Add regression cross-reference with automated fix-PR test execution

.github/agents/maui-expert-reviewer.md

@@ -111,6 +111,7 @@ Every bug fix needs a regression test. Modified code must be checked against git
 - CHECK: Test covers the specific scenario from the issue report, not a generic case
 - CHECK: Shared code changes are tested on all affected platforms
 - CHECK: Previously-fixed issue numbers are cross-referenced when modifying the same code area
+- CHECK: If `regression-check/risks.json` exists and contains `REVERT` entries, list the affected fix PRs/issues and require author acknowledgment that the reverted fix is intentional. The regression cross-reference script (`Find-RegressionRisks.ps1`) detects when a PR deletes lines that were previously added by a labeled bug-fix PR.
 - CHECK: UI tests run on all applicable platforms unless there is a specific technical limitation
 - CHECK: Snapshot baselines updated across all platforms when changing background color, font, or layout
 - CHECK: Screenshot size matches capture method — a size mismatch means the capture changed, not the rendering

.github/scripts/BuildAndRunHostApp.ps1

@@ -219,9 +219,13 @@ Write-Success "Test project: $TestProject"
 
 #region Run Tests
 
-# Determine the filter to use
+# Determine the filter to use.
+# NOTE: The CI pipeline `maui-pr-uitests` (definition 313) uses `TestCategory=`
+# (see eng/pipelines/common/ui-tests-steps.yml lines 116-164). NUnit accepts
+# both `Category=` and `TestCategory=` but Cake's RunTestWithLocalDotNet uses
+# `TestCategory=` so we mirror that here for byte-for-byte parity with CI.
 if ($Category) {
-    $effectiveFilter = "Category=$Category"
+    $effectiveFilter = "TestCategory=$Category"
     Write-Step "Running UI tests with category: $Category"
 } else {
     $effectiveFilter = $TestFilter
@@ -233,27 +237,30 @@ if ($Platform -eq "android") {
     Write-Info "Clearing Android logcat buffer before test..."
     & adb -s $DeviceUdid logcat -c
 
-    # Dismiss any ANR dialogs that may have appeared during build/deploy.
-    # The emulator can sit idle during long builds, causing SystemUI ANR.
-    Write-Info "Dismissing any system dialogs before test..."
+    # Wait for Android settings service to be available.
+    Write-Info "Waiting for Android settings service..."
+    $settingsReady = $false
+    for ($i = 0; $i -lt 30; $i++) {
+        $settingsCheck = & adb -s $DeviceUdid shell settings get global device_name 2>&1
+        if ($settingsCheck -and $settingsCheck -notmatch "Can't find service|error") {
+            $settingsReady = $true
+            Write-Success "Settings service ready (device_name=$settingsCheck)"
+            break
+        }
+        Write-Info "  Settings service not ready yet (attempt $($i+1)/30)..."
+        Start-Sleep -Seconds 5
+    }
+    if (-not $settingsReady) {
+        Write-Warn "Settings service may not be ready — tests might fail"
+    }
+
+    # Do NOT force-stop or restart the app here. Appium's UiAutomator2
+    # driver handles app lifecycle via appPackage/appActivity capabilities.
+    # Manual restart causes double-stop issues and the app ends up in a
+    # bad state. Just dismiss any system dialogs and let Appium handle it.
     & adb -s $DeviceUdid shell am broadcast -a android.intent.action.CLOSE_SYSTEM_DIALOGS 2>$null
-    & adb -s $DeviceUdid shell input keyevent KEYCODE_ENTER 2>$null
-    & adb -s $DeviceUdid shell input keyevent KEYCODE_BACK 2>$null
-    Start-Sleep -Seconds 1
     & adb -s $DeviceUdid shell input keyevent KEYCODE_WAKEUP 2>$null
-    & adb -s $DeviceUdid shell input keyevent KEYCODE_MENU 2>$null
     Start-Sleep -Seconds 1
-
-    # Check for lingering ANR dialogs via window dump
-    $windowDump = & adb -s $DeviceUdid shell dumpsys window 2>$null | Select-String "Application Not Responding|ANR"
-    if ($windowDump) {
-        Write-Warn "ANR dialog detected — force-dismissing..."
-        & adb -s $DeviceUdid shell input keyevent KEYCODE_HOME 2>$null
-        Start-Sleep -Seconds 2
-        & adb -s $DeviceUdid shell am broadcast -a android.intent.action.CLOSE_SYSTEM_DIALOGS 2>$null
-        & adb -s $DeviceUdid shell input keyevent KEYCODE_BACK 2>$null
-        Start-Sleep -Seconds 1
-    }
 }
 
