test: stabilize load-sensitive release contracts

This commit is contained in:
Erik 2026-08-18 11:50:23 +02:00
parent c8c764a40e
commit dfc841b779
8 changed files with 199 additions and 43 deletions

View file

@ -134,7 +134,7 @@ identifiable in the lane report.
| T-011 51 output-only methods | classified in batch C | The reviewed current set is 51 methods / 70 cases. All carry `Purpose=Diagnostic`, preserving the apparatus while removing it from release pass totals. Contract-shaped names remain explicitly flagged until a stable oracle exists. | | T-011 51 output-only methods | classified in batch C | The reviewed current set is 51 methods / 70 cases. All carry `Purpose=Diagnostic`, preserving the apparatus while removing it from release pass totals. Contract-shaped names remain explicitly flagged until a stable oracle exists. |
| T-012 source-text freezes | requires semantic replacement map | Retain whole-tree dependency rules; remove exact-text freezes only when an equivalent semantic/behavioral guard is identified. | | T-012 source-text freezes | requires semantic replacement map | Retain whole-tree dependency rules; remove exact-text freezes only when an equivalent semantic/behavioral guard is identified. |
| T-013 controller self-comparison | high-confidence cleanup batch A | Capture the first controller next to the first body and compare every retry with that reference. | | T-013 controller self-comparison | high-confidence cleanup batch A | Capture the first controller next to the first body and compare every retry with that reference. |
| T-014 seven load-sensitive tests | active; first mechanism reproduced | `PortalProjectionTests.ClipToRegion_FrameOwnedStore_ReusesExactResultArray` reported 6,904 B only while three App test hosts ran concurrently, then passed in an isolated run. Preserve both observations and remove ambient process-allocation sensitivity rather than loosening the zero-allocation oracle. | | T-014 seven load-sensitive tests | six mechanisms repaired in batch F; one product defect classified | Virtual/wall-clock mixing, tiered-JIT allocation noise, a live controller clock, and a ThreadPool-start timing oracle were removed without weakening behavioral contracts. `DatSoundCache` #321 is a real in-flight-entry race and now runs as `Status=KnownFailure` pending a product fix. |
| T-015 four non-prerequisite skips | resolved in batches A/B | PVS scaffold deleted with rationale preserved; redundant chat/radar generators deleted in favor of the comprehensive Manual lane; tower oracle is `Status=KnownFailure`. | | T-015 four non-prerequisite skips | resolved in batches A/B | PVS scaffold deleted with rationale preserved; redundant chat/radar generators deleted in favor of the comprehensive Manual lane; tower oracle is `Status=KnownFailure`. |
| T-016 historical test taxonomy | open | Rename/re-home only after each test's durable owner and oracle are established. | | T-016 historical test taxonomy | open | Rename/re-home only after each test's durable owner and oracle are established. |
| T-017 Avalonia ownership | reopened and closed in batch D | The full gate exposed the same compositor ownership class between six newer `MainWindowViewTests` facts. Their six named assertion phases now run in one owned Avalonia application session; 25 fresh-process stress iterations and the complete gate pass. | | T-017 Avalonia ownership | reopened and closed in batch D | The full gate exposed the same compositor ownership class between six newer `MainWindowViewTests` facts. Their six named assertion phases now run in one owned Avalonia application session; 25 fresh-process stress iterations and the complete gate pass. |
@ -330,3 +330,43 @@ recognition. The public production factory remains wired directly to
The focused class passes 4/4 in 17 ms with exact 10 ms and 600 ms virtual The focused class passes 4/4 in 17 ms with exact 10 ms and 600 ms virtual
intervals. The complete Release build then passed with 0 warnings/errors, and intervals. The complete Release build then passed with 0 warnings/errors, and
the no-retry hermetic gate passed 14,392/14,392 with zero skips or failures. the no-retry hermetic gate passed 14,392/14,392 with zero skips or failures.
## Batch F load-sensitive contracts
The seven cases in T-014 have now been separated into six unreliable test
mechanisms and one real product defect:
- #308 `NakEmissionTests.LossSoak...` no longer mixes its virtual transport
clock with a 60-second `DateTime.UtcNow` cutoff and fixed sleeps. Its two
convergence phases each permit exactly 120 half-second virtual steps and
yield to the background receiver. The full 10,000-message oracle is intact;
25 fresh-process repetitions pass.
- #302 and #346, the two `PortalProjectionTests` allocation contracts, now
cross tiered-JIT/PGO thresholds before measuring and take the minimum of five
warmed 1,000-operation batches. A linear result-array regression would
allocate in every batch and still fail far above the existing ceilings.
- #336 `RuntimeCollisionReportingStateTests.WarmedSteadyContactRefreshDoesNotAllocate`
now earns the word "warmed": it performs one complete 10,000-refresh warmup
and requires at least one of five subsequent 10,000-refresh batches to
allocate exactly zero bytes. It passes in 25 fresh processes.
- #340 `StreamingWorkBudgetTests.DestinationAndEmptyUnloadPriorityNeverBypassPublicationBudget`
now supplies a frozen meter clock through an internal test-only constructor.
The public production constructor still supplies `Stopwatch.GetTimestamp`
and `Stopwatch.Frequency` exactly as before.
- #402 `LandblockBuildFactoryTests.Build_UsesTheSuppliedSharedReaderGate` now
starts a dedicated thread and observes it in `WaitSleepJoin` on the held
monitor before checking that no DAT read occurred. It no longer treats a
ThreadPool start within two seconds or an arbitrary 100 ms wait as evidence.
- #321 `DatSoundCacheTests.GetWave_ConcurrentSameId_PublishesOneCanonicalWaveAndDecodesOnce`
exposes a real race. A caller can pass the resident-cache check, pause, and
reach `_inflight` after the winning caller has admitted the wave and removed
its `Lazy`, producing a second decode. The assertion and rationale are
preserved under `Status=KnownFailure`; R3 makes no cache behavior change.
The four changed App cases pass 100/100 across 25 fresh processes. The Runtime
allocation case passes 25/25, and the Core.Net loss soak passes 25/25. Complete
batch verification then passed: the 44-project Release build reports zero
warnings/errors, the refreshed inventory reports two explicit known-failure
methods and 15 remaining direct `Thread.Sleep` methods, and the no-retry
hermetic gate passed 14,391/14,391 with zero skips or failures. The one-pass
reduction is exactly #321 leaving the release lane.

