vitest-mobile@0.4.3

Patch Changes

• 706afc9: Fix bootstrap failing on iOS for projects that depend on react-native-reanimated@4. The React Native community CLI's --force-pods path calls pod install with RCT_NEW_ARCH_ENABLED='0' even when the template defaults the New Architecture to on, which trips Reanimated's podspec assertion. The harness builder now prepends ENV['RCT_NEW_ARCH_ENABLED'] = '1' to the scaffolded Podfile so the env is corrected inside the same pod-install process, before any podspecs are evaluated.

vitest-mobile@0.4.2

Patch Changes

• 60e4368: Fix bootstrap failing on iOS for projects that depend on react-native-reanimated@4. The React Native community CLI's pod install step was passing RCT_NEW_ARCH_ENABLED='0' on a virgin scaffold because its New Architecture detector relies on a Pods xcodeproj that doesn't exist yet, which Reanimated's podspec asserts against. The harness builder now pre-seeds the file the detector reads, so pod install runs with the New Architecture enabled like the template expects.

vitest-mobile@0.4.1

Patch Changes

• 0124380: ### fix(harness): prevent bootstrap hang from npx install prompts

  Add --yes to every npx call in the harness builder so npx
  auto-confirms package downloads instead of prompting. Under the
  spinner, stdin is 'ignore' — any prompt would hang indefinitely:

  • npx --yes @react-native-community/cli init (scaffold step, runs
    before the project's node_modules exist, so npx always needs to
    resolve the package over the network)

  • npx --yes react-native build-ios and npx --yes react-native build-android (build steps; defensive — these usually resolve
    locally after npm install)

    While here, also switch buildAndroid from a raw gradlew assembleDebug to npx react-native build-android for symmetry with
    iOS and to let the RN CLI handle codegen + Gradle wrapper setup.
    Pass --tasks assembleDebug explicitly: the CLI defaults its task
    prefix to bundle (producing an .aab), not assemble (producing
    the .apk we need). The --help text is misleading on this point.

vitest-mobile@0.4.0

Minor Changes

• be23517: Fix component hot reload during pause() and improve build robustness.

  HMR fix: Editing a component file while paused previously triggered a full
  app reload instead of a live Fast Refresh update. The root cause was twofold:

  1. Components loaded with __ReactRefresh disabled had no family registrations,
     so performReactRefresh() couldn't map old→new component types.
  2. Restoring the real Refresh runtime during pause caused Metro to detect a
     boundary status change ("invalidated boundary") and call
     performFullRefresh().

  The fix installs a registration-only shim that builds component families from
  initial load (forwarding register()) while preventing implicit self-accept
  (isLikelyComponentType() → false). During pause, performFullRefresh() is
  suppressed and performReactRefresh() is triggered manually.

  Build improvements:

  • Use npx react-native build-ios / build-android instead of raw
    xcodebuild/gradlew — simplifies the build step and lets RN CLI handle
    pod install, gem setup, and gradle wrapper automatically.
  • Detect and recover from stale build locks (killed/interrupted builds no
    longer block subsequent runs indefinitely).
  • Clean incomplete binaries and DerivedData before rebuilding.
  • Add --verbose flag to all CLI commands (streams child-process output
    instead of using a spinner).

Patch Changes

• 2035ecc: Anchor the generated Metro config's vitest-stubs/ lookup at the active
  workspace's node_modules/vitest-mobile/ instead of the cached harness's.

  The cached harness's node_modules/vitest-mobile is installed via file:,
  so npm creates a symlink to whichever workspace first built the cache.
  Two workspaces with the same RN version + native modules + vitest-mobile
  version share a cache key, and the second one would hit Metro errors like

  Failed to get the SHA-1 for: <other-workspace>/node_modules/vitest-mobile/src/metro/vitest-stubs/empty.js.
    Potential causes:
      1) The file is not watched. Ensure it is under the configured `projectRoot` or `watchFolders`.
  

  — because the symlink target lives outside the second workspace's
  projectRoot and watchFolders, so Metro's file map doesn't track it.

  Resolving the stubs from projectRoot instead is safe: computeCacheKey
  already includes the vitest-mobile package version, so the workspace's
  stubs are guaranteed to match the harness's on every run. The fix is
  template-only — existing cached harness binaries continue to work
  unchanged, no rebuild required.

