test

Run the requested suites (default: all).

• Everything: pixi run test [preset] — builds the preset and installs the npm packages, then runs ctest: zest's bootstrap stages, its runner check, then unit_tests and system_tests, each from the repo root with --snapshot-dir=tests/snapshots (both are required for snapshot tests; ctest handles them for you), and beside them the integration tests, one ctest test per module (ipc_lsp_integration) that runs the module's integration/*.test.ts on node's test runner. ctest --preset <preset> -L unit (or system, bootstrap, integration) runs one label.
• Integration tests alone: pixi run integration-test [preset] (ctest -L integration). ctest passes each driver's path in KOTATSU_<DRIVER> (e.g. KOTATSU_LSP_STUB_SERVER); a test whose driver is unset or missing fails, it is never skipped.
• Coverage (Linux and macOS): pixi run coverage builds the coverage preset (clang source-based coverage), runs every stage with each process writing its profile into build/coverage/profiles/, and reports on include/kotatsu and src in build/coverage/report/: summary.md (lines, functions and regions per module), html/index.html and coverage.lcov. Its arguments go to ctest (pixi run coverage -L unit). CI's linux | clang | coverage leg puts the table in the job summary and uploads the report as the coverage-report artifact. zest runs on async, cli and support, so part of their coverage is the runner's own; code only the compiler evaluates (constexpr) never runs and counts as not covered.

Filtering tests means invoking a binary directly — replicate what ctest does (repo root, snapshot dir) and add the filter:

./build/<preset>/unit_tests --snapshot-dir=tests/snapshots --test-filter=Suite.Case
./build/<preset>/system_tests --snapshot-dir=tests/snapshots --test-filter='async_io_*'

(Suite.* and bare positional patterns also work.) Rerun integration tests by running node the way ctest does (ctest --preset <preset> -N -V -L integration prints its command and environment) — from the repo root, through pixi (the host node is not the project's), with the driver's path set (.exe on Windows) — and add a name pattern:

KOTATSU_LSP_STUB_SERVER=build/<preset>/lsp_stub_server pixi run node --import ./tests/check_npm_packages.ts --test --test-timeout=20000 --test-name-pattern='^hover_' tests/ipc/lsp/integration/requests.test.ts

Rules:

• This skill runs and reports — nothing else. Never fix code or tests from here, never skip or weaken anything, never pass --update-snapshots (snapshot updates are a deliberate act in the main conversation).
• When running a binary directly, build the preset first (the build skill); a stale test binary silently tests old code, and so does a stale driver such as lsp_stub_server.

Report back, per suite: pass/fail and counts. For each failure: test name, file:line, the assertion message or snapshot-diff excerpt (trimmed), and the exact filter command to reproduce it.