Add android e2e setup + simulator run iOS e2e

[dewabisma]

Summary

Added android emu and ios simu e2e test scripts.

The commit history shown here is buggy because I extend from other PR that already merged to main then I pull main to this branch. But the files changed should be correct. We can ignore and it will resolve it properly. Like how sometimes we have it before.

---

Note

Low Risk
Changes are limited to test automation (Gradle test runner settings, androidTest stub, and dev scripts); production app behavior is unaffected aside from standard instrumentation test dependencies.

Overview
Adds Patrol-native Android E2E wiring so patrol_test/ Dart tests run as instrumentation tests: PatrolJUnitRunner, clearPackageData between runs, AndroidX Test Orchestrator, and a parameterized MainActivityTest that lists and runs each Dart test.

Introduces patrol_android_emulator.sh and patrol_ios_simulator.sh for local runs—auto-pick or boot an emulator/simulator, optional single-file or full-suite targets, debug/release builds, and test secrets via .env.test or TEST_IMPORT_MNEMONIC (--dart-define, not bundled assets). Both invoke patrol test in one step (unlike the existing physical-iOS device script).

^{Reviewed by Cursor Bugbot for commit 7c38e21. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 7c38e21. Configure here.}

[cursor]

Wrong simulator runtime selected

Medium Severity

When several runtimes define the same simulator name, simulator_udid_for_name uses tail -1 on matching lines, which picks the last runtime block in simctl output. That usually maps to an older iOS runtime, not the newest one the comment describes, so -s can boot the wrong UDID and E2E runs on an unintended OS version.

 

^{Reviewed by Cursor Bugbot for commit 7c38e21. Configure here.}

[cursor]

Reused emulator skips boot wait

Medium Severity

If a running emulator is auto-selected or chosen with -d, the script never calls wait_for_emulator_boot, unlike the path that starts a new AVD. Patrol can start while sys.boot_completed is still unset, leading to flaky or failed instrumentation runs.

 

^{Reviewed by Cursor Bugbot for commit 7c38e21. Configure here.}

[n13]

Review

Solid, well-documented E2E addition. The Android native Patrol wiring is correct (PatrolJUnitRunner + clearPackageData + AndroidX Orchestrator + parameterized MainActivityTest), package placement matches applicationId/patrol android.package_name (com.quantus.wallet), MainActivity import from the namespace package is right, .env.test is already gitignored, and the empty-array expansions correctly guard set -u on macOS Bash 3.2.

Requesting changes for two confirmed correctness bugs + a DRY concern before merge.

Correctness (agree with Bugbot)

1. Android reused emulator skips boot wait (patrol_android_emulator.sh). wait_for_emulator_boot only runs when a fresh AVD is booted. When an emulator is auto-selected or passed with -d, it's skipped. adb devices reports device before sys.boot_completed=1, so Patrol can start against a not-fully-booted emulator → flaky runs. Fix (also a DRY win): call wait_for_emulator_boot "$DEVICE_SERIAL" once after the serial is resolved, on every path.

2. iOS wrong runtime selected (patrol_ios_simulator.sh). simulator_udid_for_name uses tail -1, but the comment claims "prefer the newest runtime" — simctl list devices section ordering isn't guaranteed, so -s can boot the wrong OS version (same non-determinism in default_simulator_name). More robust: parse xcrun simctl list devices available -j and pick the highest runtime version.

DRY (repo rule)

The three patrol scripts now triplicate large verbatim blocks: the .env.test/TEST_IMPORT_MNEMONIC → DART_DEFINES block (~16 lines), the TARGET_ARGS builder, SCRIPT_DIR/APP_ROOT resolution, and the common arg-parse loop. Recommend extracting scripts/lib/patrol_common.sh (e.g. patrol_resolve_app_root, patrol_collect_dart_defines, patrol_build_target_args) sourced by all three scripts.

Minor nits

• Stale comment in MainActivityTest.java: "...run/filtered individually by xcodebuild's Android equivalent (the orchestrator)" — "xcodebuild" is an iOS copy-paste leftover.
• patrol_ios_simulator.sh passes PATROL_DEVICE="${DEVICE_LABEL:-$DEVICE_ID}", preferring the possibly-ambiguous name over the already-resolved UDID; passing $DEVICE_ID is unambiguous when duplicate names exist.

Verdict: Request changes — prioritize the Android boot-wait fix and the DRY extraction; the iOS runtime selection and nits are quick follow-ups.

[n13]

Re-review (commits 2456ccf → 4c0f52b)

The three items from the previous review were all addressed — thank you. Unfortunately the DRY refactor introduced a critical regression that breaks every platform-specific flag in all three scripts, so requesting changes one more time. Reviewed in a clean worktree with the bugs reproduced by actually running the scripts.

