fix(app): #343 — a wounded render loop defers the native release instead of throwing over the real failure
Some checks are pending
Headless portability / portable-headless (ubuntu-latest) (push) Waiting to run
Headless portability / portable-headless (windows-latest) (push) Waiting to run
Headless portability / linux-graphical (push) Waiting to run
Headless portability / linux-vulkan (push) Waiting to run

Root cause pinned by IL-decompiling Silk.NET.Windowing.Common:
ViewImplementationBase._inRenderLoop is set at DoRender/DoUpdate entry
and cleared ONLY on normal return, so a throwing frame callback leaves
it armed forever and any later Dispose -> Reset throws "You cannot call
Reset inside of the render loop", exit 82, replacing the original
wounding exception in the report.

The fix mirrors Silk's own bracket exactly: GameWindow._renderLoopArmed
set at OnUpdate/OnRender entry, cleared only on their normal return —
deliberately NOT in a finally, so it tracks the wound the same way
Silk's private field does. ReleaseNativeWindow checks it before
disposing: armed -> best-effort Close() (swallowed so it can never
become the reported failure), no Dispose, and a new terminal status
CompleteWithDeferredNativeRelease with Error kept null — the original
exception stays the primary report. Healthy paths (OnClosing's
in-loop completion, Run()'s tail release) are byte-unchanged, and the
new PublishNativeWindow parameters default to null so every existing
caller and test behaves identically.

Sabotage: disabling the armed-check flipped the deferral test to
Expected CompleteWithDeferredNativeRelease / Actual Complete —
the guard is what the test exercises. Clean-room suite 11,262 / 6 / 1,
the 1 being #340's documented load flake (passed standalone; second
recorded firing noted in its entry).

Queue: #344 done, #343 done; next #345's instrumented mechanism
session, then #341's boundary hunt.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
Erik 2026-08-07 10:28:29 +02:00
parent 52bdf4df71
commit 6c6664a685
4 changed files with 194 additions and 10 deletions

View file

@ -257,7 +257,19 @@ mid-flight-teleport-only. Next-session queue: #344, #343, then S4b/S6.
## #343 — Shutdown after a wounded render loop: Silk Reset called inside the render loop, exit 82
**Status:** OPEN. LOW-MEDIUM — only reachable after a render-frame exception
**Status:** FIXED 2026-08-08. Root cause pinned by IL-decompiling Silk:
`ViewImplementationBase._inRenderLoop` is cleared only on a frame callback's
NORMAL return, so a throwing callback leaves it armed forever and the
disposal's `Reset()` throws. The fix mirrors that bracket exactly
(`GameWindow._renderLoopArmed`, deliberately not cleared in a finally),
and `ReleaseNativeWindow` defers when armed: best-effort `Close()`, no
`Dispose()`, new terminal status `CompleteWithDeferredNativeRelease` with
`Error` kept null so the ORIGINAL wounding exception remains the primary
report. Sabotage-verified; healthy paths byte-unchanged. Clean-room suite
11,262 passed / 6 skipped / 1 failed = #340's documented flake, which
passed standalone (its second recorded firing).
**Original status:** OPEN. LOW-MEDIUM — only reachable after a render-frame exception
(#339's crash was the trigger), but it turns a diagnosable failure into an
`AbandonedIncomplete` shutdown. `GameWindowLifetime.ReleaseNativeWindow:294`
calls `ViewImplementationBase.Dispose``Reset` while the render loop is

