[FastDeploy] Retry first run-as query to absorb post-install data-dir race (XA0137) (#11809)

The Fast Deployment install-vs-`run-as` race (#7821) fires `XA0137` (`run-as: couldn't stat /data/user/0/<pkg>: No such file or directory`) on the **primary user (user 0)** ~daily in CI. The existing `EnsureUserIsRunning` mitigation (#11322) only covers secondary users and explicitly skips user 0 — and `am start-user -w 0` wouldn't help anyway, since user 0 is already running; the real issue is the brief window after `pm install` before `/data/user/0/<pkg>` is stat-able through `run-as`.

## Changes

`src/Xamarin.Android.Build.Debugging.Tasks/Tasks/FastDeploy.cs`:

- **`QueryInternalPathWithRetry`** — wraps the first `run-as <pkg> pwd` query in `CheckAppInstalledAndDebuggable` with a bounded poll (10 attempts × 500 ms) that retries while the result matches the transient race signature, before falling through to `RaiseRunAsError`/XA0137. Applies to **all** users, including user 0; no-op when the directory already exists.
- **`IsTransientRunAsStatRace`** — matches the `couldn't stat … No such file or directory` signature only, so genuine `run-as` failures (permission denied, not debuggable, corrupt install, etc.) still surface immediately.

```csharp
async Task<string> QueryInternalPathWithRetry ()
{
    const int maxAttempts = 10;
    var delay = TimeSpan.FromMilliseconds (500);
    string result = await Device.RunAs (packageInfo, "pwd");
    for (int attempt = 1; attempt < maxAttempts && IsTransientRunAsStatRace (result); attempt++) {
        LogDiagnostic ($"run-as could not stat the data directory for {packageInfo.PackageName} yet (attempt {attempt}/{maxAttempts}); retrying in {delay.TotalMilliseconds:0} ms. Output: {result?.Trim ()}");
        await Task.Delay (delay, CancellationToken);
        result = await Device.RunAs (packageInfo, "pwd");
    }
    return result;
}
```

`EnsureUserIsRunning` is left intact; the retry sits after it, so secondary users retain both mitigations.

## Notes

No unit-test project exists for `Xamarin.Android.Build.Debugging.Tasks`; the path is exercised by the on-device `MSBuildDeviceIntegration` install tests where the race was observed.

Co-authored-by: simonrozsival <374616+simonrozsival@users.noreply.github.com>

Author: Copilot

src/Xamarin.Android.Build.Debugging.Tasks/Properties/AssemblyInfo.cs

@@ -17,3 +17,5 @@
 
 // The following GUID is for the ID of the typelib if this project is exposed to COM
 [assembly: Guid ("42bb9aea-f4d8-4b43-8956-8e1fee857697")]
+
+[assembly: InternalsVisibleTo ("Xamarin.Android.Build.Tests, PublicKey=0024000004800000940000000602000000240000525341310004000011000000438ac2a5acfbf16cbd2b2b47a62762f273df9cb2795ceccdf77d10bf508e69e7a362ea7a45455bbf3ac955e1f2e2814f144e5d817efc4c6502cc012df310783348304e3ae38573c6d658c234025821fda87a0be8a0d504df564e2c93b2b878925f42503e9d54dfef9f9586d9e6f38a305769587b1de01f6c0410328b2c9733db")]

src/Xamarin.Android.Build.Debugging.Tasks/Tasks/FastDeploy.cs

@@ -362,7 +362,7 @@ async Task CheckAppInstalledAndDebuggable (string packageName)
 			packageInfo.UserId = UserID;
 			packageInfo.PackageName = packageName;
 			await EnsureUserIsRunning ();
-			packageInfo.InternalPath = packageInfo.InternalPath ?? await Device.RunAs (packageInfo, "pwd");
+			packageInfo.InternalPath = packageInfo.InternalPath ?? await QueryInternalPathWithRetry ();
 			if (packageInfo.InternalPath.IndexOf ("Permission denied", StringComparison.OrdinalIgnoreCase) >= 0) {
 				packageInfo.InternalPath = await Device.RunAs (packageInfo, "readlink", "-f", ".");
 			}
@@ -395,6 +395,53 @@ async Task CheckAppInstalledAndDebuggable (string packageName)
 			return;
 		}
 
