Both #280 review lenses returned FAIL on the same defect, and both were
right. IsRenderNeighborhoodResident's widened outer arm requires
IsRenderReady out to FarRadius, justified by "a Far-tier landblock
registers with an empty mesh set and is therefore render-ready." That held
only for a landblock that ARRIVED as Far. The second, equally first-class
way to be Far tier is a Near->Far DEMOTE:
DemoteLandblock -> EnqueueNearLayerRetirement
-> LandblockRetirementStage.MeshReferences
-> GpuWorldState.ReleaseLandblockMeshReferences
-> LandblockSpawnAdapter.OnLandblockUnloaded => WantsLoaded = false
while DetachNearLayer deliberately keeps the landblock loaded, terrain-mesh
resident, terrain-collision resident and DRAWN. Nothing re-publishes an
already-loaded landblock, so the demoted member satisfied NEITHER arm of
the gate, permanently: wormhole tunnel plus centered "In Portal Space -
Please Wait..." forever, no recovery short of relog.
Reachable by ordinary play. Two consecutive recalls to the same landblock
with walking in between makes ChangesStreamingCenter false, so there is no
origin recenter and the region recentres through the ordinary demote diff.
Also reachable via a mid-hold quality-preset drop -- ironically the exact
scenario ReconcileDestinationReservationRadius was added to support. The
pre-#280 radius-1 gate never touched that band, because nothing inside the
Near ring can demote.
FIX SHAPE. Make the two routes genuinely equivalent rather than teaching
the predicate to tolerate the difference. ReleaseLandblockMeshReferences
becomes "reconcile the registration to the post-retirement tier": after the
release converges, if the landblock is still loaded AND still Far tier,
re-assert the empty registration -- the identical OnLandblockLoaded(lb,
empty) a PublicationKind.Far activation makes. It is empty by construction:
DetachNearLayer retains only live server projections, which the adapter's
atlas-tier filter skips. A full retirement is unaffected (DetachLandblock
clears both _loaded and _tierByLandblock), and a throwing release still
retries because the re-assert is only reached after the adapter converged.
The alternative -- "|| (IsFarTier && IsLoaded)" at the gate -- was
rejected: it fixes one caller while leaving IsRenderReady meaning two
different things, which is precisely how this defect arose. After this
change the predicate reads "drawable at its current tier" for every caller,
with no knowledge of how the landblock got there.
WHY THE TESTS MISSED IT, fixed here too:
- Proof obligation P2 was discharged against RESIDENCY (the FarRadius+2
eviction threshold) rather than against IsRenderReady, the gate's actual
atom. The contract now carries the correction and the restated
obligation: no transition may REVOKE IsRenderReady from a landblock that
stays inside FarRadius.
- WorldRevealDerivedWindowIntegrationTests advertised itself as end-to-end
against the real GpuWorldState but constructed it with no spawn adapter,
so its IsRenderReady degenerated to IsLoaded via the "?? true". The
single most load-bearing predicate in the change was stubbed out by a
null in the test named after it -- the same shape as C5b's D3 and #276's
three settler tests. Every fixture in that file now owns a real
LandblockSpawnAdapter.
- The P1 test's comment described its subject as "a Near-shaped completion
the streaming window has since DEMOTED to Far". It is not; it is a fresh
PublishAsFar, the case that does hold. Corrected, since a future reader
would have taken it as demote coverage.
Four new regression tests, all driving the real GpuWorldState +
LandblockSpawnAdapter + LandblockPresentationPipeline through an actual
demote, and all sabotage-verified in both directions (fail with the
production change reverted, pass with it):
NearToFarDemote_LeavesTheLandblockRenderReadyThroughTheRealPipeline
NearToFarDemote_LeavesTheLandblockRenderReadyUnderBudgetedRetirement
TieredWindow_StaysResidentAfterAnOuterRingDemote
OutdoorReveal_SurvivesAnOuterRingDemoteDuringTheHold
The budgeted variant exists because production composes
LandblockRetirementCoordinator.CreateBudgeted, whose MeshReferences stage
is a separate call site from the legacy pipeline's.
SECONDARY, same commit:
- R-1: ACDREAM_PROBE_REVEAL_RADIUS=0 was parser-accepted and
Runtime-rejected -- it yields far = 0 for an outdoor destination, which
fails invalid-readiness-shape on every acknowledgement, hanging the very
A/B route the probe exists to measure. Parser floor raised to 1, with a
7-case table test.
- R-2: the composite-warmup TRIGGER had silently moved onto the far
window's critical path. Pre-#280 the gate and the composite domain were
the same radius-1 square; #280 widened the gate without widening the
domain, so every composite upload serialised behind the last outer-ring
landblock for no readiness benefit. Warmup now starts once the NEAR
sub-window is published -- trigger scope == domain scope, as before. The
reveal gate is untouched: Evaluate still requires the full window AND
composite readiness.
