi

Metric	always wipe	threshold
Cache warmth	frequently cold	usually warm
Disk safety	fine	fine

Exercise 3: Target the test cache narrowly¶

Before — go clean -testcache to re-run one package expires all results, so the whole suite re-runs next time.

After:

go test -count=1 ./onepackage/...

Metric	global `-testcache`	targeted `-count=1`
Tests re-run	entire suite	one package

Exercise 4: Re-warm efficiently after a necessary wipe¶

Before — after go clean -modcache, the first build serially re-downloads everything on the critical path.

After — pre-download in a parallel/early step:

go clean -modcache
go mod download        # bulk download up front, before build/test steps

Metric	download during build	pre-download
Critical-path time	download + build	build only (deps ready)

Exercise 5: Cold-build verification without slowing every job¶

Before — every PR runs a clean build "to be safe," paying full recompile each time.

After — run the cold build only on a schedule or pre-release:

# nightly job only
go clean -cache && go build ./... && go test ./...

Metric	clean build every PR	scheduled cold build
PR feedback time	slow (cold)	fast (warm)
Reproducibility check	redundant	still covered

Exercise 6: Clear fuzz cache to bound disk, keep seeds¶

Before — long fuzzing runs grow the generated corpus cache unbounded on a runner.

After:

go clean -fuzzcache     # clears generated corpus, keeps testdata/fuzz seeds

Metric	unbounded corpus	cleared cache
Disk used by fuzz	grows	reset to seeds
Seed corpus	intact	intact

Measurement checklist¶

Do not go clean -cache/-modcache per CI job if you persist caches.
Clean module cache by disk threshold on persistent runners.
Use -count=1 for targeted test re-runs, not global -testcache.
go mod download up front after a necessary -modcache wipe.
Run cold builds on a schedule, not every PR.
-fuzzcache to bound fuzz corpus disk without losing seeds.