+		/// <summary>
+		/// Issues the first <c>run-as &lt;pkg&gt; pwd</c> query, retrying briefly while the
+		/// per-user data directory is not yet stat-able through <c>run-as</c>.
+		/// </summary>
+		/// <remarks>
+		/// <para>Immediately after <c>pm install</c>, the per-user data directory
+		/// <c>/data/user/N/&lt;pkg&gt;</c> may not yet be stat-able through <c>run-as</c>,
+		/// even for the primary user (id 0). During that window <c>run-as</c> returns
+		/// <c>run-as: couldn't stat /data/user/N/&lt;pkg&gt;: No such file or directory</c>,
+		/// which otherwise raises <c>XA0137</c> and disables Fast Deployment. This races
+		/// install on the primary user ~daily in CI. Poll for a bounded period to let the
+		/// directory materialize before giving up. See
+		/// https://github.com/dotnet/android/issues/7821 and
+		/// https://github.com/dotnet/android/issues/11808.</para>
+		/// <para>Retry policy: up to 10 attempts with a 500 ms delay between each, giving
+		/// a maximum wait of 4.5 seconds before the error is surfaced as <c>XA0137</c>.
+		/// Only the transient <c>couldn't stat … No such file or directory</c> signature
+		/// (detected by <see cref="IsTransientRunAsStatRace"/>) triggers a retry; all other
+		/// <c>run-as</c> failures are surfaced immediately.</para>
+		/// </remarks>
+		async Task<string> QueryInternalPathWithRetry ()
+		{
+			const int maxAttempts = 10;
+			var delay = TimeSpan.FromMilliseconds (500);
+			string result = await Device.RunAs (packageInfo, "pwd");
+			for (int attempt = 1; attempt < maxAttempts && IsTransientRunAsStatRace (result); attempt++) {
+				LogDiagnostic ($"run-as could not stat the data directory for {packageInfo.PackageName} yet (attempt {attempt}/{maxAttempts}); retrying in {delay.TotalMilliseconds:0} ms. Output: {result?.Trim ()}");
+				await Task.Delay (delay, CancellationToken);
+				result = await Device.RunAs (packageInfo, "pwd");
+			}
+			return result;
+		}
+
+		/// <summary>
+		/// Returns <see langword="true"/> when a <c>run-as</c> result matches the transient
+		/// install-vs-run-as race signature (<c>couldn't stat … No such file or directory</c>),
+		/// i.e. the per-user data directory has not yet materialized after <c>pm install</c>.
+		/// </summary>
+		internal static bool IsTransientRunAsStatRace (string result)
+		{
+			if (string.IsNullOrEmpty (result)) {
+				return false;
+			}
+			return result.IndexOf ("couldn't stat", StringComparison.OrdinalIgnoreCase) >= 0 &&
+				result.IndexOf ("No such file or directory", StringComparison.OrdinalIgnoreCase) >= 0;
+		}
+
 		/// <summary>
 		/// Ensures the secondary Android user targeted by this deployment is in the
 		/// 'running' state before any <c>run-as</c> query is issued against it.

src/Xamarin.Android.Build.Tasks/Tests/Xamarin.Android.Build.Tests/DebuggingTasksTests.cs

@@ -74,4 +74,54 @@ public void TestResolveToolsExists ()
 		}
 
 	}
+
+	/// <summary>
+	/// Unit tests for <see cref="FastDeploy"/> helper methods that do not require a device.
+	/// </summary>
+	[TestFixture]
+	public class FastDeployTests
+	{
+		// Canonical transient race output produced by the Android run-as tool when
+		// the per-user data directory has not yet materialized after pm install.
+		static readonly string [] TransientRaceOutputs = {
+			"run-as: couldn't stat /data/user/0/com.example.app: No such file or directory",
+			"run-as: couldn't stat /data/user/10/com.example.app: No such file or directory",
+			// Verify case-insensitivity of the detection.
+			"run-as: Couldn't Stat /data/user/0/com.example.app: No Such File Or Directory",
+			// Extra surrounding whitespace / newlines as they may appear in raw adb output.
+			"  run-as: couldn't stat /data/user/0/com.example.app: No such file or directory\n",
+		};
+
+		// Genuine run-as failures that must NOT be swallowed by the retry loop.
+		static readonly string [] NonTransientOutputs = {
+			// Null / empty — first guard in the implementation.
+			null,
+			"",
+			// Successful pwd output — the data directory already exists.
+			"/data/user/0/com.example.app",
+			// Package not debuggable.
+			"run-as: package 'com.example.app' is not debuggable",
+			// Package not installed.
+			"run-as: package 'com.example.app' is unknown",
+			// Permission denied (SELinux / policy).
+			"run-as: couldn't stat /data/user/0/com.example.app: Permission denied",
+			// Only one of the two required substrings — must not match.
+			"run-as: couldn't stat /data/user/0/com.example.app",
+			"No such file or directory",
+		};
+
+		[TestCaseSource (nameof (TransientRaceOutputs))]
+		public void IsTransientRunAsStatRace_ReturnsTrueForRaceSignature (string output)
+		{
+			Assert.IsTrue (FastDeploy.IsTransientRunAsStatRace (output),
+				$"Expected transient-race detection for: {output}");
+		}
+
+		[TestCaseSource (nameof (NonTransientOutputs))]
+		public void IsTransientRunAsStatRace_ReturnsFalseForNonTransientOutput (string output)
+		{
+			Assert.IsFalse (FastDeploy.IsTransientRunAsStatRace (output),
+				$"Expected no transient-race detection for: {output}");
+		}
+	}
 }