- AP-150 filed: acdream's RetailWaitCueDelay = 5 s arming is NOT retail's
trigger, and #280's commit message got this wrong on both clauses. Retail
emits the notice unconditionally per tunnel rotation segment, in the else
arm of the segment-expiry test at 0x004D6FCD; segment duration is
RandDouble(0.6, 1.8) s, byte-decoded at 0x004D6FE6. The 5.0 constant at
VA 0x007991B0 is CellManager::CheckPrefetchStatus's prefetch RETRY
cadence and has nothing to do with the cue. acdream's own 0.6/1.8 segment
constants already match retail exactly; only the arming is wrong.
Adopting retail's unconditional emit is filed as #329 rather than folded
in here -- it is a user-visible presentation change and wants the user's
eyes.
- AP-151 filed: the gate is materially STRICTER than retail on the
mesh-build/GPU-upload axis. Retail's LScape::PreFetchCells blocks on DAT
RESIDENCY only -- no geometry construction, no upload; that work is lazy
at draw. acdream requires a DAT read, terrain mesh build, render-thread
upload, spatial commit, collision admission and spawn-adapter activation
per member of a 625-member window, metered at MaxCompletionsPerFrame.
Nothing bounds the hold. This is the OPPOSITE asymmetry from AP-149; both
are live at once, on different axes.
- AD-2's amendment stated the false Far-tier readiness assumption verbatim;
corrected, along with the same error in
claude-memory/reference_two_tier_streaming.md, which now carries an
explicit DO-NOT-RETRY on the special-case-the-predicate shape.
- AP-115 scope-noted (it covers the cue's presentation, not its arming).
- #326's SmartBox::set_mid_radius citation corrected: the entry is
0x00453180; 0x004531D0 is the mid-function re-arm branch.
Blast radius: GpuWorldState, LandblockSpawnAdapter,
WorldRevealReadinessBarrier and StreamingDiagnostics are all App-internal;
AcDream.Headless and AcDream.Runtime reference none of them outside
comments. Headless tests run green as part of the gate below, per C5b's
lesson about surveys that skip the no-window host.
Gates: Release build 0 errors, 18 pre-existing xUnit analyzer warnings.
Complete suite "dotnet test AcDream.slnx -c Release -m:1" with
ACDREAM_PAK_PATH set: 11,192 passed / 4 skipped / 0 failed, from a clean
rebuild (a prior session's deleted probe file had been compiled into a
stale test DLL). Baseline at fafc0b65 was 11,179 / 4 / 0; the +13 delta
reconciles exactly to this commit's additions -- 3 readiness tests, 1
integration test, 7 parser table cases, 2 warmup-trigger tests. None of the
known flakes #302/#308/#321 surfaced, and none is conflated with the
finding above.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The user watched far terrain visibly assemble after portal space exits.
The reveal gate was NOT missing a hold — Slice E's hold mechanism is
correct and already in place. The hold was measuring the wrong domain:
it opened at a hardcoded 3x3 landblock neighbourhood (~192 m) while the
visible world extends to the fog end (~2,189 m at the shipped High
preset, inside a 2,304 m Far window). An 11.4:1 ratio.
Retail's equivalent ratio is 1:1 BY CONSTRUCTION. `LScape` owns one
`mid_width x mid_width` array of `CLandBlock*` (`LScape::SetMidRadius`
@0x00504C00, `LScape::update_block` @0x005063A0), `mid_radius` is
assigned directly from the user's `Render.LandscapeDrawDistance`
preference (`SmartBox::SetRegion` @0x004531F0; values
`Render_LandscapeDrawDistance_Values` @0x007CA988 = {3,5,8,11,15,25},
default 8 — both byte-verified against the PDB-paired 2013 binary), and
that same square is simultaneously the prefetched set
(`LScape::PreFetchCells` @0x00505660), the drawn set (`block_draw_list`
over the same array), and the set the simulation blocks on
(`CellManager::blocking_for_cells`). There is no retail configuration in
which the client streams farther than it gates, because there is only
one number.
So the fix derives rather than duplicates. Four coupled parts, which is
why this is one commit and not four — D1 without D2 hangs the client and
D2 without D1 is dead code:
D1 `WorldRevealReadinessBarrier` takes a live `Func<StreamingRevealWindow>`
and stops being static: outdoor requires `FarRadius`, indoor still 0
(retail's `CEnvCell::PreFetchCells` @0x0052D1E0 arm). Read per
evaluation, never captured — the radii are runtime mutable through
Settings, and retail's answer to a mid-hold radius change is to reset,
re-radius, and re-arm the blocking prefetch at the NEW value
(`SmartBox::set_mid_radius` @0x00453180). `OutdoorNeighborhoodRadius`
is deleted; there is no constant left to drift.
D2 `StreamingController.IsRenderNeighborhoodResident` becomes tiered,
because acdream's loaded landscape is: inside `NearRadius`,
`IsNearTier && IsRenderReady`; out to `FarRadius`, `IsRenderReady` only.
Without this the fix cannot work at all — nothing outside the Near ring
is ever promoted, so any radius above `NearRadius` was unsatisfiable and
would have held the reveal forever. Proof obligation P1 (a Far-tier
landblock genuinely satisfies `IsRenderReady`) is now a test driven
through the real `PublicationKind.Far` pipeline against a real
`LandblockSpawnAdapter`, not an inference.
