acdream/src/AcDream.App/Rendering/Wb/EntityClassificationCache.cs
Erik 565c351f93 feat(render): Campaign V slice V4t-2 — the world texture stack crosses to GpuTextureSlot
The rest of V4t. The composite, particle and shared-atlas texture paths now
hand out the device's GpuTextureSlot instead of a raw 64-bit
ARB_bindless_texture handle, and GroupKey, CachedBatch and ObjectRenderBatch
carry that slot. WbDrawDispatcher, EnvCellRenderer and ParticleRenderer retire
their interim GlBindlessHandleTable instances and share the device's one
table, exactly as V4t-1 did for terrain. Nothing about world submission
changes otherwise: these three renderers are still raw GL, still bind binding
9 themselves, and still draw the same geometry in the same order.

**What produces a slot now.** CompositeTextureArrayCache's GL backend interns
each array's handle when it makes it resident and retires the entry when it
makes it non-resident, so the pair is created and destroyed together and the
cache above it never learns a device exists — the fake backend its tests use
mints a stand-in slot. TextureCache.AcquireParticleTexture does the same for
the one-layer particle arrays it owns, including on its rollback path.
ObjectMeshManager registers each shared atlas's wrap/clamp handles at batch
upload; registration is idempotent by handle, so the many batches sharing an
atlas share its entry.

**Slot release is stricter than what it replaces, not looser.** The interim
tables never released anything — the class comment said so — and they grew
without bound. The device's table has a fixed 16,384-slot capacity, so an
unreleased entry is now a leak with an end. Every producer therefore retires
its entry: the composite backend at MakeNonResident, the particle backend at
MakeNonResident, and ObjectMeshManager when a retiring atlas's PHYSICAL
retirement completes — the point at which its handles are already non-resident
and its texture already deleted. That last one needs the handles snapshotted
at eviction, because ManagedGLTextureArray.Dispose zeroes its own copies as
its first act. Teardown deliberately does not release: the device is being torn
down alongside its callers, so there is nothing left to recycle a slot into,
and deferring work through a possibly-disposed retirement queue would turn a
clean shutdown into a throw.

**The default value became load-bearing, and that is the one real hazard here.**
BindlessTextureLocation could say "not resolved" with handle 0, because no
texture has handle 0. A slot index has no spare value — default(GpuTextureSlot)
is real slot 0 — so a positional record would have turned every
budget-rejected or still-uploading composite into a silent read of whichever
texture registered first. That is the magenta-placeholder failure shape one
layer down. The type is now a struct storing the slot one-based, so default IS
Unresolved, with a test pinning both halves: default is unresolved, and a
location naming slot 0 is resolved and distinguishable from it. Elsewhere the
sentinel is already exact — GpuTextureSlot.Unassigned is 0xFFFFFFFF, which is
common.glsl's ACDREAM_TEXTURE_NONE — so the classify path's "no texture yet"
test and the particle billboard's untextured branch are unchanged in meaning.

**GroupKey ordering is preserved because the key never ordered anything.**
Handle→slot is a bijection (the device interns one slot per resident handle),
so the same (entity, batch) pairs bucket together as before. The key reaches
equality, hashing and the scene-digest fingerprints — never a comparator:
opaque and translucent groups sort by cull mode then camera distance, the
delayed-alpha path by viewer distance then submission ordinal, and group
enumeration follows the persistent dictionary's insertion order, which a
changed hash does not disturb. The digests hash the slot index where they
hashed the handle; both sides of the render-shadow comparison compute them the
same way, so the value changing is invisible to it. Read
CompareOpaqueSubmissionOrder, CompareTransparentSubmissionOrder and
AlphaFingerprintComparer before doubting this — sort-order drift is a
pixel-visible regression class this project has hit, and it is why the check
was made before the retype rather than after.

**One visibility change, forced rather than chosen.** BindlessTextureLocation
was public and now holds an internal contract type, so it is internal;
ObjectRenderBatch.TextureSlot is internal on an otherwise public class for the
same reason. Nothing outside this assembly and its InternalsVisibleTo test
assemblies named either.

**SkyRenderer keeps its interim table**, and the report should say why: the
sky's textures are minted by SkyRenderer itself from TextureCache's raw GL
texture names, which this slice does not retype, so it would be the one
consumer registering handles it produced — a different shape from the world
stack. The offline gate also masks the sky band, so the one automated
instrument here cannot see a sky regression. V4f owns that renderer.

