make check is the one canonical local verification command. It runs, in
order, the same gates as CI:
| Target | Command | Catches |
|---|---|---|
make tidy-check | go mod tidy -diff | untidy go.mod / go.sum |
make fmt-check | gofmt -l . | unformatted files |
make vet | go vet ./... | suspicious constructs |
make lint | pinned golangci-lint | lint findings (CI enforces this) |
make test | go test ./... | test failures |
make race | go test -race ./... | data races |
CI (.github/workflows/ci.yml) runs each of these targets as a separate step
so a failure names the gate; it must stay equivalent to make check. Add new
gates to the Makefile first, then to check and CI together.
make test-backtest runs the one-config EMA-cross regression
(testdata/backtest/ema-cross, docs/arch/testing.md). It is part of make test; run it alone while
refactoring backtest orchestration.
- Before pushing, and before saying work is done: run full
make check. - While iterating, run the narrowest target that covers your change, for
example
go test ./path/to/pkgormake test. These are for feedback only and never replacemake checkbefore a push. - Documentation-only changes need no Go checks, but say so in the PR.
- There is no separate “quick” target: on a warm cache the whole of
make checktakes about 15 seconds, so a subset would add a second command to keep in sync without saving meaningful time.
In PR descriptions and handoff comments, record each command you ran, its
exit status, and the relevant failure lines (not the whole log), e.g.
make check: exit 0 or make lint: exit 2, errcheck at foo.go:12. List any
gate you did not run and why; “tests pass” without a command is not evidence
(see docs/workflows/implementation-checklist.org).