Files
sanderling/examples/sample-app
pj 69002b07b1 fix(runner,sample-app): surface silent errors and demo a failing property (#15)
* fix(runner): surface non-deadline WaitForIdle errors

Previously the WaitForIdle return value was discarded entirely, hiding
real driver failures (gRPC transport errors, sidecar crashes) behind
the expected deadline-exceeded case. Log non-deadline errors so they
are visible without changing control flow.

* chore(sample-app): drop unused uptime_millis extractor

Registered in SampleApplication but never consumed by spec.ts.

* fix(sample-app): drop trivial appIsRunning property

app_state was hardcoded to 'running' so the property was a tautology
that could never fail. Removing both the extractor and the property
is the simplest fix; demo-grade properties that can fail land next.

* feat(sample-app): add Reset button that zeroes clickCount

Pairs with the next commit's tap-reset action so the fuzzer can
violate clickCountNeverDecreases and demonstrate uatu actually
finding a property violation.

* feat(sample-app): add tap-reset action to exercise Reset button

Weighted at 10/122, fuzzer reaches it within a short run. Pairs with
the Reset button to demonstrate uatu detecting the
clickCountNeverDecreases violation.

* fix(runner): filter WaitForIdle errors via context state, not errors.Is

errors.Is(err, context.DeadlineExceeded) misses gRPC's wrapped
status.DeadlineExceeded, so every step under the maestro driver
logged a spurious warning. Check idleCtx.Err() instead — captures
both deadline-fired and parent-canceled cases regardless of how the
driver wraps them.

* chore(sample-app): tune action weights so demo violates in ~30s

Prior weights left tap-reset rare enough that short demo runs missed
the violation by chance. Bumped to 30/107, with typeUsername reduced
since username noise doesn't help exercise clickCount.

* refactor(runner): route warnings through slog

Adds Options.Logger (defaults to slog.Default()) and converts the
three warning sites that were using fmt.Printf. Progress line stays
on Printf since it's user-facing UI, not a log. Makes the warnings
testable via a capturing handler.

* test(runner): assert WaitForIdle driver errors are logged

Captures slog output via TextHandler into a buffer and asserts the
warning message + injected error text appear when the mock driver
returns a non-context error from WaitForIdle. Guards against a
regression of the silent-error swallow.
2026-04-18 18:48:25 +07:00
..

uatu sample app

Tiny Android app wired to the uatu SDK, plus a TypeScript spec that drives it. Use it as a reference for integrating uatu into your own app.

Prerequisites

  • uatu CLI on PATH (see getting started)
  • Android SDK installed (uatu auto-discovers adb and emulator under $ANDROID_HOME, ~/Library/Android/sdk, or the Homebrew cask; nothing to export if you use a standard install)
  • just task runner

Install the app

just install

Run a test

just test

If no device is connected, uatu boots the single AVD it finds. With multiple AVDs, pick one:

just AVD=Pixel_7 test

Persistent settings can live in .env alongside the justfile:

AVD=Pixel_7
DURATION=5m

Traces land in ./runs/<timestamp>/.

How the pieces connect

  • android/build.gradle.kts depends on io.github.priyanshujain:sdk-android from Maven Central
  • android/src/main/kotlin/.../SampleApplication.kt calls Uatu.start(this) and registers snapshot extractors (app_state, click_count)
  • spec.ts imports @uatu/spec (see package.json), reads those snapshots, asserts properties on them, and weights the actions the fuzzer picks from
  • just test invokes uatu test against the installed APK on the connected device (or the AVD named via AVD=)