**Gates.** GL offline pixel gate vs cb2a70b8, measured twice: 31 and 22
differing pixels of 563,200 (5.50e-05, 3.91e-05). The first is above the
plan's documented 15-23 px band, so a control was measured rather than
assumed: two same-commit captures at this tree differ by 19 px, and — the
decisive number — a capture at V4t-1 and a capture at this commit differ by
9 px, fewer than the same-commit control. Maximum channel delta is 41-52 in
every pair including the controls, i.e. the differing pixels are drawn from
one flickering population, not from moved geometry. tools/run-repeat-connected-gate.ps1
-Runs 3: 3/3 RENDERED on both the desktop witness and the client capture. One
Vulkan composition-host run with VK_LAYER_KHRONOS_validation proven inserted
by the loader: zero errors, zero warnings, converged ownership ledger. App
tests 4,077 / 3 skips and the complete Release suite 9,140 / 5 — both the
4,075 and 9,138 baselines plus the two tests added here.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-28 12:43:50 +02:00

204 lines
9.4 KiB
C#

using System.Collections.Generic;
namespace AcDream.App.Rendering.Wb;
/// <summary>
/// Cache of per-entity classification results for static entities (those NOT
/// in <c>GameWindow._animatedEntities</c>). Holds one
/// <see cref="EntityCacheEntry"/> per cached entity. The cache is opaque
/// w.r.t. classification logic — it simply stores what callers populate.
///
/// <para>
/// <b>Key composition:</b> entries are keyed by the tuple
/// <c>(EntityId, LandblockHint)</c>, NOT by <c>EntityId</c> alone. Issue #53
/// uncovered that older hydration IDs were not globally unique across all
/// static-entity paths: their former byte-aligned counters overflowed at
/// more than 256 items per landblock and wrapped into the <c>lbY</c> byte.
/// The current <c>0x8XXYYIII</c> scenery and <c>0x4XXYYIII</c> interior
/// allocators reserve 12-bit counters and fail before aliasing. Keying
/// by the tuple is correct-by-construction ONLY when the hint identifies the
/// entity's OWNING landblock — callers must derive it via
/// <c>WbDrawDispatcher.ResolveCacheLandblockHint</c> (the entity's
/// ParentCellId landblock when present, canonicalized <c>0xXXYYFFFF</c>),
/// never a call-site landblock. The #119 "broken stairs + water barrel" was
/// exactly this: the bucket draw path hinted every entity with the PLAYER's
/// landblock, so colliding ids from different landblocks shared a key and
/// served each other's batches.
/// </para>
///
/// <para>
/// <b>Invariants:</b>
/// <list type="bullet">
/// <item><see cref="Populate"/> overwrites any existing entry for the same (id, lb) tuple (defensive).</item>
/// <item><see cref="InvalidateEntity"/> sweeps all entries with the given <c>EntityId</c>
/// regardless of <c>LandblockHint</c>; idempotent (no-throw on missing id).</item>
/// <item><see cref="InvalidateLandblock"/> walks all entries; entries whose
/// <see cref="EntityCacheEntry.LandblockHint"/> equals the argument are removed.</item>
/// <item>All operations are render-thread only. No internal locking.</item>
/// </list>
/// </para>
///
/// <para>
/// <b>Audit foundation:</b> see
/// <c>docs/research/2026-05-10-tier1-mutation-audit.md</c> for why static
/// entities can be cached and what invalidation is needed.
/// </para>
///
/// <para>
/// <b>Accessibility:</b> <c>internal</c>. <see cref="EntityCacheEntry"/> and
/// <see cref="CachedBatch"/> both transitively reference the <c>internal</c>
/// <see cref="GroupKey"/>; surfacing the cache as <c>public</c> would create
/// inconsistent-accessibility errors. Cross-assembly access for the test
/// project comes via <c>InternalsVisibleTo("AcDream.Core.Tests")</c> on
/// <c>AcDream.App.csproj</c>.
/// </para>
/// </summary>
internal sealed class EntityClassificationCache
{
private readonly Dictionary<(uint EntityId, uint LandblockHint), EntityCacheEntry> _entries = new();
/// <summary>Number of cached entities — for diagnostics.</summary>
public int Count => _entries.Count;
/// <summary>
/// Look up an entity's cached classification. Keyed by both
/// <paramref name="entityId"/> AND <paramref name="landblockHint"/> to
/// preserve defensive isolation if a future hydration path introduces an
/// entity-id collision across landblocks.
/// Returns <c>true</c> with the entry on hit; <c>false</c> with
/// <paramref name="entry"/> set to <c>null</c> on miss.
/// </summary>
public bool TryGet(uint entityId, uint landblockHint, out EntityCacheEntry? entry)
=> _entries.TryGetValue((entityId, landblockHint), out entry);
/// <summary>
/// Insert or overwrite a cache entry for the
/// <c>(<paramref name="entityId"/>, <paramref name="landblockHint"/>)</c>
/// tuple. Defensive: if an entry already exists, replaces it.
/// </summary>
public void Populate(
uint entityId,
uint landblockHint,