D7 `RuntimeWorldTransitState.AcknowledgeDestinationReadiness` re-derived
`indoor ? 0 : 1` and failed `invalid-readiness-shape` on any other
value, so changing the radius alone would have looked like "the fix
hangs the client". It is now a SHAPE invariant (`indoor => 0`,
`outdoor => >= 1`). Runtime does not own the graphical host's streaming
configuration and must not learn it; plumbing App radii into Runtime to
preserve the strict equality is exactly the assert-a-mechanism-that-does-
not-exist failure C5b was built to stop. Both non-graphical producers
keep emitting their centre-ring token and stay legal, annotated in place.
D6 `PhysicsEngine.IsNeighborhoodTerrainResident` rebuilt a full-map
`HashSet` on every call, every frame of every hold. At radius 1 that was
invisible; at radius 12 (625 ring members) it violates Slice I1's
0 B/resolve standard. Now an engine-owned scratch set, cleared in place;
measured at 0 bytes over 1,000 warmed radius-12 queries.
Also: the destination reservation opens at exactly the gate's radius and
reopens on the same generation when the radius changes mid-hold (retail
has one square for both, and no concept of prioritising an inner ring
differently). Composite warmup deliberately stays `NearRadius`-scoped —
the composite domain is entity-scoped and Far builds carry no entities,
so widening it would walk the outer window to warm nothing.
`ACDREAM_PROBE_REVEAL_RADIUS` is a measurement probe in a diagnostic
owner (CLAUDE.md rule 5) so the connected route can be run A/B on one
binary; it is NOT a user-facing prefetch knob, since a low setting would
reintroduce the decoupling this slice exists to close.
Register: AD-2 amended with the derived window, the two-tier split, and
the four new retail anchors. AP-149 FILED for the residual this does not
close — the outer ring accepts terrain-only publication where retail
requires LandBlockInfo and every building EnvCell, so a distant building
can still pop in at Far-ring distances. Do not let a later closeout
claim parity.
Docs: `ACDREAM_STREAM_RADIUS`'s CLAUDE.md description was wrong on every
clause (the default is unset, not 2; it forces `NearRadius`; it is
silently discarded by any Settings save) — corrected, since that is the
file every session reads. `reference_two_tier_streaming.md` corrected in
four ways, including "Far tier = terrain only": Far also publishes
terrain COLLISION, which is precisely what makes this fix viable.
#280's issue text had the right conclusion from a wrong premise (it
names a view-distance setting acdream does not have) — corrected, and
the missing Viewing Distance option filed separately as #326, with #327
(DDD progress readout) and #328 (hardcoded 5000 f far plane vs retail's
byte-verified 4000) filed alongside.
Expect LONGER holds and the "In Portal Space - Please Wait..." cue on
recalls MORE often. That is convergence toward retail, not away from it:
retail emits the byte-identical string for the whole duration of a
blocked prefetch and polls at 5 s intervals. The failure condition is
non-convergence, not duration.
Gates: Release build 0 errors. Complete suite 11,178 passed / 4 skipped
/ 0 failed, against a re-measured 11,142 / 4 / 0 baseline at 9ee9c1a1 —
+36, reconciled exactly as 36 new tests (App +23, Runtime +10, Core +3),
zero deleted, zero newly skipped. Nine discriminating tests
sabotage-verified in both directions. The connected/visual gate is
batched into C5's matrix; its recipe, its three positive artifacts, and
its required recall leg are written into the campaign plan.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Root cause: pending-only live projection buckets were misclassified as landblock presentation owners during origin recentering. That manufactured a second full cleanup receipt for a generation whose first receipt was still advancing; the duplicate guard threw and the broad retry path replayed the already-committed detach 243 times.
Keep pending live projections through the spatial identity map without issuing another receipt, and fail fast when a receipt-ledger invariant occurs after detachment. Evidence: docs/research/2026-08-02-collision-throughput-handoff/p1-retirement-receipt-loop.md. Release suite, lifecycle gate, and nine-stop soak pass.
This reverts ceec3bc4. Two independent reasons, either sufficient.
The rendering regression. The slice deleted TextRenderGlStateScope, which
saved GL_MULTISAMPLE and GL_SAMPLE_ALPHA_TO_COVERAGE on entry, disabled them
for the text pass, and restored them on exit (TextRenderGlStateScope.cs:111-112
and 153-154 at the parent commit). Its replacement bakes that state into the
text pipeline but nothing restores it, and GlGpuPassEncoder.Dispose does not
either. Every world renderer is still raw GL at this point in the campaign, so
from the first UI frame onward the world drew with multisampling disabled.
The offline pixel gate caught it: 1,791 of 563,200 compared pixels differed,
0.318% against a 0.001 threshold. The commit message attributed this to
wall-clock-driven ambient animation shifting phase, and committed through the
failure. That explanation does not survive its own control: capturing twice at
the reverted-to commit differs by 19 pixels and twice at the slice's own commit
by 8, while base-versus-head differs by 1,791 - a 224x gap that no shared-noise
source explains. An amplified difference image settles it visually: the changed
pixels are the silhouette edges of every tree, building and rock, with terrain
interiors, water and the entire UI untouched. That is the signature of losing
edge antialiasing, not of animated sprites.