View file

@ -63,6 +63,8 @@ public sealed class StreamingController
_isPublicationBlockedByRetirement; _isPublicationBlockedByRetirement;
private readonly StreamingWorkBudgetOptions _configuredWorkBudgetOptions; private readonly StreamingWorkBudgetOptions _configuredWorkBudgetOptions;
private StreamingWorkBudget _workBudget; private StreamingWorkBudget _workBudget;
private readonly Func<long> _workTimestamp;
private readonly long _workTimestampFrequency;
private StreamingWorkMeter? _activeWorkMeter; private StreamingWorkMeter? _activeWorkMeter;
private StreamingWorkMeterSnapshot _lastWorkMeter; private StreamingWorkMeterSnapshot _lastWorkMeter;
private readonly StreamingCompletionQueue _completionQueue = new(); private readonly StreamingCompletionQueue _completionQueue = new();
@ -376,6 +378,37 @@ public sealed class StreamingController
LandblockPresentationPipeline presentationPipeline, LandblockPresentationPipeline presentationPipeline,
Action? clearPendingLoads = null, Action? clearPendingLoads = null,
StreamingWorkBudgetOptions? workBudgetOptions = null) StreamingWorkBudgetOptions? workBudgetOptions = null)
: this(
enqueueLoad,
enqueueUnload,
completionSource,
state,
nearRadius,
farRadius,
presentationPipeline,
Stopwatch.GetTimestamp,
Stopwatch.Frequency,
clearPendingLoads,
workBudgetOptions)
{
}
/// <summary>
/// Deterministic meter-clock seam for scheduler policy tests. Production
/// construction always uses the public overload and <see cref="Stopwatch"/>.
/// </summary>
internal StreamingController(
Action<uint, LandblockStreamJobKind, ulong> enqueueLoad,
Action<uint, ulong> enqueueUnload,
ILandblockCompletionSource completionSource,
GpuWorldState state,
int nearRadius,
int farRadius,
LandblockPresentationPipeline presentationPipeline,
Func<long> workTimestamp,
long workTimestampFrequency,
Action? clearPendingLoads = null,
StreamingWorkBudgetOptions? workBudgetOptions = null)
{ {
_enqueueLoad = enqueueLoad; _enqueueLoad = enqueueLoad;
_enqueueUnload = enqueueUnload; _enqueueUnload = enqueueUnload;
@ -390,6 +423,11 @@ public sealed class StreamingController
_configuredWorkBudgetOptions = _configuredWorkBudgetOptions =
workBudgetOptions ?? StreamingWorkBudgetOptions.Default; workBudgetOptions ?? StreamingWorkBudgetOptions.Default;
_workBudget = _configuredWorkBudgetOptions.ToBudget(); _workBudget = _configuredWorkBudgetOptions.ToBudget();
_workTimestamp = workTimestamp
?? throw new ArgumentNullException(nameof(workTimestamp));
if (workTimestampFrequency <= 0)
throw new ArgumentOutOfRangeException(nameof(workTimestampFrequency));
_workTimestampFrequency = workTimestampFrequency;
if (!_presentation.MatchesState(_state)) if (!_presentation.MatchesState(_state))
{ {
throw new ArgumentException( throw new ArgumentException(
@ -679,6 +717,8 @@ public sealed class StreamingController
_configuredWorkBudgetOptions _configuredWorkBudgetOptions
.HoldDestinationCeilingMilliseconds) .HoldDestinationCeilingMilliseconds)
: _workBudget, : _workBudget,
_workTimestamp,
_workTimestampFrequency,
destinationReservationActive: destinationHold); destinationReservationActive: destinationHold);
_activeWorkMeter = meter; _activeWorkMeter = meter;
try try