  Unblocks running tests from multiple checkouts of the same repo (and
  fixes CI tarball-restore scenarios where the originating workspace
  isn't present on the runner).

• a3eeb10: Default the device picker to an existing simulator/AVD instead of "Create new"

  The interactive device picker (used by bootstrap and boot-device) now
  defaults to an existing device rather than prompting to create a dedicated
  one. On iOS the pre-selected device is the most recently booted simulator,
  matching Expo CLI's heuristic; on Android it's the first available AVD.

  "Create new dedicated simulator/AVD" is still available at the bottom of
  the list for users who want isolation. The non-interactive (CI) fallback
  is unchanged — it still auto-creates a project-scoped device.

  This prevents vitest-mobile from stealing Expo CLI's default simulator:
  previously, creating and booting a VitestMobile-* sim made it macOS
  Simulator.app's "most recently used" device, so Expo would target it
  on the next expo start → i press.

vitest-mobile@0.3.1

Patch Changes

• 846d725: Pin install-expo-modules to 0.14.21 so the Expo autolinking step
  succeeds on Linux runners (Android-only CI).

  install-expo-modules@0.14.18 (the version currently published behind
  upstream's latest tag) ships without the process.platform === 'darwin'
  gate around its final pod install --repo-update step, so on Linux it
  crashes with ENOENT spawn pod and bootstrap aborts before
  patchAppDelegateForExpo can run. 0.14.21 (newest stable) has the gate
  restored. Switch from @latest to a pinned version so we pick up the
  fix regardless of which install-expo-modules version upstream tags as
  latest next.

vitest-mobile@0.3.0

Minor Changes

• d348a55: The harness builder now auto-wires Expo modules autolinking whenever a user
  declares any expo-* (or expo, or @expo/*) entry in
  nativePlugin({ harness: { nativeModules } }). This unblocks testing
  components from libraries that pull Expo modules under the hood — e.g.
  expo-blur, expo-haptics, expo-image, etc.

  Previously the harness was a vanilla React Native template (use_native_modules!
  only — no use_expo_modules!), so listing expo-blur in nativeModules got
  the JS dep installed but no native pod, and JS-side renders crashed with
  Cannot read property 'BlurView' of undefined because
  expo-modules-autolinking had never run.

  The builder now detects Expo-shaped names in nativeModules and runs
  npx install-expo-modules@latest --non-interactive against the scaffolded
  project to wire up the Podfile (use_expo_modules!), settings.gradle
  (useExpoModules()), MainApplication, and AppDelegate. Two
  post-processing patches keep the result compatible with vitest-mobile's
  own pipeline:

  1. The CLI integration's bundle-root rename (index →
     .expo/.virtual-metro-entry) is reverted, because vitest-mobile
     rewrites /index.bundle requests onto its prebuilt bundle directly
     and never consults Expo CLI's resolver.
  2. The missing bindReactNativeFactory(factory) call is inserted into
     AppDelegate.swift. SDK 54+'s ExpoAppDelegate.recreateRootView
     reads its own factory property and fatalErrors if it's unset;
     install-expo-modules's Swift transform doesn't add the bind call,
     but the from-scratch Expo bare template does.

  Cache key bumps to fmt6 so users with v5 binaries that listed Expo modules
  as deps (no autolinking pipeline) get a fresh build the next time they
  bootstrap.

  Heuristic: a nativeModules entry triggers the Expo wiring when its name
  matches expo, expo-*, or @expo/*. Modules outside that pattern still
  go through the React Native community CLI's autolinking, which the
  scaffolded RN template already supports.

vitest-mobile@0.2.2

Patch Changes

• 320c60b: Fix vitest-mobile bundle (and any other CLI command that statically reads the
  metro customizer from a vitest config) to look at metro.customize on the
  plugin options.

  readMetroCustomizerFromConfig was inspecting stored.metro as if it were the
  customizer function itself, but nativePlugin({ metro }) stashes a
  MetroOptions object ({ bundle, customize, babelPlugins }) on the plugin
  instance — so the customizer was never picked up and the bundle was built with
  only the harness-anchored base resolver. Any user resolver hook (e.g. monorepo
  #src/* rewrites or react-native condition pinning) silently dropped on the
  floor in pre-built bundles, while the in-process Vitest pool path was unaffected
  because it reads options.metro.customize directly.

