Skip to content

Update History

Bubbleshum edited this page May 22, 2026 · 17 revisions

Update History

Running changelog of WPR fixes / features / compatibility changes. Most recent entries are at the top.

22/05/2026

  • Compat: Max and the Magic Marker promoted from Legacy → Playable (end-to-end after the three fixes below)
  • Compat: Pac-Man promoted from Legacy → Playable (first-hand verification on this fork)
  • Fix: Max and the Magic Marker crashed on window close with an AccessViolationException originating in FNA3D_AddDisposeTexture from ~Texture(). Root cause: the Texture finalizer ran after GraphicsDevice.Dispose had already destroyed the FNA3D device, so the native dispose call dereferenced freed device state. Fix: guard the native dispose call in Src/ThirdParty/fna/src/Graphics/Texture.cs on !GraphicsDevice.IsDisposed. Shim-only — no reinstall
  • Fix: Real touchscreen input never registered in Max and the Magic Marker (and any other XNA game polling SDL touch). Root cause: SDL2's touch device IDs are 64-bit (SDL_TouchID), but TouchPanel.LastActiveTouchId was declared int, so the value was truncated as it round-tripped through TouchPanel and the polling path called SDL_GetNumTouchFingers(bogusId) — which always returned 0. Widened the field to long in Src/ThirdParty/fna/src/Input/Touch/TouchPanel.cs and dropped the (int) casts in Src/ThirdParty/fna/src/FNAPlatform/SDL2_FNAPlatform.cs. Shim-only — no reinstall
  • Fix: Single-finger drawing in Max and the Magic Marker refused to start a stroke — one real finger was being reported as two simultaneous touches. Root cause: SDL synthesises mouse events from touch events by default (SDL_HINT_TOUCH_MOUSE_EVENTS = 1); we already filtered the synthetic events in our event pump, but the mouse-as-touch poll still read the polluted SDL_GetMouseState, writing a phantom finger into slot 7 alongside the real one. Max's drawing logic treats a 2-touch frame as a multi-finger gesture and refuses to start a stroke. Fix: set SDL_HINT_TOUCH_MOUSE_EVENTS = "0" in the Windows init block of Src/ThirdParty/fna/src/FNAPlatform/SDL2_FNAPlatform.cs so SDL stops synthesising mouse events from touch entirely. Shim-only — no reinstall
  • Compat: Z0MB1ES (on teh ph0ne) confirmed Playable
  • Compat: I Dig It promoted from Broken → Playable (end-to-end after the two fixes below)
  • Fix: I Dig It crashed at frame 1 of IDigItApp.Update with FileLoadException(0x80131515) wrapping NotSupportedException: A non-collectible assembly may not reference a collectible assembly on its first reference to its bundled Chipmunk.dll physics wrapper. Root cause: the AssemblyLoadContext.Default.Resolving handler in ApplicationLaunch loaded every sibling DLL it found in the install dir into the collectible user ALC and handed the resulting assembly back to its Default-ALC requestor. The CLR then refused to bind a non-collectible static reference to that collectible assembly, so the JIT failed at the IDigItApp.Update call site. Fix: split the handler's two cases — the main game DLL still routes into the user ALC (FNA's ContentTypeReaderManager.Type.GetType legitimately needs the collectible type back, and loading it into Default would re-trigger the launch-1-statics contamination bug guarded by the existing comment), but sibling DLLs now load into the Default ALC so non-collectible static refs bind cleanly. Trade-off: those siblings persist across launches; for self-contained native-wrapper libs like Chipmunk that's harmless. Added a _CurrentMainAssemblyName field to drive the discrimination. Shim-only — no reinstall
  • Fix: After clearing the Chipmunk crash, I Dig It got past the splash but still wedged just before the menu rendered, this time with MissingMethodException: LeaderboardReader.get_TotalLeaderboardSize(). Our shim had TotalLeaderboardSize declared as a method; XNA exposes it as an int property. Adding only that surfaced the next missing member (get_PageStart), so the full property surface (CanPageDown, CanPageUp, IsDisposed, IsSynchronizedWithLiveServer, Leaderboard, PageSize, PageStart, TotalLeaderboardSize) plus EndPageDown/EndPageUp/EndRead
    • PageDown/PageUp + IDisposable.Dispose was filled in in one pass. All stubs report "no leaderboard available" (false / 0 / empty / fresh reader), so the game treats the leaderboard screen as empty rather than crashing — appropriate given there's no live Xbox LIVE backend. Shim-only — no reinstall
  • Compat: Castlevania Puzzle promoted from Partial → Playable (story / arcade mode now start; previously tapping a save slot silently no-op'd)
  • Fix: Castlevania Puzzle's save-slot screen rendered correctly but tapping a slot to start the game did nothing. The slot tap fires DashResourceProvider.getLocalizedBinary, which uses the standard WP7 localization-fallback idiom try { OpenStream("dat\\data\\items.en-GB") } catch { OpenStream("dat\\data\\items") }. Our FNA TitleContainer.OpenStream was catching FileNotFoundException (and every other exception), logging it to [wpr-ex], and returning null instead of rethrowing. The game's catch therefore never fired, the fallback to the unsuffixed file never ran, and the subsequent new BinaryReader(null) threw ArgumentNullException inside Game.Update — which Game.Tick swallows. The save-slot UI stayed visible, so the user perceived the "Start Game" button as missing. Fix: rethrow at the end of both catch blocks in TitleContainer.OpenStream after the Content/Scenes/ fallback has been given a chance to run. [wpr-ex] diagnostics still log on the way out. Likely had silent secondary effects in any other game using the same localization-fallback pattern. Shim-only — no reinstall
