Merge pull request #351 from winnerspiros/copilot/fix-black-screen-issue-81afa2e2-8602-4671-849b-80662d737660

Android startup black-screen: force safe-mode fallback on persistent Vulkan RGB565 surface + always feed native watchdog heartbeat

Author: winnerspiros

.github/workflows/ci.yml

@@ -227,7 +227,7 @@ jobs:
       - name: Setup JDK 11
         uses: actions/setup-java@v5
         with:
-          distribution: microsoft
+          distribution: temurin
           java-version: 11
 
       - name: Install .NET 10.0.x
@@ -294,9 +294,13 @@ jobs:
             [ -d "$sdk_dir" ] || continue
             for sdk in $(ls "$sdk_dir" 2>/dev/null | grep '^MacOSX[0-9].*\.sdk$' | sort -rV); do
               cand="$sdk_dir/$sdk"
-              hdr_count=$(ls "$cand/usr/include" 2>/dev/null | wc -l | tr -d ' ')
+              cand_real=$(python3 -c "import os, sys; print(os.path.realpath(sys.argv[1]))" "$cand" 2>/dev/null || echo "$cand")
+              # Skip candidates that resolve back to the unversioned MacOSX.sdk alias.
+              # Re-linking MacOSX.sdk to itself creates a broken self-referential symlink.
+              [ "$(basename "$cand_real")" = "MacOSX.sdk" ] && continue
+              hdr_count=$(ls "$cand_real/usr/include" 2>/dev/null | wc -l | tr -d ' ')
               if [ "${hdr_count:-0}" -gt 5 ]; then
-                REAL_SDK="$cand"
+                REAL_SDK="$cand_real"
                 REAL_XCODE="$xapp"
                 break 2
               fi
@@ -306,9 +310,11 @@ jobs:
           if [ -z "$REAL_SDK" ]; then
             for sdk in $(ls /Library/Developer/CommandLineTools/SDKs 2>/dev/null | grep '^MacOSX[0-9].*\.sdk$' | sort -rV); do
               cand="/Library/Developer/CommandLineTools/SDKs/$sdk"
-              hdr_count=$(ls "$cand/usr/include" 2>/dev/null | wc -l | tr -d ' ')
+              cand_real=$(python3 -c "import os, sys; print(os.path.realpath(sys.argv[1]))" "$cand" 2>/dev/null || echo "$cand")
+              [ "$(basename "$cand_real")" = "MacOSX.sdk" ] && continue
+              hdr_count=$(ls "$cand_real/usr/include" 2>/dev/null | wc -l | tr -d ' ')
               if [ "${hdr_count:-0}" -gt 5 ]; then
-                REAL_SDK="$cand"
+                REAL_SDK="$cand_real"
                 break
               fi
             done
@@ -327,9 +333,13 @@ jobs:
             MACOS_SDK="$ACTIVE_SDKS_DIR/MacOSX.sdk"
             # Resolve REAL_SDK to its canonical (symlink-free) path to prevent ELOOP errors.
             REAL_SDK_CANON=$(python3 -c "import os, sys; print(os.path.realpath(sys.argv[1]))" "$REAL_SDK" 2>/dev/null || echo "$REAL_SDK")
-            sudo rm -rf "$MACOS_SDK"
-            sudo ln -sfn "$REAL_SDK_CANON" "$MACOS_SDK"
-            echo "Created MacOSX.sdk symlink -> $REAL_SDK_CANON"
+            if [ "$REAL_SDK_CANON" = "$MACOS_SDK" ]; then
+              echo "WARNING: resolved SDK points to active MacOSX.sdk alias; skipping relink to avoid self-referential symlink"
+            else
+              sudo rm -rf "$MACOS_SDK"
+              sudo ln -sfn "$REAL_SDK_CANON" "$MACOS_SDK"
+              echo "Created MacOSX.sdk symlink -> $REAL_SDK_CANON"
+            fi
           else
             echo "WARNING: no valid macOS SDK found; build may fail"
           fi

osu.Android/OsuGameActivity.cs

@@ -763,6 +763,7 @@ await Task.WhenAll(uris.Select(async uri =>
         // check-then-set in SurfaceChanged (lines ~732-734) is not a concurrency concern
         // because no two SurfaceChanged calls can overlap on the single UI thread.
         private volatile bool setFormatPending;
+        private volatile int setFormatAttempts;
 
         public IntPtr GetSurfaceGlobalRef()
         {
@@ -889,6 +890,7 @@ public void SurfaceChanged(ISurfaceHolder holder, global::Android.Graphics.Forma
                 && !AndroidStartupSafeMode.IsActive)
             {
                 setFormatPending = true;
+                setFormatAttempts = 1;
 
                 // Log to Runtime so the mid-session RGB565 reset is visible in the main log
                 // (and therefore in the notification overlay). Performance log gets the same
@@ -924,13 +926,37 @@ public void SurfaceChanged(ISurfaceHolder holder, global::Android.Graphics.Forma
                 return;
             }
 
+            if (format == global::Android.Graphics.Format.Rgb565
+                && LogManagement.IsVulkanConfigured()
+                && setFormatPending
+                && !AndroidStartupSafeMode.IsActive)
+            {
+                setFormatAttempts++;
+
+                if (setFormatAttempts >= 2)
+                {
+                    Logger.Log(
+                        "[osu!] Surface remained RGB565 after RGBA8888 request; restarting to enter safe-mode OpenGL fallback.",
+                        LoggingTarget.Runtime,
+                        LogLevel.Important);
+                    Logger.Log(
+                        "[osu!] Vulkan surface-format recovery failed (RGB565 persisted); killing process for safe-mode restart.",
+                        LoggingTarget.Performance,
+                        LogLevel.Important);
+
+                    try { global::Android.OS.Process.KillProcess(global::Android.OS.Process.MyPid()); }
+                    catch (Exception e) { Debug.WriteLine($"[osu!] Failed to kill process after persistent RGB565 detection: {e.Message}"); }
+                }
+            }
+
             // Release the pending-format guard once the surface is confirmed RGBA8888.
             // This allows future RGB565 detection (e.g. after a display-mode change that
             // would legitimately reset the format) while still blocking a second spurious
             // fire during the immediate teardown+recreate that follows our own SetFormat call.
             if (format == global::Android.Graphics.Format.Rgba8888 && setFormatPending)
             {
                 setFormatPending = false;
+                setFormatAttempts = 0;
                 Debug.WriteLine("[osu!] Surface format confirmed RGBA8888 — pending-format guard released.");
             }
 

osu.Android/OsuGameAndroid.cs

@@ -2247,6 +2247,18 @@ public override void SetHost(GameHost host)
             // still make the enqueue side spin on its lock.
             CrashDiagnostics.WriteAliveMarker("OsuGameAndroid.SetHost (about to start HangWatchdog)");
 
+            // Keep the native pthread watchdog fed even when verbose diagnostics are off.
+            // The native watchdog is always armed from Activity.OnCreate, but HangWatchdog
+            // (which normally forwards heartbeats) is verbose-gated below.
+            try
+            {
+                host.UpdateThread.Scheduler.AddDelayed(static () => NativeWatchdog.Heartbeat(), 1_000, true);
+            }
+            catch (Exception e)
+            {
+                Debug.WriteLine($"[osu!] Failed to schedule native watchdog heartbeat: {e.Message}");
+            }
+
             // Start the hang watchdog only when verbose diagnostics are enabled.
             // It writes /proc/self/task snapshots to native_crash.log on stalls — valuable
             // during debugging but adds a dedicated background thread and periodic file I/O