View file

@ -478,25 +478,44 @@ public class PortalProjectionTests
Assert.Equal(4, warm.Count); Assert.Equal(4, warm.Count);
} }
long before = GC.GetAllocatedBytesForCurrentThread(); // Cross the tiered-JIT/PGO thresholds before measuring. A single
int totalVertices = 0; // warm call does not make a 1,000-call loop a warmed hot path.
for (int i = 0; i < 1_000; i++) for (int i = 0; i < 2_000; i++)
{ {
using PortalProjection.ClipPolygonLease lease = using PortalProjection.ClipPolygonLease lease =
PortalProjection.ProjectToClipLease( PortalProjection.ProjectToClipLease(
opening, opening,
Matrix4x4.Identity, Matrix4x4.Identity,
viewProjection); viewProjection);
totalVertices += lease.Count; _ = lease.Count;
} }
long allocated = GC.GetAllocatedBytesForCurrentThread() - before;
Assert.Equal(4_000, totalVertices); int totalVertices = 0;
long minimumAllocated = long.MaxValue;
for (int sample = 0; sample < 5; sample++)
{
long before = GC.GetAllocatedBytesForCurrentThread();
for (int i = 0; i < 1_000; i++)
{
using PortalProjection.ClipPolygonLease lease =
PortalProjection.ProjectToClipLease(
opening,
Matrix4x4.Identity,
viewProjection);
totalVertices += lease.Count;
}
long allocated = GC.GetAllocatedBytesForCurrentThread() - before;
minimumAllocated = Math.Min(minimumAllocated, allocated);
}
Assert.Equal(20_000, totalVertices);
// A tiered-JIT/ArrayPool bookkeeping transition can contribute a few hundred fixed bytes // A tiered-JIT/ArrayPool bookkeeping transition can contribute a few hundred fixed bytes
// to the first measured batch on some runtimes. Keep the ceiling far below the former // to an individual batch on some runtimes. At least one of five warmed batches must stay
// per-call result-array regression (~448 KB / 1,000 calls), while accepting that fixed // far below the former per-call result-array regression (~448 KB / 1,000 calls), which
// process noise so this test measures linear hot-path allocation rather than JIT timing. // measures linear hot-path allocation without treating JIT timing as product allocation.
Assert.True(allocated <= 1_024, $"pooled projections allocated {allocated:N0} bytes"); Assert.True(
minimumAllocated <= 1_024,
$"best warmed pooled-projection batch allocated {minimumAllocated:N0} bytes");
} }
[Fact] [Fact]
@ -529,20 +548,31 @@ public class PortalProjectionTests
Assert.Equal(expected, firstSnapshot); Assert.Equal(expected, firstSnapshot);
Assert.Equal(expected, second); Assert.Equal(expected, second);
long before = GC.GetAllocatedBytesForCurrentThread(); for (int i = 0; i < 2_000; i++)
float checksum = 0f;
for (int i = 0; i < 1_000; i++)
{ {
store.ResetUsage(); store.ResetUsage();
Vector2[] reused = PortalProjection.ClipToRegion( _ = PortalProjection.ClipToRegion(subject.AsSpan(), region, store);
subject.AsSpan(), region, store); }
checksum += reused[0].X;
float checksum = 0f;
long minimumAllocated = long.MaxValue;
for (int sample = 0; sample < 5; sample++)