CachedBatch[] batches,
CachedSelectionPart[]? selectionParts = null)
{
_entries[(entityId, landblockHint)] = new EntityCacheEntry
{
EntityId = entityId,
LandblockHint = landblockHint,
Batches = batches,
SelectionParts = selectionParts ?? [],
};
}
/// <summary>
/// Remove all cache entries for the given <paramref name="entityId"/>,
/// regardless of which landblock they were populated under. Sweep is
/// needed because we may have entries for the same Id under different
/// LandblockHints if any hydration path produced colliding Ids
/// historically (defensive even though current paths shouldn't produce
/// duplicates per-LB). Was <c>O(1)</c> before the #53 tuple-key change;
/// now <c>O(n)</c>, but called rarely (only on entity despawn).
/// </summary>
public void InvalidateEntity(uint entityId)
{
if (_entries.Count == 0) return;
List<(uint, uint)>? toRemove = null;
foreach (var key in _entries.Keys)
{
if (key.EntityId == entityId)
{
toRemove ??= new List<(uint, uint)>();
toRemove.Add(key);
}
}
if (toRemove is null) return;
foreach (var k in toRemove) _entries.Remove(k);
}
/// <summary>
/// Remove every cache entry whose <see cref="EntityCacheEntry.LandblockHint"/>
/// equals <paramref name="landblockId"/>. Used by the streaming pipeline
/// when a landblock demotes from near to far or unloads. No-op if no
/// entries match.
/// </summary>
public void InvalidateLandblock(uint landblockId)
{
if (_entries.Count == 0) return;
// Collect the keys to remove first to avoid mutating the dict during iteration.
// Buffered locally because the typical case removes ~all entries in the LB
// (which is still small relative to the total cache).
List<(uint, uint)>? toRemove = null;
foreach (var key in _entries.Keys)
{
if (key.LandblockHint == landblockId)
{
toRemove ??= new List<(uint, uint)>();
toRemove.Add(key);
}
}
if (toRemove is null) return;
foreach (var k in toRemove) _entries.Remove(k);
}
#if DEBUG
/// <summary>
/// Asserts that the cached entry for <paramref name="entityId"/> still
/// matches what fresh classification would produce. Catches the prior
/// Tier 1 bug class — silent caching of mutable per-frame state — by
/// firing <see cref="System.Diagnostics.Debug.Assert"/> when any cached
/// field has drifted from live state.
///
/// <para>
/// Caller passes per-batch live state (Key, TextureSlot, RestPose)
/// reconstructed from the same path the populate ran. The cache iterates
/// its stored entries in parallel and asserts equality.
/// </para>
///
/// <para>
/// As of Phase 4 (commit f16604b) this method is exercised by unit tests
/// only; the dispatcher's cache-hit branch fires a simpler predicate assert
/// (<c>!isAnimated</c>) at production hit time. Wiring the full live-state
/// cross-check into the per-entity branch is the spec section 6.5 stretch
/// goal and remains open as a follow-up. Zero cost in Release; the method
/// stays here so the regression-class guard is locked behind tests.
/// </para>
/// </summary>
public void DebugCrossCheck(uint entityId, uint landblockHint, IReadOnlyList<CachedBatch> liveBatches)
{
if (!_entries.TryGetValue((entityId, landblockHint), out var entry)) return;
System.Diagnostics.Debug.Assert(
entry.Batches.Length == liveBatches.Count,
$"EntityClassificationCache: batch count mismatch for entity {entityId}: cached={entry.Batches.Length} live={liveBatches.Count}");
for (int i = 0; i < entry.Batches.Length && i < liveBatches.Count; i++)
{
var cached = entry.Batches[i];
var live = liveBatches[i];
System.Diagnostics.Debug.Assert(
cached.Key.Equals(live.Key),
$"EntityClassificationCache: GroupKey drift for entity {entityId} batch {i}");
System.Diagnostics.Debug.Assert(
cached.TextureSlot == live.TextureSlot,
$"EntityClassificationCache: texture slot drift for entity {entityId} batch {i}");
System.Diagnostics.Debug.Assert(
MatrixApproxEqual(cached.RestPose, live.RestPose, epsilon: 1e-5f),
$"EntityClassificationCache: RestPose drift for entity {entityId} batch {i}");
}
}
private static bool MatrixApproxEqual(System.Numerics.Matrix4x4 a, System.Numerics.Matrix4x4 b, float epsilon)
{
return System.MathF.Abs(a.M11 - b.M11) <= epsilon && System.MathF.Abs(a.M12 - b.M12) <= epsilon &&
System.MathF.Abs(a.M13 - b.M13) <= epsilon && System.MathF.Abs(a.M14 - b.M14) <= epsilon &&
System.MathF.Abs(a.M21 - b.M21) <= epsilon && System.MathF.Abs(a.M22 - b.M22) <= epsilon &&
System.MathF.Abs(a.M23 - b.M23) <= epsilon && System.MathF.Abs(a.M24 - b.M24) <= epsilon &&
System.MathF.Abs(a.M31 - b.M31) <= epsilon && System.MathF.Abs(a.M32 - b.M32) <= epsilon &&
System.MathF.Abs(a.M33 - b.M33) <= epsilon && System.MathF.Abs(a.M34 - b.M34) <= epsilon &&
System.MathF.Abs(a.M41 - b.M41) <= epsilon && System.MathF.Abs(a.M42 - b.M42) <= epsilon &&
System.MathF.Abs(a.M43 - b.M43) <= epsilon && System.MathF.Abs(a.M44 - b.M44) <= epsilon;
}
#endif
}