  While there: collapse the three plugin-options readers
  (readNativeModulesFromConfig, readMetroCustomizerFromConfig,
  readBabelPluginsFromConfig) onto a single typed extractor
  (readVitestMobilePluginOptions) that returns NativePluginOptions[] for the
  matching projects. Each per-field reader is now a tiny pluck function over the
  shared extractor — so a future change to the plugin-options shape fails the
  type-checker in one place instead of silently dropping options in three. The
  extractor also normalizes the legacy top-level nativeModules field into
  harness.nativeModules so consumers stay strictly typed.

  Internal: also extract the readers from cli/index.ts into a new internal
  cli/config-readers.ts module (no public API change) so they can be
  unit-tested without dragging the cac dispatcher in. Adds regression tests
  covering all readers, including the metro-customizer bug fix and the legacy
  nativeModules compat path.

vitest-mobile@0.2.1

Patch Changes

• 66f4239: Add metro.babelPlugins option to inject extra Babel plugins into Metro's transform pipeline.

  Native modules like react-native-reanimated require compile-time Babel transforms (e.g. worklet directives) that Metro won't apply unless the plugin is explicitly wired in. Previously, users had no way to add these — worklet transforms were silently skipped in both watch mode and pre-built bundles.

  New option: metro.babelPlugins.

  nativePlugin({
    harness: { nativeModules: ['react-native-reanimated'] },
    metro: { babelPlugins: ['react-native-reanimated/plugin'] },
  });

  Plugins are resolved from the harness project's node_modules and injected into the generated Metro transformer shim. They run before vitest-mobile's own plugins so worklet transforms etc. are applied before the test wrapper inspects the output. Works in both live Metro (watch mode) and bundle pre-builds.

  Auto-injection for known modules. When a native module listed in harness.nativeModules has a well-known companion Babel plugin (currently just react-native-reanimated → react-native-reanimated/plugin), the harness builder automatically adds it to babel.config.js during bootstrap.

  CLI plumbing. The bundle command now reads metro.babelPlugins from the vitest config and passes them through to the bundler, so pre-built bundles match live-Metro output.

vitest-mobile@0.2.0

Minor Changes

• fd223b8: Restructure plugin options and rewrite tye device-side runtime.

  Breaking: plugin options

  nativePlugin(...) options are now nested into harness / device / metro groups:

  // Before
  nativePlugin({ platform: 'ios', headless: true, bundle: true, metro: customizerFn });
  
  // After
  nativePlugin({ platform: 'ios', device: { headless: true }, metro: { bundle: true, customize: customizerFn } });

  Removed promptForNewDevice and skipIfUnavailable. Missing environments now throw instead of soft-skipping.

  Runtime rewrite

  • setup.ts, ControlBridge, TestRegistry, TestRunnerService, run.ts, state.ts collapsed into a single HarnessRuntime root that owns connection, state, registry diff, and RPC forwarding.
  • Generated test-registry module replaced by require.context in test-context.ts (backed by new inline-app-root-plugin).
  • Explorer UI walks Vitest's native task tree directly; old TestTreeNode parallel tree deleted.
  • ReactNativeRunner simplified to (config, runtime) — delegates onCollected/onTaskUpdate to the runtime.
  • Task state (collectedFiles, taskState) lives on the runtime instance; entries are append-only across reruns.

  Other

  • New vitest-compat-plugin.ts rewrites Vitest dist for Hermes/Metro. Hermes-safe node:* stubs added.
  • Pool-side: PauseController, Metro log tailing, bundle server, and pool message types extracted into dedicated modules.

• 1434ccd: Per-project device ownership, interactive device picker, and a big CLI UX polish pass.

  Project-scoped devices (iOS + Android).