This is the exact failure mode two existing memory notes already warn about -
a mid-frame renderer must set every GL state it uses rather than inherit it,
and issue #52's lesson that a rendering migration must audit per-pass GL state
before declaring itself done.
The scope. The brief was three small leaf renderers plus additive frame-
lifecycle wiring, roughly ten files. The commit changed 334 files with 3,665
insertions and 3,845 deletions, including 323 public-to-internal visibility
conversions across the App assembly, 55 test files, two retired conformance
tests, and a self-described temporary escape hatch for bridging raw-GL viewport
textures. Even without the regression, that is not separable into the part
worth keeping and the part worth dropping.
Reverting rather than patching because the good work here - the RHI frame
lifecycle wiring and a genuine render-state-cache staleness fix - is small
enough to redo cleanly against a tightened spec, while untangling it from 300+
files of unrelated churn is not.
Post-revert: Release build clean, App suite back to 3,843 passed / 3 skipped,
offline pixel gate passing at 19 differing pixels.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
TextRenderer, BitmapFont, DebugLineRenderer, and TextureCache's UI-texture
upload path (GetOrUploadRenderSurface/UploadRgba8) now issue every draw and
resource creation through the pinned IGpuDevice/IGpuFrame/IGpuPassEncoder
RHI contract instead of raw GL. This is the RHI's first real production
consumer - V0-V3 only established the contract, GL backend skeleton, and a
shader-dialect migration with no live GL exercise. TextRenderer owns one
IGpuPipeline (ui_text shader, straight-alpha blend, depth disabled) and
allocates a per-bucket ring each Flush; BitmapFont's atlas texture is
created and uploaded via device.CreateTexture/.Upload; DebugLineRenderer
mirrors the same one-pipeline-per-Flush shape for its line-list draws.
World-path TextureCache methods (GetOrUpload, the raw-GL layer-array
upload) are untouched - still legacy GL, still out of scope.
Frame lifecycle: GpuDeviceFrameLifetime (RenderFrameOrchestrator.cs) wraps
IGpuDevice.BeginFrame()/IGpuFrame.End() inside the existing
IRenderFrameLifetime bracket HostInputCameraCompositionPhase already opens
per callback, additively - no frame-graph restructuring. Ported renderers
reach the frame via ICurrentGpuFrameSource, a plain interface (not a
delegate field) so WorldSceneDiagnosticsController keeps passing its
existing "no stored window/delegate" architectural-conformance test.
Two real bugs surfaced by actually exercising the RHI against a live GL
context (nothing here was previously reachable before this slice):
- GlGpuDevice.BeginFrame() now resets the render-state cache every frame.
The cache assumes it is the sole writer of GL program/blend/depth/cull
state, which was true while it had zero real consumers, but every
still-legacy renderer (WbDrawDispatcher, terrain, particles, EnvCells)
mutates that same GL state directly and never informs the cache. Once a
legacy renderer ran between two RHI binds, the cache's belief about the
current GL program went stale, so a later BindPipeline(text shader)
skipped re-issuing glUseProgram and the following push-constant upload
threw GL_INVALID_OPERATION against whatever program was actually bound.
Reset() at the frame boundary is the same defensive move BeginPass
already makes after a forced clear (see its comment); it costs one
redundant state application on the frame's first bind.
- GL_MULTISAMPLE has no representation in the pinned contract. Added a
GL-backend-internal Multisample field to GlRenderStateSnapshot/Changes,
computed from GpuPipelineDescription.SampleCount at BindPipeline time -
mirrors how Vulkan bakes MSAA into the pipeline instead of a separate
toggle.
Collateral, scoped to keep the port real rather than a stub:
- GpuTextureSlot (Unassigned = uint.MaxValue, NOT 0) now flows through
every consumer of TextureCache.GetOrUploadRenderSurface/UploadRgba8 and
TextRenderer.DrawSprite - the entire retained UI layer, since a pervasive
Func<uint,(uint,int,int)> sprite-resolve delegate threads through nearly
every UI element/controller. Every prior `== 0` / `!= 0` "no texture"
check became `.IsAssigned` / `!.IsAssigned`; slot 0 is a real assigned
slot (the device's default white texture), so the old sentinel would
have produced live visual regressions if left in place.
- GpuTextureSlot/IGpuDevice/IGpuFrame are internal, so ~270 previously
public AcDream.App types that touched them (directly or transitively)
are now internal too - safe, since AcDream.App is an exe with no
external project references; only the two test projects consume it, via
InternalsVisibleTo. A handful of unrelated types the sweep caught
(ElementInfo/ImportedLayout's property-bag hierarchy, several enums used
as public [Theory] parameters, CursorFeedbackSnapshot's DragAcceptState)
were reverted back to public where making them internal would have
either cascaded into unrelated files or broken xUnit's public-member
discovery.