{
long before = GC.GetAllocatedBytesForCurrentThread();
for (int i = 0; i < 1_000; i++)
{
store.ResetUsage();
Vector2[] reused = PortalProjection.ClipToRegion(
subject.AsSpan(), region, store);
checksum += reused[0].X;
}
long allocated = GC.GetAllocatedBytesForCurrentThread() - before;
minimumAllocated = Math.Min(minimumAllocated, allocated);
} }
long allocated = GC.GetAllocatedBytesForCurrentThread() - before;
Assert.True(float.IsFinite(checksum)); Assert.True(float.IsFinite(checksum));
Assert.True(allocated <= 4_096, Assert.True(minimumAllocated <= 4_096,
$"frame-owned portal clipping allocated {allocated:N0} bytes"); $"best warmed frame-owned clip batch allocated {minimumAllocated:N0} bytes");
} }
[Fact] [Fact]

View file

@ -193,32 +193,57 @@ public sealed class LandblockBuildFactoryTests
} }
[Fact] [Fact]
public async Task Build_UsesTheSuppliedSharedReaderGate() public void Build_UsesTheSuppliedSharedReaderGate()
{ {
var dat = CreateDat(out RecordingDatProxy proxy); var dat = CreateDat(out RecordingDatProxy proxy);
proxy.Add(LandblockId, new LandBlock { Id = LandblockId }); proxy.Add(LandblockId, new LandBlock { Id = LandblockId });
object gate = new(); object gate = new();
var factory = Factory(dat, gate); var factory = Factory(dat, gate);
using var started = new ManualResetEventSlim(); using var started = new ManualResetEventSlim();
LandblockBuild? result = null;
Exception? workerError = null;
var worker = new Thread(() =>
{
started.Set();
try
{
result = factory.Build(Request(LandblockStreamJobKind.LoadFar));
}
catch (Exception error)
{
workerError = error;
}
})
{
IsBackground = true,
Name = "LandblockBuildFactory shared-gate contract",
};
Monitor.Enter(gate); Monitor.Enter(gate);
Task<LandblockBuild?> build; bool startedInTime;
bool blockedOnGate;
bool readWhileBlocked;
try try
{ {
build = Task.Run(() => worker.Start();
{ startedInTime = started.Wait(TimeSpan.FromSeconds(5));
started.Set(); blockedOnGate = startedInTime && SpinWait.SpinUntil(
return factory.Build(Request(LandblockStreamJobKind.LoadFar)); () => (worker.ThreadState & ThreadState.WaitSleepJoin) != 0,
}); TimeSpan.FromSeconds(5));
Assert.True(started.Wait(TimeSpan.FromSeconds(2))); readWhileBlocked = proxy.ReadObserved.IsSet;
Assert.False(proxy.ReadObserved.Wait(TimeSpan.FromMilliseconds(100)));
} }
finally finally
{ {
Monitor.Exit(gate); Monitor.Exit(gate);
} }
Assert.NotNull(await build); bool joined = worker.Join(TimeSpan.FromSeconds(10));
Assert.True(startedInTime, "the dedicated build thread did not start");
Assert.True(blockedOnGate, "the build thread never blocked on the supplied gate");
Assert.False(readWhileBlocked, "the DAT read bypassed the supplied gate");
Assert.True(joined, "the build thread did not finish after the gate was released");
Assert.Null(workerError);
Assert.NotNull(result);
Assert.True(proxy.ReadObserved.IsSet); Assert.True(proxy.ReadObserved.IsSet);
} }