View file

@ -42,6 +42,18 @@ public sealed class GameWindow :
private readonly WorldEvents _worldEvents;
private readonly HostQuiescenceGate _hostQuiescence = new();
private IWindow? _window;
// #343: mirrors Silk's own ViewImplementationBase._inRenderLoop guard,
// which we cannot read directly (it is a private field on Silk's
// internal type). Set true before OnUpdate/OnRender run and cleared
// only on their normal (non-throwing) return — exactly like Silk's own
// bracket around DoUpdate/DoRender — so a frame callback that throws
// leaves this stuck true, same as Silk's real guard. GameWindowLifetime
// reads it before disposing the native window, so a wounded loop defers
// the release instead of calling Reset() while Silk still thinks it is
// mid-frame (which throws "You cannot call `Reset` inside of the render
// loop!" and would otherwise bury whatever exception actually wounded
// the loop). See docs/ISSUES.md #343.
private bool _renderLoopArmed;
private SilkWindowCallbackBinding? _windowCallbacks;
private GameWindowGraphics? _graphics;
// Campaign V slice V6h: borrowed, not owned — _graphics owns the context and
@ -722,7 +734,11 @@ public sealed class GameWindow :
_startupQuality = startup.Quality;
_window = Window.Create(options);
_lifetime.PublishNativeWindow(_window);
IWindow window = _window;
_lifetime.PublishNativeWindow(
window,
isRenderLoopArmed: () => _renderLoopArmed,
requestClose: window.Close);
_displayFramePacing.BindSurface(
new SilkDisplayFramePacingSurface(_window));
// The fixed binding preserves main Render before post-render pacing,
@ -1505,15 +1521,24 @@ public sealed class GameWindow :
private void OnUpdate(double dt)
{
// #343: armed before the callback body runs, cleared only on normal
// return. Deliberately NOT a try/finally — if Tick throws, the flag
// must stay true (mirroring Silk's own stuck _inRenderLoop guard),
// not get cleared on the way out.
_renderLoopArmed = true;
using var _updStage = _frameProfiler.BeginStage(
AcDream.App.Diagnostics.FrameStage.Update);
_frameGraphs.Tick(new AcDream.App.Update.UpdateFrameInput(dt));
_renderLoopArmed = false;
}
// Performance overlay state — updated every ~0.5s and written to the
// window title so there's zero rendering cost (no font/overlay needed).
private void OnRender(double deltaSeconds)
{
// #343: see OnUpdate above — armed on entry, cleared on every normal
// exit path below, left stuck true if anything here throws.
_renderLoopArmed = true;
Vector2D<int> size = _window!.Size;
// Campaign V slice V6h: swapchain currency is the one piece of
// presentation the RHI contract deliberately leaves to the host (plan
@ -1521,7 +1546,10 @@ public sealed class GameWindow :
// out-of-date one is recreated at a frame boundary, the only safe point.
// The handoff below stays exactly one call on both backends.
if (_vulkanGraphics is { } vulkan && !vulkan.PrepareFrame())
{
_renderLoopArmed = false;
return;
}
try
{
_frameGraphs.Render(
@ -1535,10 +1563,12 @@ public sealed class GameWindow :
when (_vulkanGraphics is not null)
{
_vulkanGraphics.RequestRecreate();
_renderLoopArmed = false;
return;
}
_vulkanGraphics?.NoteFrameClosed();
_renderLoopArmed = false;
}
// IsEntityCurrentlyMoving REMOVED (2026-07-09): it powered a cache-bypass