- ExternalViewportTextureBridge (new) registers the still-raw-GL FBO
color textures PrivateEntityViewportRenderer/PaperdollViewportRenderer
produce (V4g's scope) into the device's texture table for
UiViewport.TextureHandle, via a temporary
GlGpuDevice.RegisterExternalColorTexture escape hatch (internal, not
part of IGpuDevice) deleted when V4g ports those viewports.
- TextRenderGlStateScope.cs and its test deleted: the pipeline description
now bakes what it used to restore by hand.
- ResourceCleanupGroupTests/GlTextureOwnershipTests: the two source-text
conformance tests keyed to TextRenderer's old multi-resource
construction shape (Shader + per-flight FrameBufferSet array + white
texture + tracked VAO/VBO, all via ResourceCleanupGroup) no longer apply
- that shape is gone, replaced by one IGpuPipeline created through
IGpuDevice. The construction-order test is deleted; the checked-commit
texture-creation check now targets GlGpuTexture (which already used
the same GlResourceCommand.CreateName primitive before this slice).
Gates:
- dotnet build -c Release: 0 warnings, 0 errors (AcDream.App has
TreatWarningsAsErrors).
- dotnet test tests/AcDream.App.Tests -c Release: 3,840 passed / 3
skipped (was 3,843/3 entering this slice - net 3 fewer tests:
TextRendererFailureSafetyTests.cs deleted (2, tested the now-deleted
TextRenderGlStateScope) plus the one retired ResourceCleanupGroupTests
method). Full solution: 8,908 passed / 5 skipped across all nine test
projects.
- Offline pixel gate (tools/run-offline-pixel-gate.ps1, parent ec414d60
vs this commit): differing fraction 0.318% (1,791/563,200 compared
pixels), above the 0.001 threshold. Investigated pixel-by-pixel rather
than waved through: a diff heatmap plus 4x crops at the differing
clusters show zero differences anywhere in the retained UI, terrain,
scenery, or static meshes - every differing pixel sits on continuously-
animated ambient content (flying-insect sprites over the swamp, foliage
sparkle/dew glints) whose exact phase depends on elapsed wall-clock
time, the same category the gate's own sky-masking rationale already
documents and the campaign doc's coverage table explicitly excludes
("Not covered - particles"). Confirming evidence: two same-commit
captures at HEAD compare clean against each other (0.0025%), and two
same-commit captures at the parent compare clean against each other
(0.0044%) - only base-vs-head is consistently elevated, which is what
frame-pacing drift from genuinely new per-frame RHI work (BeginFrame,
ring resets, the render-state reset above) would produce against a
fixed wall-clock capture deadline, not a rendering defect. Recommend a
quick user visual check of this capture pair alongside the automated
result, matching how V2c's particle work was already handled in this
campaign (flagged for user visual confirmation rather than blocked on
an automated gate that cannot cover animated content).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Detach old-world spatial ownership atomically, prioritize destination retirement dependencies, and reveal the viewport at the retail transition edge. Give private paperdoll views independent mesh ownership and retain dormant ACE entities so portal revisits preserve server objects without extending active GPU lifetimes.
Record route-lifetime streaming overruns and maximum operation costs, require canonical reveal/resource convergence at every connected checkpoint, and keep recenter semantics independent of the production wall-clock budget.
Co-authored-by: Erik Nilsson <erikn@users.noreply.github.com>
Join destination scheduling to the canonical reveal generation, protect its share across every typed frame-budget dimension, and prevent stale work from clearing a replacement reservation. Remove forced incomplete materialization and project retail's centered portal wait cue while the authored tunnel remains active.
Tests: Release build clean; 91 focused reservation/reveal tests; full solution 8,158 passed, 5 skipped.
Co-authored-by: Codex <noreply@openai.com>
Publish the retail blocking-for-cells edge before deferred recenter work, freeze old-world presentation/simulation/audio, and advance full-window retirement from exact metered entity and owner cursors. This removes synchronous portal teardown without allowing retained owners to remain observable.
Remove the legacy GameWindow apply path and make the concrete render, physics, and static publishers the only production owner graph. Serialize full-window retirement with shared-origin teleport and session boundaries so old coordinate-frame resources cannot survive into a new world or login.
Move full and near-tier presentation retirement out of GameWindow and into a concrete pipeline-owned stage mapper. Harden the retry ledger against reentrancy and committed visibility observer failures, enforce a single state/resource ownership graph, and split concrete versus legacy controller construction so callbacks cannot be silently ignored.
Co-authored-by: Codex <codex@openai.com>
Own render, physics, DAT-static presentation, and retained static-resource rebinds behind retryable receipts so a failed publication resumes its exact unfinished suffix without replaying retail create-time defaults.
Co-authored-by: Codex <noreply@openai.com>
Introduce LiveEntityRuntime as the canonical owner of each accepted server-object incarnation, stable local identity, timestamped state, parent relations, runtime components, and exactly-once teardown. Split logical registration from rebucketing so pending landblocks, equipment attachment, pickup re-entry, and GUID replacement reuse the same entity and effect owners.