Critical: platform option callbacks run in a subshell — -d, -a, -s are all silently ignored

In lib/patrol_common.sh, patrol_parse_runner_args invokes the platform callback via command substitution:

consumed="$("$platform_option_fn" "$@")" || consumed=0

Command substitution runs the callback in a subshell, so assignments like DEVICE_SERIAL=…, AVD_NAME=…, DEVICE_ID=…, SIMULATOR_NAME=…, DEVICE_UDID=… are lost when the subshell exits. The echo 2 consumed-count still reaches the parent, so the flag and its value are shifted away without error — a textbook silent failure. Reproduced on all three scripts:

• patrol_android_emulator.sh -d emulator-9999 → serial discarded, falls into auto-select. With another emulator running it would silently test the wrong device; with none it errors with "No running emulator found" despite an explicit serial.
• patrol_ios_simulator.sh -d <udid> → UDID discarded; the script booted a different simulator (the default-name fallback) and would have run tests on it.
• patrol_ios_device.sh -d <udid> → UDID discarded → "ERROR: No iOS device found. Connect a device or pass its UDID."
• -s "iPhone 16 Pro" and -a <avd> are equally ignored.

The pre-refactor inline parsers (7c38e21) handled these flags correctly, so this is a pure regression from 2456ccf.

Suggested fix: don't use command substitution for the callback. Call it directly and pass the consumed count through a global, e.g.:

# patrol_common.sh
if [[ "$platform_option_fn" != "-" ]] && "$platform_option_fn" "$@"; then
  shift "$PLATFORM_CONSUMED"
  continue
fi

# callers
patrol_android_emulator_platform_option() {
  case "$1" in
    -d|--device) DEVICE_SERIAL="${2:?--device requires a serial}"; PLATFORM_CONSUMED=2; return 0 ;;
    -a|--avd)    AVD_NAME="${2:?--avd requires a name}";           PLATFORM_CONSUMED=2; return 0 ;;
  esac
  return 1
}

Please add a quick smoke test of each flag before pushing (-d, -a, -s, --release, a target file) — none of the flag paths currently work.

Bug: first_booted_simulator breaks on parenthesized simulator names

The sed in patrol_ios_simulator.sh:

sed -E 's/^[[:space:]]+([^(]+) \(([A-F0-9-]+)\).*/\1\t\2/'

[^(]+ cannot cross a (, so for stock names like iPhone SE (3rd generation) (UDID) (Booted) the regex never reaches the UDID group, sed passes the line through unmodified, and DEVICE_LABEL/DEVICE_ID both become the raw untrimmed line — which is then handed to patrol test --device. Verified with a synthetic line: plain names produce name\tudid, parenthesized names produce garbage. Since the auto-select path is currently the only working path (see bug above), anyone with an SE-class simulator booted hits this. Note simulator_udid_for_name doesn't have this problem (its awk uses literal index() and matches the UDID pattern directly) — consider reusing that approach for the booted case too.

Previous review items — confirmed fixed

• ✅ Android boot wait on every path — wait_for_emulator_boot "$DEVICE_SERIAL" now runs once after the serial is resolved, covering auto-select, -d, and fresh-boot paths.
• ✅ iOS newest-runtime selection — the awk version-key approach in simulator_udid_for_name / default_simulator_name correctly parses the -- iOS X.Y -- section headers and picks the highest version; verified the logic against real simctl output shape. (JSON would still be more robust, but this works.)
• ✅ DRY extraction — lib/patrol_common.sh with patrol_resolve_app_root / patrol_collect_dart_defines / patrol_build_target_args is exactly the right structure; the shared-flag parsing is the right idea too, it just needs the subshell fix above.

Also re-verified: Gradle wiring (PatrolJUnitRunner + clearPackageData + orchestrator 1.5.1) is standard and correct, MainActivityTest package placement matches applicationId/patrol package_name (com.quantus.wallet) with MainActivity imported from the namespace package, .env.test is gitignored, and the ${ARR[@]+"${ARR[@]}"} expansions behave under macOS Bash 3.2 set -u (tested).

Nits (carried over, still open)

• MainActivityTest.java still has the iOS copy-paste leftover: "run/filtered individually by xcodebuild's Android equivalent (the orchestrator)".
• PATROL_DEVICE="${DEVICE_LABEL:-$DEVICE_ID}" still prefers the possibly-ambiguous name over the resolved UDID.
• Tiny: in simulator_udid_for_name, ver is never reset when a non-iOS section header (tvOS/watchOS) starts; harmless for iPhone names but worth a guard.

Verdict: request changes — the flag-parsing regression must be fixed (and the flags actually exercised) before merge; the sed name bug should go in the same pass.

[n13]

Re-review (commits ebdc0c2 → 2e563c2)

Both blocking issues from the last review are fixed. Verified by executing all three scripts against stubbed adb/emulator/xcrun/patrol/xcodebuild binaries under macOS stock Bash 3.2, exercising every flag path.