View file

@ -34,6 +34,14 @@ internal enum GameWindowLifetimeStatus
RetryableIncomplete,
Complete,
CompleteWithCleanupFailures,
// #343: the resource-shutdown transaction converged, but the native
// window's release was deferred because Silk's per-frame render-loop
// guard was still armed (a render/update callback threw without
// reaching its own reset line). Disposing here would trade whatever
// wounded the loop for Silk's unrelated "Reset inside of the render
// loop" exception. Distinct from AbandonedIncomplete: nothing failed,
// the native release just did not happen yet.
CompleteWithDeferredNativeRelease,
AbandonedIncomplete,
}
@ -46,6 +54,7 @@ internal sealed record GameWindowLifetimeReport(
public bool IsTerminal => Status is
GameWindowLifetimeStatus.Complete
or GameWindowLifetimeStatus.CompleteWithCleanupFailures
or GameWindowLifetimeStatus.CompleteWithDeferredNativeRelease
or GameWindowLifetimeStatus.AbandonedIncomplete;
}
@ -145,6 +154,12 @@ internal sealed class GameWindowLifetime
private GameWindowShutdownRoots? _roots;
private ResourceShutdownTransaction? _transaction;
private IDisposable? _nativeWindow;
// #343: caller-supplied signal for "is Silk's render loop still armed"
// and a best-effort close request, both optional so every existing
// caller/test that publishes a bare IDisposable keeps disposing exactly
// as before. Null means "assume not armed" (unchanged prior behavior).
private Func<bool>? _isNativeRenderLoopArmed;
private Action? _requestNativeClose;
private bool _shutdownRootsPublished;
private bool _nativeReleaseAttempted;
private bool _completing;
@ -169,12 +184,17 @@ internal sealed class GameWindowLifetime
_shutdownRootsPublished || _injectedTransactionFactory is not null;
internal bool RetainsShutdownGraph => _roots is not null || _transaction is not null;
public void PublishNativeWindow(IDisposable nativeWindow)
public void PublishNativeWindow(
IDisposable nativeWindow,
Func<bool>? isRenderLoopArmed = null,
Action? requestClose = null)
{
ArgumentNullException.ThrowIfNull(nativeWindow);
if (_nativeReleaseAttempted || _nativeWindow is not null)
throw new InvalidOperationException("A native window is already lifetime-owned.");
_nativeWindow = nativeWindow;
_isNativeRenderLoopArmed = isRenderLoopArmed;
_requestNativeClose = requestClose;
}
public void PublishShutdownRoots(GameWindowShutdownRoots roots)
@ -237,7 +257,12 @@ internal sealed class GameWindowLifetime
if (_report.Status == GameWindowLifetimeStatus.RetryableIncomplete)
{
AbandonRetainedRootsAfterTerminalFailure();
Exception? nativeFailure = ReleaseNativeWindow();
(_, Exception? nativeFailure) = ReleaseNativeWindow();
// A deferred native release (#343) is not a failure — it is a
// decision not to dispose while Silk's render-loop guard is
// still armed. Never synthesize a native failure that didn't
// happen; the transaction's own error stays the sole reported
// cause, exactly as if the native step had been skipped.
Exception terminalError = nativeFailure is null
? _report.Error ?? new InvalidOperationException(
"Shutdown did not converge before native fallback.")
@ -256,8 +281,16 @@ internal sealed class GameWindowLifetime
if (_report.IsTerminal && !_nativeReleaseAttempted)
{
Exception? nativeFailure = ReleaseNativeWindow();
if (nativeFailure is not null)
(NativeReleaseOutcome outcome, Exception? nativeFailure) = ReleaseNativeWindow();
if (outcome == NativeReleaseOutcome.Deferred)
{
_report = new GameWindowLifetimeReport(
GameWindowLifetimeStatus.CompleteWithDeferredNativeRelease,
"native window",
_report.CleanupFailures,
null);
}
else if (nativeFailure is not null)
{
_report = new GameWindowLifetimeReport(
GameWindowLifetimeStatus.AbandonedIncomplete,
@ -284,20 +317,54 @@ internal sealed class GameWindowLifetime
"Shutdown roots must publish before completion starts."));
}
private Exception? ReleaseNativeWindow()
private enum NativeReleaseOutcome
{
Released,
Deferred,
Failed,
}
private (NativeReleaseOutcome Outcome, Exception? Error) ReleaseNativeWindow()
{
if (_nativeReleaseAttempted)
return null;
return (NativeReleaseOutcome.Released, null);
if (_isNativeRenderLoopArmed?.Invoke() == true)
{
// #343: Silk's ViewImplementationBase.Reset() (called from its
// Dispose()) throws "You cannot call `Reset` inside of the
// render loop!" when its internal render-loop guard is still
// set — which happens whenever a render/update frame callback
// threw without reaching its own reset line. Calling Dispose()
// here would trade the exception that actually wounded the loop
// for that unrelated Silk guard exception. Request a close as a
// courtesy and defer the physical release instead of disposing
// into that throw; this is a one-shot decision for this native
// window, matching every other release outcome below.
try
{
_requestNativeClose?.Invoke();
}
catch
{
// Best-effort only — the deferred outcome does not depend on
// this succeeding, and the native window's own exception
// must never become the reported failure here.
}
_nativeReleaseAttempted = true;
return (NativeReleaseOutcome.Deferred, null);
}
_nativeReleaseAttempted = true;
try
{
_nativeWindow?.Dispose();
_nativeWindow = null;
return null;
return (NativeReleaseOutcome.Released, null);
}
catch (Exception error)
{
return error;
return (NativeReleaseOutcome.Failed, error);
}
}