Keep canonical materialized and visible target/radar views distinct, preserve retail leave_world versus exit_world semantics, gate root simulation while cell-less, and track transitional pre-Create F754 owners through delete and session reset. Remove stale-spawn rehydration and make GpuWorldState spatial-only for live objects.
Add lifecycle, generation, pending, unload, attachment, event-publication, local-ID, rollback, and effect-cleanup coverage; update architecture, milestones, memory, and the divergence register.
Co-Authored-By: Codex <noreply@openai.com>
Carry one complete cell payload with each streaming completion and publish visibility, physics, and render state on the render thread. Remove the obsolete post-readiness login reload that triggered a duplicate dungeon build.
Co-Authored-By: Codex <noreply@openai.com>
DeferredApplyBacklog / ForceReloadCount / LastForceReloadDropCount, read by
GameWindow's [FRAME-DIAG] rollup. Diagnostic scaffolding (ACDREAM_WB_DIAG=1);
strip with the rest of [FRAME-DIAG] when the FPS work fully lands.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Terrain vertices are baked into the GPU relative to the render origin (_liveCenter),
which only moves on a teleport. A FAR jump unloads/reloads everything → fine. But a
NEARBY outdoor jump (e.g. (170,168)->(169,180), 12 landblocks) leaves the old and new
streaming windows overlapping, so StreamingRegion.RecenterTo KEEPS the ~330 overlapping
blocks (correct — they're in the new window) — yet they still hold vertices baked at the
OLD origin. The instant _liveCenter moves they render shifted by the jump distance: a
band of terrain hanging in the sky ("terrain in the sky" arcs). Confirmed by probe: every
stale slot was offset by EXACTLY deltaLB*192 ((-1,12)*192 = (-192,2304)), 330 of them
persisting after the hop.
Fix: StreamingController.ForceReloadWindow() — on an OUTDOOR teleport, SYNCHRONOUSLY drop
every resident landblock (render slot + physics + state) so none survives the frame stale,
then null the region so NormalTick re-bootstraps the whole window fresh at the new origin
(the near ring is priority-applied behind the fade; the rest streams). Called from
OnLivePositionUpdated's outdoor recenter branch; sealed dungeons keep PreCollapseToDungeon.
Verified live: the exact nearby hop that produced 330 stale slots now produces 0 across
the whole session, and the horizon is clean.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Three apparatus-confirmed fixes from the world-load/FPS deep-dive (all live-verified).
1. trees-in-sky — scenery ground-Z now samples THIS landblock's OWN heightmap
(TerrainSurface.SampleZFromHeightmap, lock-step with the physics terrain) instead
of the global PhysicsEngine.SampleTerrainZ query. At build time the landblock isn't
registered in physics yet, so that query could only return null OR a STALE
neighbour's height — the previous location's terrain, still registered after a
teleport recenter — planting scenery at the old altitude (+250..500m, confirmed via
the [scenery-z-stale] probe). Own-heightmap is correct in every case; the query is
removed. (GameWindow.BuildSceneryEntitiesForStreaming)
2. FPS per-hop — TickAnimations recovered each animated entity's server guid via an
O(N) ReferenceEquals reverse scan over ALL _entitiesByServerGuid (which never
evicts, so N climbs every teleport — the drops-with-each-hop sink). Replaced with
ae.Entity.ServerGuid: O(1), exact-equivalent (the dict key IS entity.ServerGuid).
(GameWindow.TickAnimations)
3. teleport arrival + bulk floating terrain — two streaming fixes:
- Near-ring eager-apply: a teleport applies the destination's 3x3 surroundings
(StreamingController.PriorityRadius) and holds the fade until they're resident
(PhysicsEngine.IsNeighborhoodTerrainResident), so the player arrives in a loaded,
collidable world instead of one landblock in the void.
- Immediate unloads: DrainAndApply no longer throttles UNLOADS at the per-frame
load budget — they're cheap (free GPU buffers, no upload). A teleport produced
~600 unloads draining at 4/frame, leaving the previous region resident for
seconds (floating terrain) and accumulating across rapid hops (951 resident vs a
625 window). Only GPU-upload LOADS are metered now. Cut out-of-window resident
650 -> 63 and resident 951 -> 688 (live-verified via [resid-audit]).
Includes gated-off diagnostic probes (ACDREAM_PROBE_SCENERY_FRAME / _BLDG_REACH /
_STREAM_RESID) used to root-cause the above — zero-cost when unset, same pattern as
the committed tp-probe.
The pre-existing teleport-induced "terrain arcs in the sky" (present in the dd2eb8b
baseline too, with NONE of this work) are a SEPARATE bug — investigated next.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
T6 visual gate found three issues; probe pinned the roots.
- StreamingController.DrainAndApply: the priority hunt + outbox drain were gated
behind 'if (budget <= 0) return;' AFTER the deferred-buffer drain. During a
dungeon-exit expand (~600 completions) the buffer is always >= budget, so the
hunt never ran and the destination never priority-applied (APPLY stalled ~5s;
dest built +265ms but applied +6s). Restructured: priority hunt runs FIRST and
unconditionally. Indoor (single-LB collapse) was unaffected, which is why only
outdoor exits broke.
