Run what the change affects. CI runs the full suite on three platforms for every pull request, so the local job is fast feedback, not a second gate. The per-commit checks are in pnpm/CONTRIBUTING.md.
pnpm/, pnpr/)Run cargo and just from the repository root; the Rust workspace lives there, not inside pnpm/.
Run just test-affected. It decides three things for you: it selects every pnpr-* crate together, it refuses to scope a change that reaches files every crate compiles against and points at just ready instead, and it runs the smoke profile in place of dependents it did not select. It prints what it selected and what it left out; --help lists its flags.
What it cannot decide is which end-to-end tests exercise your change. Smoke gives breadth across areas, not depth in the one you touched, so for a user-visible change add the suite modules for that area:
\njust test-affected -- -p pnpm-cli -E 'test(catalog::)'\nEach file under crates/cli/tests/suite/ is a module of one test target, so test(<file_stem>::) selects that file's tests.
For anything narrower, node pnpm/scripts/run-rust-tests.mjs takes the same arguments cargo nextest run does — -p <crate> for one crate, -E 'test(<name>)' for one test. Prefer -p over a package() filterset: -p restricts what cargo builds, a filterset only selects among binaries that were built anyway.
run-rust-tests.mjs, not bare cargo nextest. It strips npm_config_* and pnpm_config_* and points XDG_CONFIG_HOME and the auth npmrc at a throwaway directory, as just test does. A bare run lets your own npmrc reach the tests, which fails for you and nobody else.pnpr-* crates must be selected together. Cargo unifies features across the selection, so a lone pnpr-* crate builds without pnpr's default backend features and its backend tests skip silently. just test-pnpr selects the whole set.insta snapshots change only for a reason. Read the diff, then cargo insta review. Never accept blindly.just sweep-test-temp clears the ones older than an hour.known_failures modules hold ported tests for unimplemented behavior; just known-failures lists them. A failure there is expected, a pass is not.Do not reach for nextest's rdeps() to widen a selection. pnpm-cli holds a third of the workspace's tests and sits downstream of nearly every crate, so rdeps() on anything core selects 84% or more of the suite. Restructuring that target is pnpm/pnpm#14984.
pnpm11/)pnpm --filter <package_name> test # one package\npnpm --filter <package_name> test <file_path> # one file\npnpm --filter <package_name> test <file_path> -t <name_pattern> # one case\nThe end-to-end tests in pnpm11/pnpm/test/ run the bundled pnpm11/pnpm/dist/pnpm.mjs, not each package's lib/. After changing any TypeScript package, rebuild the bundle before running them:
pnpm --filter pnpm run compile\nSkip that and the run tests the previous bundle, passing without ever touching your change.
\nA bug present in both pnpm v11 and v12 is fixed in both, so it is tested in both. Run the TypeScript test for the scenario and the Rust test for the same scenario before calling the fix done.
\nName what you ran. \"Ran pnpm-lockfile plus the catalog e2e module; did not run the full workspace suite\" is an honest report. \"Tests pass\" after a single-crate run is not.