Lectures

We break down why, in a CLI, it is better not to write directly to os.Stdout and not to read from os.Stdin, but to inject dependencies via io.Writer/ io.Reader. We design a command contract Run(...) that returns an exit code, keep main thin, separate stdout/stderr, and configure flag.FlagSet output via SetOutput. As a result, the CLI becomes predictable and easy to test without “intercepting” global I/O.

We break down how to test a CLI in Go in a table-driven style: turning a command run into a verifiable contract args → exit code + stdout/ stderr, capturing output via bytes.Buffer, feeding stdin via strings.NewReader, adding stdin support to cases, using t.Run and a helper with t.Helper(), and also discussing when it’s better to assert whole messages vs. fragments ( strings.Contains).

We break down the golden files approach for testing CLI text output in Go: where to store snapshots ( testdata/), how to write helpers readGolden/ writeGolden, how to safely update expectations via the -update flag ( go test), and which typical mistakes destroy trust in these tests (non-determinism, mixing stdout/ stderr, “invisible” diffs forever due to the final newline).

We break down why CLI tests are “sometimes red” and turn them deterministic: we extract time via a dependency Clock/ FakeClock instead of calling time.Now() directly, lock down the time format ( UTC() + RFC3339), and also stabilize output order by sorting tasks (instead of a random range over a map). At the end: how this helps golden tests, and which mistakes most often break stability.

We break down why tests tied to “real” files next to the project fail in CI and on other machines, and how to isolate the environment. In practice, we use t.TempDir() for honest on-disk file scenarios and fstest.MapFS together with io/fs for fast unit tests without I/O. Plus: path nuances, verifying file contents, and common mistakes.