✅ Flag-parsing regression fixed

The platform callback is now invoked directly instead of via command substitution, with the consumed count passed through PLATFORM_CONSUMED — so DEVICE_SERIAL/AVD_NAME/DEVICE_ID/SIMULATOR_NAME/DEVICE_UDID assignments persist in the parent shell. Confirmed by running:

• patrol_android_emulator.sh -d emulator-9999 (with a different emulator running) → tests correctly target emulator-9999, and the boot wait runs on that path too. -a My_Custom_AVD boots the named AVD; --release and positional targets are consumed correctly; the auto-select path still works.
• patrol_ios_simulator.sh -d <udid> → the UDID reaches patrol test --device <udid> unchanged. -s "iPhone 16 Pro" resolves the newest runtime when two runtimes have the same device name (picked the iOS 26.5 UDID over 18.2 in my fixture).
• patrol_ios_device.sh -d <udid> → flows through to xcodebuild -destination "platform=iOS,id=<udid>"; --release is correctly rejected there since that script doesn't take build-mode flags.

✅ first_booted_simulator name parsing fixed

The awk match(/\([A-F0-9-]+\)/) approach handles parenthesized device names: a booted iPhone SE (3rd generation) (<udid>) (Booted) now yields the correct name+UDID pair, and -s "iPhone SE (3rd generation)" resolves correctly too (both verified against synthetic simctl output).

Footnote, not blocking: awk takes the first regex match, so a custom simulator name containing a parenthesized run of only [A-F0-9-] characters (e.g. Test (ABC)) would mis-parse. All stock Apple names are safe — "(3rd generation)" contains lowercase — so this is theoretical.

Nits carried over (non-blocking, fine as a follow-up)

• MainActivityTest.java still has the "xcodebuild's Android equivalent" copy-paste leftover in the class comment.
• PATROL_DEVICE="${DEVICE_LABEL:-$DEVICE_ID}" still prefers the possibly-ambiguous name over the resolved UDID on the auto-select and -s paths (the -d path correctly uses the UDID).
• ver in simulator_udid_for_name isn't reset when a non-iOS runtime section (tvOS/watchOS) starts; harmless for iPhone names.

Verdict: Approve. The two blockers are properly fixed, all documented flag paths now work as advertised, and the shared patrol_common.sh structure is clean.

[n13]

Re-review (head still 2e563c2)

Fresh independent verification pass — head commit is unchanged since the previous approval, so this re-confirms that verdict from scratch rather than reviewing new code. All three scripts were re-executed against stubbed adb/emulator/xcrun/patrol/idevice_id/xcodebuild binaries under macOS stock Bash 3.2 (3.2.57), exercising every documented path.

Verified by execution

patrol_android_emulator.sh

• Auto-select running emulator → boot wait → patrol test --device emulator-5554 --debug ✅
• -d emulator-9999 with a different emulator running → targets emulator-9999, boot wait included ✅
• -a My_Custom_AVD with nothing running → boots exactly that AVD ✅; with no -a/env → falls back to first listed AVD ✅
• --release + positional target → --release -t <file> reaches patrol ✅; -- separator with multiple targets → each gets its own -t ✅; unknown flag → error + usage, exit 1 ✅

patrol_ios_simulator.sh

• Booted iPhone SE (3rd generation) (<udid>) (Booted) auto-selected with correct name/UDID split (the parenthesized-name fix holds) ✅
• -d <udid> → UDID passed through unchanged ✅
• -s "iPhone 16 Pro" with the name present under both iOS 18.2 and 26.5 → boots the 26.5 UDID (newest runtime) ✅
• No booted sim, no -s → default_simulator_name picks an iPhone from the newest runtime section and correctly stops at the following -- tvOS -- header ✅
• -s "iPhone 99" → clean error with the simctl list hint, exit 1 ✅

patrol_ios_device.sh

• -d <udid> persists through the shared parser (PLATFORM_CONSUMED mechanism) ✅; --release correctly rejected since that script takes no build-mode flag ✅

Static re-checks

• Gradle wiring (PatrolJUnitRunner + clearPackageData + orchestrator 1.5.1) matches Patrol's documented Android setup; MainActivityTest package com.quantus.wallet matches applicationId/patrol android.package_name, with MainActivity imported from the namespace package.
• CI is green (Analyze passed on 2e563c2).
• shellcheck: only false-positive SC2034s (cross-file globals) and informational notes; nothing actionable.

Nits carried over (non-blocking, same as before)

• MainActivityTest.java: "xcodebuild's Android equivalent" copy-paste leftover in the class comment.
• PATROL_DEVICE="${DEVICE_LABEL:-$DEVICE_ID}" passes the (possibly ambiguous) name to patrol test --device on the auto-select and -s paths; the -d path uses the UDID.
• ver in simulator_udid_for_name isn't reset at non-iOS runtime headers; harmless for iPhone names.

Verdict: Approve. Re-verified end to end — every flag path behaves as documented, both previously-fixed blockers stay fixed, and no new issues surfaced.