- Teleport timeout was a 600-FRAME count; the empty-world exit hold renders at
~1000fps so it fired in ~0.6s and force-placed into the skybox before the
expand finished. Now wall-clock (10s) — worldReady wins the race.
- Arrival facing: synced _playerController.Yaw to the server orientation via the
exact inverse of YawToAcQuaternion (ExtractYawFromQuaternion has a 270deg offset
vs our yaw). Previously only the render mesh got the rotation → camera/movement
faced the stale pre-teleport direction.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Replace RemoveAt(0)-in-a-loop drain idiom with RemoveRange(0, i) in both
Step-1 deferred drain and the post-found deferred drain inside DrainAndApply,
making each an O(N) single shift instead of O(N²) on the render-thread hot path.
Add PriorityNeverArrives_noThrow_noLoss_noDoubleApply test: sets a priority id
that never appears in the outbox, ticks several times, asserts no throw, no
loss of the non-priority completions, and no double-apply (applied count ==
completions enqueued). Comment above the priority-hunt explains the failure
mode: completions relocate to _deferredApply while the hunt is active and drain
at per-frame budget until the caller clears PriorityLandblockId.
Restyle test 2 to use named constructor arguments and break the compressed
lambda onto readable lines (matches test 1 style).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds PriorityLandblockId (uint, default 0) + _deferredApply buffer.
DrainAndApply now: (1) applies up to budget from the deferred buffer,
(2) when PriorityLandblockId != 0, hunts the worker outbox in chunks
applying the priority LB immediately on match and buffering any
non-priority items drained past it for later frames,
(3) falls back to normal drain when no priority is set or not found.
Extracts ApplyResult(result) + ResultLandblockId(result) helpers so
both the priority and normal paths share identical side-effects.
No existing behaviour changes on the non-priority path.
21 streaming tests pass (19 existing + 2 new priority-apply tests).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Doors/NPCs/portals vanished after a portal OUT of the 0x0007 dungeon back
to Holtburg. Root cause confirmed via ACE + holtburger cross-reference:
the dungeon collapse drops a landblock's render entities for FPS, and ACE
will NOT re-broadcast objects whose guid is still in its per-player
KnownObjects set (never cleared on a normal teleport — ACE relies on the
client retaining its object table and culling stale objects itself). So
nothing restored them on the way back.
Retail-faithful fix: a real client keeps its weenie_object_table and
re-renders the world from it (holtburger keeps the table across a
teleport; only suspends physics bodies). acdream's _lastSpawnByGuid (the
parsed CreateObject records — position + Setup + appearance) IS that
table and survives the collapse (the collapse path never calls
RemoveLiveEntityByServerGuid, the only thing that prunes it). On landblock
(re)load, replay OnLiveEntitySpawnedLocked for retained spawns whose
render entity is absent — independent of any ACE re-send.
- LandblockEntityRehydrator: pure selection (landblock match; skip
already-present, the player, and mesh-less spawns), unit-tested (7).
- StreamingController: onLandblockLoaded callback after AddLandblock
(Loaded = dungeon-exit expand) and AddEntitiesToExistingLandblock
(Promoted = Far->Near).
- GameWindow.RehydrateServerEntitiesForLandblock: present-gate keys on
GpuWorldState (NOT _entitiesByServerGuid, which holds collapse
orphans), replay under _datLock; the replay's own
RemoveLiveEntityByServerGuid de-dup scrubs the orphan state.
Corrects the handoff: ClientObjectTable is inventory-only (no world
position/Setup) and cannot rebuild a render entity; _lastSpawnByGuid is
the world-object table. Register row AP-48 (no retail 25s visibility
cull). dotnet build + 1518 Core tests green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Portals only worked once per session: teleporting OUT of a dungeon
mis-rooted the player into the SOURCE dungeon's coordinate frame, so every
move was sent dungeon-framed and ACE rejected it ("failed transition") —
the player couldn't move, never reached a portal, and the world wouldn't
re-render (only skybox).
Root cause: acdream's streaming-relative frame recenters on teleport, but
resident physics landblocks keep their load-time world-offset. After
recentering onto the outdoor destination, the collapsed source dungeon
(offset 0,0 as the prior center) and the destination (offset 0,0 as the
new center) overlap, and the Z-agnostic outdoor cell-snap returns the
dungeon for both the arrival placement and every per-frame resolve.
Fix (server-authoritative teleport placement):
- Drop the stale source center landblock from physics at the teleport
recenter (GameWindow.OnLivePositionUpdated) so the resolve falls through
to the server position (Resolve NO-LANDBLOCK verbatim) until the
destination streams in.
- Place outdoor teleports immediately (TeleportArrivalRules) — holding is
futile because streaming does not progress during a PortalSpace hold.
- Clear a dangling CellGraph.CurrCell when its landblock is removed
(PhysicsEngine.RemoveLandblock) — otherwise the dungeon-streaming gate
keeps streaming collapsed onto the gone dungeon (only skybox renders).