View file

@ -32,6 +32,81 @@ public sealed class GameWindowLifetimeTests
Assert.False(lifetime.RetainsShutdownGraph);
}
[Fact]
public void ArmedRenderLoopDefersNativeReleaseInsteadOfDisposing()
{
// #343: a render/update frame callback that threw leaves Silk's
// internal render-loop guard stuck armed. Disposing the native
// window into that state throws "You cannot call `Reset` inside of
// the render loop!" and buries whatever exception actually wounded
// the loop. GameWindowLifetime must detect the still-armed signal,
// request a close instead of disposing, and record a distinct
// deferred status rather than AbandonedIncomplete-with-throw.
var calls = new List<string>();
var transaction = new ResourceShutdownTransaction(
new ResourceShutdownStage("owners",
[
new("owner", () => calls.Add("owner")),
]));
var lifetime = new GameWindowLifetime(() => transaction);
var native = new RecordingDisposable("native", calls);
bool closeRequested = false;
lifetime.PublishNativeWindow(
native,
isRenderLoopArmed: () => true,
requestClose: () => closeRequested = true);
GameWindowLifetimeReport report = lifetime.CompleteAndReleaseNativeWindow();
Assert.Equal(GameWindowLifetimeStatus.CompleteWithDeferredNativeRelease, report.Status);
Assert.Equal("native window", report.BlockedStage);
Assert.Null(report.Error);
Assert.Empty(report.CleanupFailures);
Assert.Equal(["owner"], calls);
Assert.Equal(0, native.DisposeCalls);
Assert.True(closeRequested);
Assert.True(lifetime.RetainsShutdownGraph);
Assert.True(report.IsTerminal);
// Idempotent: the armed predicate never flips back false in this
// test double, so a repeated call must keep returning the same
// deferred report without a second Dispose attempt.
GameWindowLifetimeReport repeated = lifetime.CompleteAndReleaseNativeWindow();
Assert.Same(report, repeated);
Assert.Equal(0, native.DisposeCalls);
}
[Fact]
public void UnarmedRenderLoopStillDisposesNativeWindowNormally()
{
// The healthy post-loop path (Run()'s tail, GameWindow.cs) always
// completes with the loop unarmed. This pins that the new optional
// isRenderLoopArmed/requestClose parameters do not change that path:
// Dispose still runs exactly once, in the same order as before.
var calls = new List<string>();
var transaction = new ResourceShutdownTransaction(
new ResourceShutdownStage("owners",
[
new("owner", () => calls.Add("owner")),
]));
var lifetime = new GameWindowLifetime(() => transaction);
var native = new RecordingDisposable("native", calls);
bool closeRequested = false;
lifetime.PublishNativeWindow(
native,
isRenderLoopArmed: () => false,
requestClose: () => closeRequested = true);
GameWindowLifetimeReport report = lifetime.CompleteAndReleaseNativeWindow();
Assert.Equal(GameWindowLifetimeStatus.Complete, report.Status);
Assert.Null(report.BlockedStage);
Assert.Equal(["owner", "native"], calls);
Assert.Equal(1, native.DisposeCalls);
Assert.False(closeRequested);
Assert.False(lifetime.RetainsShutdownGraph);
}
[Fact]
public void HardBarrierFailureRetriesWithoutReplayThenCompletes()
{