  • Compat: Assassin's Creed: Altaïr's Chronicles promoted from Partial → Playable (end-to-end after SignedInGamer.SignedIn deferred-invoke fix below)
  • Fix: Assassin's Creed crashed in XNAGame..ctor with a NullReferenceException inside its own SignedIn handler. Root cause: SignedInGamer.SignedIn's add accessor took two paths — a deferred 2s Task.Delay for the very first subscriber in the session, and a synchronous invoke for any later subscriber (gated by a static FirstSignInSessionDone flag). Two compounding problems: (1) SignedInGamer.Reset() was called after Activator.CreateInstance(mainType!) in ApplicationLaunch.Start, so the flag set by a previous game launch in the same WPR session carried into the new game's ctor; (2) the synchronous path then fired the just-attached handler during +=, while the game's ctor was still mid-construction, and the immediate-fire branch had no try/catch wrapper (unlike the deferred branch). Assassin's XNAGame.a accessed a this-field its ctor hadn't initialised yet, NRE'd, and the exception escaped up through Activator.CreateInstance. Fix: collapse the two paths into one — every subscription dispatches on Task.Run (delay 0 for late subscribers, 2s for the first), goes through the same _SignInGate semaphore + try/catch, and never runs synchronously from +=. Also moved SignedInGamer.Reset() to run before Activator.CreateInstance. Shim-only — no reinstall
  • Compat: Plants vs Zombies promoted from Broken → Playable (end-to-end after achievement-seed + SpriteBatch tolerance fixes below)
  • Fix: PvZ launched to a black screen because Lawn.AchievementsWidget.Draw NRE'd inside Achievements.GetAchievementItem — the game's static gAchievementList was empty, so the lookup returned null and the NRE escaped mid-SpriteBatch.Begin, wedging every subsequent frame on "Begin has been called before calling End". Root cause: PvZ stores its 18 achievement keys in an inline Achievements.ACHIEVEMENT_KEYS[] static array (built in the cctor), so none of XnaAchievementCodeExtractor's three sources — Content/xml/socialnetworks.xml.xnb, IL ldstr near AwardAchievement* callsites, or Content/Achievements/*.xnb filenames — recovered any keys at install time. The seeder wrote 0 rows; BeginGetAchievements returned 0 rows; the game's GetAchievementsCallback added 0 items to gAchievementList. Added a Source D (KnownProductCatalogues) keyed by ProductId that returns hardcoded keys recovered via decompilation, threaded productId through XnaAchievementCodeExtractor.ExtractRich and XnaAchievementSeeder.SeedAsync. Install-time change — affected games need reinstalling. Now seeds 18 rows and BeginGetAchievements: 18 rows confirms it
  • Fix: After the achievement seed worked the game flickered on main menu and in level — Lawn.GameSelector.DrawOverlay still NRE'd every frame on some other null reference, but the deeper problem was that PvZ's Sexy.Graphics layer tracks its own spritebatchBegan flag separately from FNA's beginCalled, and the two desync once any Draw exception leaves the game's flag stale. The result was a within-frame double-Begin via SetupDrawMode → EndFrame → EndDrawImageTransformed → BeginFrame, throwing InvalidOperationException and dropping every other frame. We can't reach the game's private flag from the shim. Made FNA's SpriteBatch.Begin / End tolerant of out-of-order calls: a second Begin without an intervening End soft-resets and starts a fresh batch (drops the queued sprites of the discarded batch); a stray End without a matching Begin no-ops. First 5 of each are logged so we can spot if it's firing in production. Shim-only — no reinstall
  • Feat: Diagnostic [wpr-trace] BeginGetAchievements: N rows for <ProductId> log in SignedInGamer.BeginGetAchievements confirms whether the install- time seed populated for a given game on first launch
  • Feat: Keyboard accelerometer simulator. New Controls page in the desktop sidebar binds four keys (defaults WASD) to tilt directions; readings flow through the existing Microsoft.Devices.Sensors.Accelerometer so games see them without any per-game shim work. Sensitivity slider, master toggle, in-game tilt-overlay (Avalonia for Silverlight host, FNA DrawableGameComponent for XNA), and a live-preview dial on the Controls page. Orientation-aware: the screen-relative key intent gets rotated into the device-portrait frame the WP7 sensor contract expects, so a landscape game (W = "tilt up the screen") produces the correct device-X tilt the game interprets as steer-left. Desktop orientation is inferred from the back-buffer aspect because FNA's Window.CurrentOrientation only updates from SDL display-rotation events that never fire on the desktop. Accelerometer.CurrentValue / IsDataValid / TimeBetweenUpdates were also added so games that poll instead of subscribing to ReadingChanged see live readings too
  • Compat: Hydro Thunder GO promoted from Partial → Playable (steerable via the keyboard accelerometer)
  • Compat: Uno confirmed Playable
  • Fix: Sonic 4 Episode I self-paused on every Update tick. The game's AppMain.isForeground flag is flipped true only by its Game.Activated handler, and Activated never fired on WPR because INTERNAL_isActive was initialised true (suppressing the BeforeLoop setter's no-op transition to avoid Asphalt 5's pre-Initialize KeyNotFoundException). Net effect: isForeground stayed false, the game's pause condition (!isForeground ORed into the trigger) re-armed every frame, and pause reasserted itself immediately after dismissal. Fix: defer a one-shot OnActivated to the end of the first Game.Tick, where every game's per-Update state (including Asphalt 5's) is already populated