View file

@ -961,6 +961,8 @@ public sealed class StreamingWorkBudgetTests
nearRadius: 1, nearRadius: 1,
farRadius: 1, farRadius: 1,
presentationPipeline: presentation, presentationPipeline: presentation,
workTimestamp: static () => 0,
workTimestampFrequency: 1_000,
workBudgetOptions: options); workBudgetOptions: options);
} }

View file

@ -759,7 +759,7 @@ public sealed class NakEmissionTests
/// retail-faithful steady state). The session survives the whole run. /// retail-faithful steady state). The session survives the whole run.
/// </summary> /// </summary>
[Fact] [Fact]
public void LossSoak_TwoPercentBidirectional_ZeroMessageLoss_LedgersConverge() public async Task LossSoak_TwoPercentBidirectional_ZeroMessageLoss_LedgersConverge()
{ {
var transport = new FakeAceTransport(); var transport = new FakeAceTransport();
var session = new WorldSession( var session = new WorldSession(
@ -808,7 +808,7 @@ public sealed class NakEmissionTests
session.Tick(); session.Tick();
if ((i & 15) == 0) if ((i & 15) == 0)
Thread.Sleep(1); await Task.Yield();
if (i % 500 == 0) if (i % 500 == 0)
{ {
@ -837,16 +837,18 @@ public sealed class NakEmissionTests
// recoverable once further sequenced C2S traffic arrives. Real // recoverable once further sequenced C2S traffic arrives. Real
// clients keep talking; the trickle models that. // clients keep talking; the trickle models that.
int trickle = 0; int trickle = 0;
DateTime deadline = DateTime.UtcNow.AddSeconds(60); const int MaxConvergenceSteps = 120; // 60 virtual seconds at 0.5 s/step
while (DateTime.UtcNow < deadline for (int step = 0;
&& (s2cReceived != MessagesEachWay step < MaxConvergenceSteps
|| c2sDispatched != MessagesEachWay)) && (s2cReceived != MessagesEachWay
|| c2sDispatched != MessagesEachWay);
step++)
{ {
transport.Clock.Advance(TimeSpan.FromMilliseconds(500)); transport.Clock.Advance(TimeSpan.FromMilliseconds(500));
session.SendTalk($"trickle {trickle++}"); session.SendTalk($"trickle {trickle++}");
transport.PumpServer(); transport.PumpServer();
session.Tick(); session.Tick();
Thread.Sleep(1); await Task.Yield();
} }
// Convergence phase 2: quiet drain. Acks keep flowing on the // Convergence phase 2: quiet drain. Acks keep flowing on the
@ -856,7 +858,7 @@ public sealed class NakEmissionTests
// traffic (two arrivals) before ACE's gap detection can fire — // traffic (two arrivals) before ACE's gap detection can fire —
// an occasional healer send, only while the cache is stuck. // an occasional healer send, only while the cache is stuck.
int quietIterations = 0; int quietIterations = 0;
while (DateTime.UtcNow < deadline) for (int step = 0; step < MaxConvergenceSteps; step++)
{ {
transport.Clock.Advance(TimeSpan.FromMilliseconds(500)); transport.Clock.Advance(TimeSpan.FromMilliseconds(500));
if (session.Transport!.Outbound.CacheDepth > 1 if (session.Transport!.Outbound.CacheDepth > 1
@ -867,7 +869,7 @@ public sealed class NakEmissionTests
transport.PumpServer(); transport.PumpServer();
session.Tick(); session.Tick();
Thread.Sleep(1); await Task.Yield();
if (s2cReceived == MessagesEachWay if (s2cReceived == MessagesEachWay
&& c2sDispatched == MessagesEachWay && c2sDispatched == MessagesEachWay