  • Each project now owns a specific simulator / AVD, stored in ~/.cache/vitest-mobile/devices.json keyed by project path. Running Expo, Android Studio, or another vitest-mobile project on the same machine no longer collides with your tests.
  • vitest-mobile bootstrap always shows an interactive picker: your existing simulators/AVDs plus a "Create new dedicated device" option. Current mapping is pre-selected so hitting Enter keeps your choice.
  • --device <name> skips the picker (for CI and scripts). Non-TTY bootstrap auto-creates the project-scoped device.
  • On Android, "Create new" is annotated as unavailable and refused unless the Android cmdline-tools (sdkmanager + avdmanager) are installed — with a pointer to install them or pick an existing AVD instead.
  • reset-device now respects whether vitest-mobile created the device: deletes + clears the mapping if we created it, only clears the mapping if the user picked their own device.
  • iOS: existing VitestMobile-<hash> simulators are auto-registered into the mapping on first run (no re-prompt on upgrade).
  • Concurrent runs on the same project get a per-instance secondary simulator (iOS) / emulator instance (Android).

  CLI UX polish.

  • Consistent --platform flag (breaking). Every command takes --platform <ios|android> (previously some were positional, some --platform). The old vitest-mobile build ios form prints a clear migration hint and exits non-zero. Most commands prompt (TTY) or error (non-TTY) when --platform is omitted; trim-cache / clean-devices / bundle default to both platforms; cache-key still requires an explicit platform.
  • Spinners with live step messages for build, install, bootstrap, bundle, boot-device — instead of a wall of xcodebuild / gradle / pod-install output. Spinner now animates during long builds (the underlying spawn is async; previously a sync execSync blocked the event loop).
  • Child-process output is tee'd to ~/.cache/vitest-mobile/logs/<timestamp>-<command>-<platform>.log. On failure the log path is printed; nothing is silently swallowed.
  • Ctrl+C now works during long builds (same async-spawn fix — SIGINT previously went to the blocked child instead of Node).
  • Unknown commands exit 1 with a help dump (previously exited 0 silently).

  Metro + native modules config.

  • New nativePlugin({ metro }) customizer and exported MetroConfigCustomizer type. Layers on top of the auto-generated harness-anchored base Metro config (runs before internal test transforms, so the vitest shim and test-registry stay authoritative). Composes across multiple plugin instances.
  • build, install, bootstrap, and bundle CLIs read nativeModules from the vitest config automatically; --native-modules overrides when passed.

  Setup diagnostics.

  • Simulator creation failures now report the real cause — unaccepted Xcode license, xcode-select pointing at Command Line Tools, missing iOS runtime. Preflight checks before xcodebuild catch SDK/runtime mismatch upfront.

  Build system.

  • Xcode 26.4 compat: bundled fmt pod pinned to C++17 so RCT-Folly compiles under Apple Clang 21. Harness build format bumped to v5 — existing cached binaries rebuild once.

  Fixes.

  • promptConfirm no longer leaks a stdin resume that kept bootstrap alive after accepting the prompt.

Patch Changes

• 01ce6fe: Collapse the pool worker lifecycle to fire once per user-initiated run instead of once per file, by setting test.isolate = false in the plugin's config hook.

  Why. With isolate: true (the previous default), Vitest's scheduler created one PoolRunner per test file, meaning worker.start() and worker.stop() — and the 60s / 90s handshake timeouts guarding them — fired N times per run. The React Native harness shares a single JS VM across files anyway, so the per-file isolation was a fiction maintained by singleton idempotency flags. Under isolate: false + maxWorkers: 1, Vitest bundles every file into one task with context.files = [all], and the handshake timers fire exactly once per user-initiated run (initial or HMR-driven rerun). Timer scope now matches reality.

  Changes.

  • test.isolate = false is applied by the plugin, guarded so user-level overrides win.
  • canReuse: () => true added to the pool worker.
  • Device-side handleRun refactored into a per-file loop so explorer file-start/file-done UI events still fire per file; test execution inside startTests is unchanged.
  • Removed dead code: the fallback __rerun replay path (Vitest ^4's rerunFiles is guaranteed), the _lastRunMessages map, _sessionCount, countTestsInSpecs, and the triple-keying of file identifiers. Reporter shrunk to a pass-through for __native_run_start / __native_run_end plus the run-mode teardown() await.

  No user-visible behavior change. Custom Vitest pools, test configs that explicitly set test.isolate, and the device-side test execution path are all unaffected.

vitest-mobile@0.1.1

Patch Changes

• 5c06893: Includes missing files in publish