  • Fix: Window background/foreground transitions now fire Deactivated/Activated correctly on desktop and Android. Restored the SDL_WINDOWEVENT_FOCUS_LOST → IsActive=false branch in SDL2_FNAPlatform.PollEvents that had been commented out under a //RnD marker. Symmetric FOCUS_GAINED was already wired
  • Feat: Multi-touch input is now additive with the mouse-as-touch shim. UpdateTouchPanelState previously branched either-or — if TouchPanel.MouseAsTouch was on (the default for XNA games on WPR) the real-finger poll loop was skipped entirely, capping all input at a single touch even on multi-touch hardware. New shape: real fingers fill slots 0..MAX_TOUCHES-2 from SDL_GetTouchFinger; when MouseAsTouch is on the mouse takes the last slot with synthetic finger ID int.MaxValue so it can never collide with a real finger's ID. Sonic 4's "hold D-pad + tap jump" gameplay now works
  • Fix: Asphalt 5 splash → menu tap-to-continue now registers. The per-Game.Tick FrameworkDispatcher.Update() call added on 21/05 was redundant — stock FNA's Game.Update already pumps the dispatcher at its end. Pumping twice per tick made TouchPanel.Update run twice in close succession; the second run promoted touches[0] from Pressed to Moved before CGame1.g() could read it, so h2.b (press handler) never recorded the finger and h2.d's if (i > 0) release-guard always failed. Splash-state lt.b() waiting on be.ey.fm != 0 was the surfaced case. Removed the redundant pump in FNA Game.Tick
  • Compat: Asphalt 5 promoted from Partial → Playable
  • Compat: Sonic 4 Episode I confirmed Playable (pause loop fixed, multi-touch routed)
  • Compat: Tentacles promoted from Broken → Playable (end-to-end after the fixes below)
  • Fix: FNA's ContentTypeReaderManager couldn't resolve ReflectiveReader<PressPlay.FFWD.Scene> because Type.GetType() doesn't see types in the user game's collectible ALC, and the existing string.Split(',') fallback broke on the generic argument [[PressPlay.FFWD.Scene, PressPlay.FFWD, …]] (the comma between the inner type and its assembly hint was indistinguishable from the outer delimiter). Replaced with Type.GetType's resolver-callback overload that walks AppDomain.CurrentDomain.GetAssemblies() — which returns every loaded assembly across every ALC — so generic readers parameterised by user-ALC types resolve correctly. Tentacles' Preloader scene XNB now deserialises
  • Fix: FFWD-based games call Application.LoadLevel("X") with bare scene names even though every level XNB ships under Content/Scenes/. ContentManager constructed Content/X.xnb, the file wasn't there, AssetHelper.Load<T> silently swallowed the failure and returned default(T), and the loading screen hung forever waiting on loadingProgress == 1.0f. TitleContainer.OpenStream now retries Content/<name>.xnb as Content/Scenes/<name>.xnb when the original is missing and has no subdirectory hint. Safe for non-FFWD games (fallback file simply won't exist either)
  • Fix: Microsoft.Phone.Shell.StandardTileData.Count was int in our shim but the WP7 SDK uses int?. Tentacles' live-tile updater calls set_Count(int?) every frame from a component Update and was tripping MissingMethodException ~once per tick
  • Fix: FNA.Game.RunLoop now wraps the final OnExiting(this, EventArgs.Empty) in try/catch + log. Tentacles' MetricsSender.CreateTearDownExtendedKeys NREs on GlobalManager.Instance.currentProfile when the user closes during early boot (currentProfile is null until the preloader finishes loading it). Can't fix the game code, but the host no longer surfaces its "unexpected error" dialog on close
  • Fix: SignedInGamer.SignedIn handler invocations now serialise behind a SemaphoreSlim. Tentacles' Game1.Initialize registers the same callback twice; both invocations were firing in parallel via separate Task.Delay.ContinueWiths, racing on the shared AchievementContext.Current and tripping EF Core's ConcurrencyDetector inside Gamer.GetProfile. Handlers still get their independent ~2 s delay; only the synchronous Invoke runs single-file
  • Fix: Mouse-as-touch clicks now produce GestureType.Tap gestures on desktop. SDL_MOUSEMOTION was forwarded to INTERNAL_onTouchEvent unconditionally, including hover (no button held). Hover motion ran GestureDetector.OnMoved which set activeFingerId = 1; the next MOUSEBUTTONDOWN's OnPressed(1) then saw activeFingerId != NO_FINGER and routed into pinch-init (state = PINCHING). The matching MOUSEBUTTONUP ran OnReleased_Pinch instead of Tap detection — so Tap-gated screens (Tentacles' LemmyTravelScreen, "tap to continue" prompts) never advanced from mouse but worked fine from real touch (no hover). Fix: SDL2_FNAPlatform skips the synthesised Moved event when evt.motion.state == 0. Drag (button-held motion) still goes through, so swipe/pinch/drag from mouse continue to work
  • Feat: Periodic FFWD loading-state heartbeat in FNA Game.Tick ([wpr-heartbeat], fires ~every 2 s past the first-30-ticks verbose trace cap). Reflects PressPlay.FFWD.Application and PressPlay.Tentacles.Scripts.LevelHandler static fields plus the active screen stack and the current level identity — turned a "stuck on loading screen" symptom into a precise gate-by-gate diagnosis. Best-effort, swallows all reflection errors, [Conditional("DEBUG")]-gated
  • Feat: ContentManager.Load<T> now logs the underlying exception with type/HResult/inner/stack before rethrowing ([wpr-content]). FFWD's AssetHelper.Load silently catches ContentManager.Load failures and returns default(T); without the trace, a missing ContentTypeReader or malformed XNB was invisible. Trace lets the caller's swallow stand — only adds visibility
  • Feat: Assembly-resolver diagnostics in ApplicationLaunch[wpr-resolve-user] / [wpr-resolve-default] lines log every Resolving probe with full exception details when LoadFromAssemblyPath fails