 # Capture test start time for iOS logs
@@ -306,17 +313,71 @@ $appiumLogFile = Join-Path $HostAppLogsDir "appium.log"
 $env:APPIUM_LOG_FILE = $appiumLogFile
 Write-Info "Set APPIUM_LOG_FILE: $appiumLogFile (screenshots will be saved here)"
 
+# ── TRX setup (mirrors CI: eng/cake/dotnet.cake `RunTestWithLocalDotNet`) ──
+# CI writes one trx per test run via:
+#   --logger "trx;LogFileName=<sanitized-name>.trx"
+#   --logger "console;verbosity=normal"
+#   --results-directory <test-results-dir>
+#   /p:VStestUseMSBuildOutput=false
+# We reproduce that here so STEP 3's renderer can parse authoritative
+# pass/fail counts from the TRX (instead of scraping console output, which is
+# fragile when many tests run and lines get interleaved or wrapped).
+$trxResultsDir = Join-Path $HostAppLogsDir "TestResults"
+if (-not (Test-Path $trxResultsDir)) {
+    New-Item -ItemType Directory -Path $trxResultsDir -Force | Out-Null
+}
+# Sanitize the trx file name. NUnit/MSTest reject some characters. We keep
+# alpha-numeric, dash, underscore and dot — same set Cake's
+# SanitizeTestResultsFilename uses.
+$trxBaseName = if ($Category) { "$Category-$Platform" } else {
+    ($TestFilter -replace '[^A-Za-z0-9._-]', '_')
+}
+$trxBaseName = $trxBaseName -replace '[^A-Za-z0-9._-]', '_'
+$trxFileName = "$trxBaseName.trx"
+$trxFilePath = Join-Path $trxResultsDir $trxFileName
+# Pre-clean stale TRX so we never read a previous run's results
+if (Test-Path $trxFilePath) { Remove-Item $trxFilePath -Force -ErrorAction SilentlyContinue }
+
+Write-Info "TRX file will be written to: $trxFilePath"
+
 try {
-    # Run dotnet test and capture output
-    $testOutput = & dotnet test $TestProject --filter $effectiveFilter --logger "console;verbosity=detailed" 2>&1
+    # Run dotnet test using the SAME loggers and arguments CI uses in
+    # `RunTestWithLocalDotNet` (eng/cake/dotnet.cake line 943-981).
+    $trxRunStart = Get-Date
+    $testArgs = @($TestProject, "--filter", $effectiveFilter,
+        "--logger", "trx;LogFileName=$trxFileName",
+        "--logger", "console;verbosity=normal",
+        "--results-directory", $trxResultsDir,
+        "/p:VStestUseMSBuildOutput=false")
+    Write-Info "Actual dotnet test args: $($testArgs -join ' ')"
+    $testOutput = & dotnet test @testArgs 2>&1
 
     # Save test output to file
     $testOutput | Out-File -FilePath $testOutputFile -Encoding UTF8
 
     # Output test results to the output stream so callers can capture them
     # (Write-Host goes to the Information stream which is not captured by 2>&1)
     $testOutput | ForEach-Object { Write-Output $_ }
-
+
+    # Surface the TRX path on a marker line so callers (Invoke-UITestWithRetry
+    # and Review-PR.ps1) can locate the authoritative results file regardless
+    # of where the working directory was when this script ran.
+    if (Test-Path $trxFilePath) {
+        Write-Output ">>> TRX_RESULT_FILE: $trxFilePath"
+    } else {
+        # dotnet test may have written the TRX with a slightly different name
+        # (e.g. LogFileName argument stripped on Windows, or it injected a
+        # timestamp). Fall back to scanning the results dir for any .trx
+        # written AFTER this run started — never pick up a stale TRX from a
+        # previous category that shares the same results directory.
+        $latestTrx = Get-ChildItem -Path $trxResultsDir -Filter "*.trx" -ErrorAction SilentlyContinue |
+                     Where-Object { $_.LastWriteTime -ge $trxRunStart } |
+                     Sort-Object LastWriteTime -Descending | Select-Object -First 1
+        if ($latestTrx) {
+            Write-Output ">>> TRX_RESULT_FILE: $($latestTrx.FullName)"
+        }
+    }
+
     $testExitCode = $LASTEXITCODE
 
     Write-Host ""