* fix(runner): surface focus-tap errors in InputText action
A failed Tap/TapSelector before InputText was swallowed, so text typed
into the wrong field (or no field) still reported success. Return the
error so the step fails explicitly.
* feat(sample-app): add username EditText and snapshot
Gives the spec a real EditText target (content-desc: username_field)
so the InputText action path can be exercised end-to-end. The typed
value is mirrored into MainActivity.username and surfaced as the
"username" snapshot for spec assertions.
* feat(sample-app): exercise InputText action against username field
Adds typeUsername action and usernameNeverShrinks property to the
sample spec, and extends the integration test to assert the bundled
spec emits an InputText(desc:username_field, "alice") action and that
the property correctly violates when a snapshot reports a shorter
string.
interestingTags hardcoded selectors from a specific app (etMobileNumber,
customer_row_, supplier_row_, etc.) inside the generic runner. None of
these selectors exist in the checked-in sample spec. Debug log now just
reports screen + hierarchy size; specs that want richer visibility can
log from state.ax.find themselves.
* refactor(docs): inline pandoc build into Makefile, drop scripts dir
* fix(agent): wait for deadline watcher before returning
readWithDeadline's watcher goroutine could clobber the conn's read
deadline with time.Unix(1, 0) after the main function reset it to zero.
When the Accept ctx was canceled shortly after Accept returned, the
watcher raced with close(done) in select and sometimes picked ctx.Done()
even though we were already done reading, leaving the conn unusable for
the next read (instant i/o timeout on step 1 snapshot).
Synchronize on the watcher's exit before resetting the deadline so it
can never override the reset.
* test(agent): cover readWithDeadline race on Accept ctx cancel
Drives Accept with a short-timeout ctx, cancels it right after Accept
returns, then does a Snapshot. Reliably fails without the readWithDeadline
synchronization fix (watcher goroutine overwrites the deadline to past).
The old tests loaded merchant-android uiautomator dumps from /tmp and
skipped when absent. Replace with two hermetic tests that bundle
examples/sample-app/spec.ts against a synthetic hierarchy: one checks
tapClickMe fires, the other drives the three properties through a
holds/holds/violated snapshot sequence.
Introduce examples/sample-app/spec.ts — a minimal property-based spec that
taps the sample app's "Click me" button and asserts click_count is
monotonic. Bundle-check and the trace writer test now reference the new
path.
make uatu copies the real fat JAR into assets/ before
go build -tags withsidecar. Keeping that path tracked was
the root cause of the 130 MB push rejection.
Splits embed.go so the go:embed directive only fires under
-tags withsidecar. Default builds get a stub with a nil JAR
and IsPlaceholder()=true. This removes the landmine where
make uatu overwrote a tracked placeholder file, making any
git add silently stage 130 MB.
Removes Launch + Terminate from runner.Run so the CLI can launch
the app first, wait for the SDK to connect, then start the loop.
The previous shape forced runner to launch internally which fought
with the SDK-must-be-connected-first ordering.
BundleID/ClearState fields go away too since runner no longer
launches; the CLI keeps them on its testOptions struct.
Ships with a 24-byte placeholder so fresh clones build without
requiring a sidecar build first. `make uatu` copies the real
fat JAR into internal/sidecar/assets before `go build`, so
shipping binaries carry the full sidecar (~130 MB).
Extract writes the JAR to a temp dir alongside a SHA-256 file
and skips rewrite when the checksum already matches.
Wraps each v0.1 RPC, plus a WaitForHealth helper that polls until
the sidecar reports Ready=true (used at startup before any other
calls happen). Tests stub the gRPC server in-process so they don't
need a real sidecar JAR.
Inspector parses uses-permission entries from `aapt dump permissions`
and the granter shells out to `adb shell pm grant`. Both are pluggable
so tests can drive logic without aapt or a device. Granter failures
become warnings rather than errors — Android refuses non-runtime
permissions and we'd rather keep going than abort the run.
Wires agent.Conn + driver.Driver + verifier.Verifier + trace.Writer
into the v0.1 step cycle: snapshot the SDK, push to verifier,
evaluate properties, write the trace step (with violations), release
the SDK pause, apply the next action via the driver, wait for idle.
Driver.Launch happens once before the loop and Terminate runs in
defer so even an early error tears down the app cleanly. Summary
returns step count and per-step violation records for the caller
to print or persist.
Real maestro driver will resolve selectors itself rather than
forcing the runner to look up coordinates from the hierarchy. Mock
now waits the requested duration so tests don't busy-spin and
starve the SDK fixture goroutine.
Installs globalThis.__uatu__ with extract, always, actions,
weighted, tap, inputText, and stub taps/swipes. Load runs the
bundled spec, then pulls properties + actions out of globalThis.
PushSnapshot rebuilds state.snapshots and refreshes every
extractor handle's current/previous in registration order so
chained extractors observe up-to-date values.
Properties are wired through internal/ltl as Always(Thunk(...)),
so verdicts latch to violated as soon as a predicate returns
false. NextAction resolves actions/weighted recursively with a
seedable rand source for reproducible runs.
Each WriteStep appends one JSON object per line so the trace can
be filtered with jq directly. Screenshots land under screenshots/
with zero-padded indexes. Writer is concurrency-safe; Close is
idempotent and a write after Close errors loudly rather than
silently dropping.
Supports Always over Pure/Thunk leaves; eventually/next/bounds
deferred. Once a thunk returns false under an Always, the verdict
latches to violated so the runner can surface the offending step
without later observations masking it.
Bundles a TypeScript entry into an IIFE ES2020 blob with optional
inline sourcemaps. process.env defines are JSON-quoted before being
fed to esbuild so values with quotes/backslashes survive correctly.
Returns the bundle's SHA-256 for trace meta.json.
Records every call as an Action and lets tests program hierarchy,
screenshot, health, and per-method failures. Actions() returns a
copy so test code can't mutate the driver's history.
Accept waits for an SDK HELLO then hands back a Conn. Conn.Snapshot
sends a PAUSE, blocks on the matching STATE (id-correlated), and
leaves the SDK paused until Release sends RESUME. Conn.Close sends
GOODBYE best-effort.
v0.1 supports one client at a time; transport is left to the caller
so tests can use TCP loopback while production wires via adb reverse
to localabstract:uatu-agent.
pj