spartan-ng__spartan/apps/cli-smoke

cli-smoke

Smoke tests that verify the spartan CLI against every supported workspace setup, by scaffolding real workspaces, running the generators, consuming a generated component, and building the result. This catches the class of bug that the in-memory generator unit tests cannot: broken tsconfig path mapping, missing dependencies, generated code that does not compile or is not consumable through its public alias, secondary-entrypoint wiring, and theme styles that never reach the build output.

They are smoke tests, not full end-to-end tests: they assert the output compiles, is consumable, and builds with themed styles, but they do not run the app or assert runtime behaviour.

What it does

1. global-setup starts a local verdaccio registry (the local-registry target on the root spartan project), then builds and publishes @spartan-ng/cli and @spartan-ng/brain to it under a forced version (9999.0.0-smoke.<timestamp> - unique per run so pnpm's cache can't serve a stale build). They share a version because the CLI writes the @spartan-ng/brain dependency at the installed CLI version.
2. For each cell in src/matrix.ts it scaffolds a fresh workspace (in the OS temp dir, not the repo), pins Tailwind v4 via the documented .postcssrc.json setup, installs the locally published CLI, writes components.json, runs init + ui <component>, asserts the CLI's own healthcheck reports no issues on the generated output, drops in a probe component that uses the generated code through the @spartan-ng/helm/* alias, then builds and asserts the theme CSS variables reached the output. A generator error, a healthcheck failure, a non-zero build, or missing themed CSS fails the cell.
3. global-teardown stops the registry.

The matrix

• nx monorepo x {library, entrypoint} x {buildable, non-buildable}
• nx-standalone - single-app nx workspace with no tsconfig.base.json
• acli - Angular CLI, no Nx
• all-components - generates every supported primitive into an Angular CLI app and builds them all in one ng build; the slowest cell, but fans out to its own runner. Set nightlyOnly: true in src/matrix.ts to drop it from per-PR CI if it proves too slow.

Tailwind is fixed at v4. Add or remove combinations in src/matrix.ts.

Running

# full matrix (also: `pnpm smoke`)
pnpm nx run cli-smoke:smoke

# a single cell (jest matches the whole test name, so use a substring, not an anchored id)
pnpm nx run cli-smoke:smoke --testNamePattern="nx-lib-buildable scaffolds"

Each cell installs dependencies and runs a real Angular build, so it is slow; locally the cells run single-threaded (one shared local registry). On failure the scaffolded workspace is left under <os-tmp>/spartan-cli-smoke-workspaces/<cell-id> for inspection; on success it is removed.

CI

• .github/workflows/cli-smoke.yml runs the matrix, fanning each cell out to its own runner so they run in parallel from the start - wall-clock is roughly the slowest single cell, not the sum.
  • It runs on every push/PR (no path filter) so the cli-smoke result job can be a required check that never sits pending. Cheapness on unrelated PRs comes in two layers: (1) a git diff gate before pnpm install skips the job in seconds when nothing CLI-relevant changed; (2) when it does install, the in-runner isSmokeAffected() check (src/support/affected.ts) makes the precise nx-affected call and still short-circuits to a pass if cli-smoke isn't actually affected. nrwl/nx-set-shas provides the base/head both layers use.
  • A fast drift-guard job fails CI if matrix.ts grows a cell the workflow's static matrix doesn't list (keep the two in sync).
  • cli-smoke result aggregates the matrix + drift-guard into one status to require in branch protection.
• nightly-release.yml runs the full matrix unconditionally every night (single job; no base/head, so the affected check always runs it).