test

Run the requested suites (default: all) against the requested build type (default `RelWithDebInfo`). ## Where it runs On the box that built clice (the build skill; the caller passes its id), not on this machine unless truly necessary; the tools' own tests need no build and run here. A box this skill leases because the caller passed none is reported with its id, like the build skill's, and never downed. Run the suites as CI does, in the `test-run` environment: `box run $id 'pixi run -e test-run test RelWithDebInfo'`. Windows: build on a linux box, run on a windows box, as CI does. A windows box builds nothing (no Bazel cache, only `test-run`). ```bash lin=$(box up clice-io/clice) # the linux box also restores CI's Windows cache win=$(box up clice-io/clice --os windows) # arm64: --os windows-arm, aarch64-w64-mingw32 below box push $lin && box push $win box run $lin 'pixi run build RelWithDebInfo -- --platforms=@xclang//platforms:x86_64-w64-mingw32' box cp $lin $win build/RelWithDebInfo/bin/bin build/RelWithDebInfo/bin/lib --exclude '*.runfiles' box run $win 'pixi run -e test-run unit-test RelWithDebInfo' ``` The copy is a few hundred MB (the debug info), a few minutes through the two tunnels. ## Commands - Unit tests: `pixi run -e test-run unit-test [type]` - Cases: `pixi run -e test-run cases-test [type]` - Both: `pixi run -e test-run test [type]` - The tools' own tests, which need no build: `pixi run npm test -w tools` Filtering specific tests: - Unit: `pixi run -e test-run unit-test [type] -- --test-filter=SuiteName.CaseName` (also `SuiteName.*`) - Cases: `CLICE_CASES=<path prefix> pixi run -e test-run cases-test [type]`, e.g. `CLICE_CASES=hover/` or `CLICE_CASES=hover/templates/06_template_forms`; the "Running" section of `tests/README.md` lists the rest. The compat suite (`pixi run compat-test`) and the editors' end-to-end tests (`pixi run -e editor editor-test`) need build systems and editors installed; run them only when the caller names them. This skill runs and reports — nothing else. Never fix code or tests from here, never set `UPDATE_SNAPSHOTS`. Fixing happens in the main conversation with full context. Report back, per suite: pass/fail and counts. For each failure: the test name (a case's is `<topic>/<subtopic>/<case> > <test>`), the assertion message or `snap.toml` diff excerpt (trimmed), and the exact filter command to reproduce it.