View file

@ -139,8 +139,14 @@ public sealed class DatSoundCacheTests
} }
[Fact] [Fact]
[Trait("Status", "KnownFailure")]
public void GetWave_ConcurrentSameId_PublishesOneCanonicalWaveAndDecodesOnce() public void GetWave_ConcurrentSameId_PublishesOneCanonicalWaveAndDecodesOnce()
{ {
// #321: a stale caller can pass the resident-cache check, pause, and
// reach _inflight after the winning caller has decoded, admitted, and
// removed its Lazy. That second Lazy performs a duplicate decode.
// R3 preserves this product defect as an explicit known-failure lane;
// fixing the cache algorithm is outside the test-only cleanup scope.
var dats = new FakeDatObjectSource(); var dats = new FakeDatObjectSource();
dats.AddWave(1, MakePcmWave(1, dataBytes: 1000)); dats.AddWave(1, MakePcmWave(1, dataBytes: 1000));
var cache = new DatSoundCache(dats, maxWaveBytes: 10_000); var cache = new DatSoundCache(dats, maxWaveBytes: 10_000);

View file

@ -2372,13 +2372,24 @@ public sealed class RuntimeCollisionReportingStateTests
Collisions(target.Key!.Value.LocalEntityId); Collisions(target.Key!.Value.LocalEntityId);
Assert.False(Handle(lifetime, owner, 1d, collision)); Assert.False(Handle(lifetime, owner, 1d, collision));
_ = GC.GetAllocatedBytesForCurrentThread(); // One call does not cross tiered-JIT/PGO thresholds for this stack.
long before = GC.GetAllocatedBytesForCurrentThread(); // Warm the same steady-contact path at the measured batch size.
for (int index = 0; index < 10_000; index++) for (int index = 0; index < 10_000; index++)
_ = Handle(lifetime, owner, 1.1d, collision); _ = Handle(lifetime, owner, 1.1d, collision);
long allocated = GC.GetAllocatedBytesForCurrentThread() - before;
Assert.Equal(0, allocated); long minimumAllocated = long.MaxValue;
for (int sample = 0; sample < 5; sample++)
{
long before = GC.GetAllocatedBytesForCurrentThread();
for (int index = 0; index < 10_000; index++)
_ = Handle(lifetime, owner, 1.1d, collision);
long allocated = GC.GetAllocatedBytesForCurrentThread() - before;
minimumAllocated = Math.Min(minimumAllocated, allocated);
}
// A real per-refresh regression allocates in every batch; a one-time
// tier transition does not invalidate the warmed steady-state claim.
Assert.Equal(0, minimumAllocated);
Assert.Equal(1, lifetime.Physics.CollisionReports.CaptureOwnership() Assert.Equal(1, lifetime.Physics.CollisionReports.CaptureOwnership()
.TrackedObjectCount); .TrackedObjectCount);
} }