> For AI agents: the complete documentation index is available at /llms.txt, the full documentation bundle is available at /llms-full.txt.

# Announcing Rstest 0.11

_July 6, 2026_


![9aoy](https://github.com/9aoy.png)

9aoy

[](https://github.com/9aoy)

@9aoy

![fi3ework](https://github.com/fi3ework.png)

fi3ework

[](https://github.com/fi3ework)

@fi3ework

![Rstest 0.11](https://assets.rspack.rs/rstest/rstest-banner-v0-11.png)
Rstest 0.11 is a transitional release on the road to 1.0. Its focus is to settle several public APIs through breaking changes ahead of the stable release; alongside that, it also brings faster V8 coverage and updated fake timers.

Notable changes:

- [API changes on the road to 1.0](#api-changes)
- [Faster V8 coverage](#coverage-v8)
- [Fake timers aligned with `@sinonjs/fake-timers` 15](#fake-timers)
- [Other improvements](#other-improvements)

## API changes on the road to 1.0 \{#api-changes}

To stabilize the public API before 1.0, Rstest 0.11 adjusts a few surfaces. These are breaking changes, but the migrations are straightforward.

### `TestOptions` moves to the second argument

The [`TestOptions`](/api/runtime-api/test-api/test.md#testoptions) object (`retry` / `repeats` / `timeout`) for `test` / `it` / `test.each` / `test.for` was previously accepted as the **last** argument, after the test function. With a long test body it was easy to miss, and formatters tended to push it onto a separate trailing line. It now goes in the **second** position, before the test function, matching the shape used by Vitest and Node.js' built-in test runner (`node:test`).

```diff
- test('name', () => {}, { retry: 2 });
+ test('name', { retry: 2 }, () => {});
```

The numeric timeout shorthand is unchanged — the Jest-style `test('name', fn, 1000)` still works as the last argument. `describe` now accepts a `TestOptions` object in the same second-argument position.

### Mock factories are synchronous-only

The `rs.mock` / `rs.doMock` factory type is now synchronous-only. Async factories are prone to hoisting and module-initialization issues, so TypeScript no longer suggests them. For partial mocks, keep the real module via a synchronous `importActual` import attribute and override only the exports you need:

```ts
import * as actual from './date-utils' with { rstest: 'importActual' };

rs.mock('./date-utils', () => ({
  ...actual,
  formatDate: rs.fn().mockReturnValue('2026-03-19'),
}));
```

See [Partially mock modules](/guide/basic/mock.md#partially-mock-modules) for more details.

### `pool.minWorkers` is removed

`pool.minWorkers` set a floor on retained idle workers, but it only applied under `isolate: false` — the default `isolate: true` never reuses workers, so the option had no effect. It is removed so that `pool` settles on the flat `{ type?, maxWorkers?, execArgv? }` shape; use `pool.maxWorkers` to control file-level parallelism. The internal floor still defaults to `min(maxWorkers, recommended)`, so default warm-worker behavior is unchanged.

### `shard` is CLI-only

Sharding is inherently per-invocation — every CI runner passes a different index, which a value committed to a shared config file cannot express. The `shard` config field is removed from `RstestConfig`; use the [`--shard <index>/<count>`](/guide/basic/cli.md#sharding-tests) CLI flag instead. The flag and its behavior are unchanged.

```bash
npx rstest run --shard=1/3
```

## Faster V8 coverage \{#coverage-v8}

On a large bundled project, collecting V8 coverage could spend tens of seconds turning raw V8 data into Istanbul coverage. Rstest 0.11 reworks the conversion path.

Previously, the provider ran every raw V8 entry through the generic `ast-v8-to-istanbul` converter, including many that never reach the final report. Rstest 0.11 uses a converter tailored for Rstest's output, filters out irrelevant entries before converting, and moves the conversion from each worker into the main process.

Measured on the [Rsbuild](https://github.com/web-infra-dev/rsbuild) repository's own test suite (`pnpm test --coverage --coverage.provider=v8`):

| Version         | Total time |
| --------------- | ---------: |
| baseline (0.10) |      6.05s |
| Rstest 0.11     |      1.81s |

Compared with the baseline, the reworked path is about **3.34x faster**. Resolving raw coverage in the main process further cuts Rstest's own processing time (about **2x** on the same project) and lowers CPU usage. The coverage output is unchanged.

There is no configuration change — projects using `coverage.provider: 'v8'` get the speedup automatically.

## Fake timers aligned with `@sinonjs/fake-timers` 15 \{#fake-timers}

Rstest 0.11 upgrades to `@sinonjs/fake-timers@15.4.0` and aligns the fake timers runtime with its API. Two methods are now available:

- `rstest.jumpTimersByTime(ms)` jumps the clock forward by `ms` without running the timers in between, so any timer scheduled within that span fires at most once, at the destination.
- `rstest.setTickMode({ mode })` switches the clock between manual and auto-advancing modes.

```ts title="example.test.ts"
import { expect, rstest, test } from '@rstest/core';

test('jump without running intermediate timers', () => {
  rstest.useFakeTimers();
  const fn = rstest.fn();

  setInterval(fn, 1000);
  rstest.jumpTimersByTime(5000);

  expect(fn).toHaveBeenCalledTimes(1);
});
```

`rstest.setSystemTime(date)` also works on its own now — without a prior `rstest.useFakeTimers()` call it pins only the global `Date` and leaves the real timers intact, so you can freeze the system clock without taking over `setTimeout` / `setInterval`.

This release also closes a Browser Mode gap. Browser Mode previously aliased `@sinonjs/fake-timers` to a minimal stub that kept a clock-shaped object but did not actually fake browser globals such as `setTimeout` or `Date`. Rstest 0.11 removes that stub and bundles the real implementation, so fake timers behave consistently across Node mode and Browser Mode — including `Date` / `performance`, animation frames, microtasks, and string durations. See [Fake Timers](/api/runtime-api/rstest/fake-timers.md) for more details.

## Other improvements \{#other-improvements}

- **Clearer error for `rs.spyOn` on ESM namespace exports.** Previously, spying on a read-only ESM namespace export produced an opaque failure; it now produces an actionable error explaining why the binding cannot be reassigned.
- **`new URL()` and Wasm resolve from on-disk source.** Assets referenced via `new URL(..., import.meta.url)` and Wasm modules now resolve from the original source location, matching runtime behavior.
- **Shared module state under `isolate: false`.** Imported module state is now shared and context-bound APIs stay live across files when `isolate: false`, so cross-file singletons behave as expected.
- **Browser Mode project isolation.** Each project now runs in its own Rsbuild instance, fixing cross-project module resolution in multi-project browser runs.
- **OS-native `testPath`.** `testPath` is now surfaced with the platform's native path separators.
- **Custom `runCLI` argv.** Programmatic callers can pass a custom `argv` array to `runCLI`, while `process.argv` remains the default.

## Upgrade

Upgrade the `@rstest/*` packages to version 0.11 to get these improvements. Rstest 0.11 includes a few breaking API changes — see [API changes on the road to 1.0](#api-changes) for the migration steps.

For a full list of changes, see the [v0.11.0 release notes](https://github.com/web-infra-dev/rstest/releases/tag/v0.11.0).