Shared vitest configuration for TypeScript LIBRARY packages: the pinned reporters, the test excludes, the coverage block, and defineLibraryVitestConfig, which wraps a package’s own projects in them.
The Obsidian plugin configuration (@mnaoumov/obsidian-dev-toolkit/test-runners/vitest-config) builds on the same reporters and excludes, so the two cannot drift apart.
| Function |
Description |
| defineLibraryVitestConfig |
Defines a library package’s vitest configuration: its own projects, wrapped in the shared reporters, excludes and coverage block, with console output from passing tests suppressed. |
| Type |
Description |
| VitestReporter |
One entry of a vitest reporters list: the name of a built-in reporter ('default', 'junit', 'minimal', …) or a reporter instance. Spelled out rather than derived from vitest’s own option type, whose parts (ReporterName, InlineReporter) are internal to vitest and exported from none of its entry points. Naming the two shapes directly keeps a reporters list a plain array a caller pushes onto, which the option type itself cannot be, because it is a union of an array with a single value. Vitest’s third accepted shape, a [name, options] tuple, is deliberately left out: no repository on this toolchain uses one. |
| Variable |
Description |
| DEFAULT_VITEST_REPORTERS |
The reporters every repo on this toolchain prints through. Pinned, because vitest’s fallback is not one reporter but two. An unset reporters resolves to isAgent ? 'minimal' : 'default', and std-env’s isAgent is true inside every AI coding session — it reads AI_AGENT plus a dozen tool-specific variables. MinimalReporter extends the default one with summary: false, silent: 'passed-only', and a printTestModule that returns early unless the module FAILED. So the SAME npm run test, on the SAME tree, prints a line per test file for a developer and not one for a session, and swallows a passing test’s console output in the second case only. Measured on the unit-tests:script-utils project of obsidian-dev-utils (43 files, vitest 5.0.1, 2026-09-23): 46 per-file lines over a 58-line log pinned, 0 over a 13-line log unset. A gate should not report differently depending on who runs it, and a console.log a developer adds to find something should not vanish because a session ran the suite. The cost of the pin is the ~45 lines above. What this deliberately does NOT carry is a reporter naming the file a dead pool worker took with it, because vitest 5 already does that itself. emitUnexpectedExit is (code, signal) => … there, building its message from the exit code, the signal, the runner state and the files the worker held. Driven against real runs on 2026-09-23 in all four shapes: a worker SIGKILLed mid-file, one killed at import before collection, the same under minimal, and the realistic one — the dying file among 42 others in a parallel run — which reported 42 passed (43) AND named the file. Under vitest 4 none of that existed and the reader was left subtracting over file names the log did not contain, which is what a reconciling reporter was written for; on 5 it would be a module, a test file and a coverage burden for every consumer, carrying a claim that is no longer true. |
| DEFAULT_VITEST_TEST_EXCLUDE |
Folders vitest never collects tests from. |
| LIBRARY_VITEST_COVERAGE |
The coverage options of a library package: v8 over its own src/, without the generated barrels, the declarations, the tests and the test helpers. The threshold is the test:coverage script’s to set. |