fix(app): Phase A.1 — make LandblockStreamer synchronous (DatCollection isn't thread-safe)

Second hotfix attempt for the "ball of spikes" terrain corruption.
The previous _datLock fix was insufficient because dat reads happen
from many render-thread code paths I didn't enumerate (animation
tick, OnLiveMotionUpdated, OnLivePositionUpdated, the live spawn
hydration, ApplyLoadedTerrain) and locking each is invasive and
fragile.

DatReaderWriter's DatCollection is fundamentally not thread-safe:
DatBinReader's internal buffer position is shared per-database, so
two concurrent .Get<T> calls corrupt each other's read state. The
ArgumentOutOfRangeException at DatBinReader.ReadBytesInternal in
the failure log is the smoking gun — one read started reading a
LandBlock, another moved the reader's position, the first one
asked for the wrong number of bytes.

Until Phase A.3 introduces a thread-safe dat wrapper (or until we
preload all dats into pure in-memory dictionaries), the streamer
runs synchronously: EnqueueLoad invokes the load delegate inline
on the calling thread and writes the result to the outbox in a
single call. The render-thread DrainCompletions loop picks it up
on the same frame.

API surface unchanged — Channel-based outbox, EnqueueLoad/Unload,
DrainCompletions, Start (now no-op), Dispose all preserved. Move
back to async loading is a single-class change once dat thread
safety lands.

Cost: visible frame hitch when crossing landblock boundaries
(rendering the new landblock is now on the render thread). For
default 5×5 the hitch is one landblock per cardinal step, ~50ms
worst case. Acceptable for the MVP — correctness over hitches.

Updated the off-thread test to assert the new synchronous contract
(loader runs on the calling thread). The other 4 tests still pass
unchanged because their spin-drain pattern works with synchronous
delivery too.

The previous _datLock from commit c991fb2 stays in place as
defensive belt-and-suspenders — it's free in synchronous mode and
keeps the contract documented at every dat-reading entry point.

212 tests green.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Erik 2026-04-11 22:56:19 +02:00
parent c991fb23ce
commit 531c9f9349
2 changed files with 52 additions and 31 deletions

View file

@ -8,22 +8,39 @@ using AcDream.Core.World;
namespace AcDream.App.Streaming;
/// <summary>
/// Background worker that services landblock load and unload requests off
/// the render thread. Loads are executed on a dedicated thread via a
/// caller-supplied delegate (the production instance wraps
/// <see cref="LandblockLoader.Load"/>); completed results are posted to
/// an outbox channel the render thread drains once per OnUpdate.
/// Services landblock load/unload requests by invoking a caller-supplied
/// load delegate (the production instance wraps
/// <see cref="LandblockLoader.Load"/>) and posting results to an outbox
/// the render thread drains once per OnUpdate.
///
/// <para>
/// Unloads are passed through the same channel as a <see cref="LandblockStreamResult.Unloaded"/>
/// record so the render thread can release GPU state on the next drain —
/// the worker never touches GPU resources directly.
/// <b>Currently runs synchronously on the calling thread.</b> The original
/// Phase A.1 design ran loads on a dedicated worker thread, but DatReaderWriter's
/// <c>DatCollection</c> is not thread-safe — concurrent reads from a worker
/// and the render thread (animation tick, live spawn handlers) corrupt
/// internal buffer state and produce half-populated <c>LandBlock.Height[]</c>
/// arrays which render as wildly distorted terrain. Until Phase A.3 introduces
/// a thread-safe dat wrapper, loads are synchronous: <see cref="EnqueueLoad"/>
/// invokes the load delegate inline and writes the result to the outbox in
/// a single call. This causes a frame hitch when crossing landblock
/// boundaries, but the rendering is correct.
/// </para>
///
/// <para>
/// The Channel-based outbox + <see cref="DrainCompletions"/> API is
/// preserved so the move back to async loading is a single-class change
/// when DatCollection thread safety lands.
/// </para>
///
/// <para>
/// Unloads pass through the outbox as <see cref="LandblockStreamResult.Unloaded"/>
/// records so the render thread can release GPU state on the next drain —
/// the streamer never touches GPU resources directly.
/// </para>
///
/// <remarks>
/// Threading: <see cref="EnqueueLoad"/> / <see cref="EnqueueUnload"/> may
/// be called from any thread; <see cref="DrainCompletions"/> must be called
/// from a single consumer thread (the render thread in production).
/// Threading: synchronous mode means all methods must be called from the
/// same thread (the render thread in production).
/// </remarks>
/// </summary>
public sealed class LandblockStreamer : IDisposable
@ -39,7 +56,9 @@ public sealed class LandblockStreamer : IDisposable
private readonly Channel<LandblockStreamJob> _inbox;
private readonly Channel<LandblockStreamResult> _outbox;
private readonly CancellationTokenSource _cancel = new();
#pragma warning disable CS0649 // _worker stays declared for the future async path; unused in synchronous mode.
private Thread? _worker;
#pragma warning restore CS0649
private int _disposed;
public LandblockStreamer(Func<uint, LoadedLandblock?> loadLandblock)
@ -52,34 +71,31 @@ public sealed class LandblockStreamer : IDisposable
}
/// <summary>
/// Start the worker thread. Must be called before enqueueing jobs.
/// Calling twice is a no-op.
/// No-op in synchronous mode. Preserved on the API surface so callers
/// don't need to change when async loading returns in Phase A.3.
/// </summary>
public void Start()
{
if (System.Threading.Volatile.Read(ref _disposed) != 0)
throw new ObjectDisposedException(nameof(LandblockStreamer));
if (_worker is not null) return;
_worker = new Thread(WorkerLoop)
{
IsBackground = true,
Name = "acdream.landblock-streamer",
};
_worker.Start();
// No worker thread in synchronous mode.
}
public void EnqueueLoad(uint landblockId)
{
if (System.Threading.Volatile.Read(ref _disposed) != 0)
throw new ObjectDisposedException(nameof(LandblockStreamer));
_inbox.Writer.TryWrite(new LandblockStreamJob.Load(landblockId));
// Synchronous mode: invoke the load delegate inline. The result lands
// in the outbox and DrainCompletions picks it up later in the same
// (or next) frame.
HandleJob(new LandblockStreamJob.Load(landblockId));
}
public void EnqueueUnload(uint landblockId)
{
if (System.Threading.Volatile.Read(ref _disposed) != 0)
throw new ObjectDisposedException(nameof(LandblockStreamer));
_inbox.Writer.TryWrite(new LandblockStreamJob.Unload(landblockId));
HandleJob(new LandblockStreamJob.Unload(landblockId));
}
/// <summary>