21/05/2026

  • Fix: Asphalt 5 sat on a blank loading screen — StartupMode enum integer values now match the WP7 SDK (Launch=1, Activate=2), and FNA's Game.IsActive no longer fires a spurious Activated event at first frame
  • Fix: Tentacles crashed at exit on a null ApplicationCurrentMemoryUsageMicrosoft.Phone.Info.DeviceExtendedProperties now returns the WP7 memory counters (ApplicationCurrentMemoryUsage, ApplicationPeakMemoryUsage, ApplicationMemoryUsageLimit)
  • Fix: XNA games that wait on MediaPlayer / song-finished callbacks before advancing past their splash now progress — Game.Tick auto-pumps FrameworkDispatcher.Update() once per update, matching WP7 XNA 4.0 behaviour
  • Feat: Per-game wpr_game_debug.log and the in-engine [wpr-trace] output are now #if DEBUG-gated via WprDebugTrace — Release builds elide the trace formatting and file listener entirely (no log spam, no per-frame cost)
  • Feat: Richer assembly-resolver diagnostics in ApplicationLaunchResolving failures log the underlying exception type, HResult and inner exception instead of the opaque "Operation is not supported" surface error
  • Compat: Asphalt 5 promoted from Broken → Partial (loads splash, gameplay TBD)
  • Compat: Final Fantasy promoted to Playable
  • Compat: Fruit Ninja confirmed Playable (no longer blocked on GamerProfile.GetGamerPicture)

17/05/2026

  • Fix: Small bug on second launch of XNA games, where resources arent released fully
  • Feat: Added Windows based notifications
  • Feat: Added game icon as window icon for XNA

Clone this wiki locally