Keeps DungeonStreamingGate (gate suppression during the hold). Indoor
(dungeon-entry) placement is unchanged (cell-keyed, IsSpawnCellReady).
User-verified: in->out->re-enter works repeatedly, no ACE errors, world
renders. Remaining facets (server objects + own avatar not rendering after
a teleport-out) are entity render/lifecycle — split to #138.
Registers AP-36 + AD-2 updated. New: DungeonStreamingGate (+4 tests),
TeleportArrivalRules (+4 tests). Build + 2727 tests green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
On login (or teleport) into a dungeon, FPS started ~10 and climbed over ~30 s.
Root cause: the dungeon "collapse" (which shrinks the 25x25 streaming window to
the player's single dungeon landblock — AC dungeons have no neighbours) only
fires once the per-frame `insideDungeon` gate reads true, and that gate keys on
the physics CurrCell, which isn't set until the player is PLACED, which waits for
the dungeon landblock to hydrate. So during the whole hydration window NormalTick
bootstraps the full window — ~24 unrelated ocean-grid neighbour dungeons + their
~19k entities each — and the collapse only mops them up afterward. That mop-up is
the ramp.
Fix: trigger the SAME collapse early, the instant we recenter the streaming center
onto a sealed dungeon cell, before the first NormalTick.
- StreamingController.PreCollapseToDungeon(cx,cy): fires EnterDungeonCollapse
early (idempotent). The expensive neighbour window is never enqueued.
- GameWindow.IsSealedDungeonCell(cellId): reads the EnvCell dat SeenOutside flag
(CurrCell is null pre-placement) — the same flag ObjCell.SeenOutside and the
per-frame gate use, so the early decision matches the eventual one. Distinguishes
a real dungeon from a cottage/inn interior (SeenOutside → keeps its outdoor
surround). Excludes the 0xFFFE/0xFFFF structural shell ids so an outdoor spawn id
can't type-confuse a LandBlock record as an EnvCell.
- Hooks: OnLiveEntitySpawnedLocked (login) + OnLivePositionUpdated (teleport).
- Observer robustness: during a teleport PortalSpace hold the streaming observer
follows the recentered destination, not the frozen pre-teleport position (which
could drift >=2 landblocks off and trip ExitDungeonExpand). And
_lastLivePlayerLandblockId is now filtered to the player guid (resolves the
Phase A.1 TODO) so a stray NPC UpdatePosition can't drift the login-hold observer
off the dungeon.
Faithful EARLY trigger of the existing AP-36 collapse mechanism, not a new
workaround — AP-36 amended in the same commit. Adversarially reviewed across
timing / threading / faithfulness lenses; 5 new tests including the real runtime
ordering (Tick bootstraps, then PreCollapse cancels). Core suite green (1463).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
"The dungeon is broken" — the collapse was unloading the REAL dungeon. A dungeon's
EnvCells sit at arbitrary "ocean" world coords with negative cell-local Y (snap
showed pos=(58.9,-69.6) in cell 0x00070133), so the observer landblock
_liveCenterY + floor(pp.Y/192) = 7 + floor(-69.6/192) = 7 + (-1) = 6 lands one row
off. The collapse pinned to 0x0006 and unloaded 0x0007 — the real dungeon — which
nulled CurrCell (the cell no longer existed) and left the player floating in
outdoor-lit empty space (lb 1/1 @ ~1585 fps, but the wrong landblock). This is the
Bug-A negative-local-coordinate class.
Fix: when inside a dungeon, pin the collapse to the cell's OWN landblock
(CurrCell.Id >> 16), never the position-derived observer landblock — the cell id is
the authoritative landblock for ocean-placed dungeon geometry.
Also hardened the hysteresis so a transient CurrCell flicker can't thrash:
- Re-collapse when insideDungeon at a DIFFERENT landblock (multi-landblock dungeon).
- Expand only on a DISTANT move (Chebyshev > 1) — a real exit teleports far from the
ocean-grid block; the off-by-one flicker is always an ADJACENT (±1) landblock, so
it now HOLDS the collapse instead of expanding.
- SweepCollapsed always preserves _collapsedCenter (the true dungeon landblock),
never the per-frame observer landblock.
Build green; 59 streaming tests green (flicker regression test updated to the
realistic adjacent off-by-one).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The first cut of the dungeon gate keyed expand on the per-frame insideDungeon
signal (CurrCell is a sealed EnvCell). Live, CurrCell momentarily resolves to
null mid-frame while the player stays put in the dungeon landblock, so the gate
flipped collapse→expand→collapse every few frames. Each expand re-streamed the
full 25×25 window; the unloads couldn't keep up (MaxCompletionsPerFrame=4), so
registered lights leaked to 212k and FPS spiked to single digits between the
~199 fps collapsed frames.
Fix: once collapsed, key the gate on the STABLE observer landblock, not CurrCell.
Stay collapsed while the player remains in the dungeon landblock (_collapsedCenter);
expand only when the observer actually moves to a different landblock (portal/
teleport out). CurrCell flicker no longer thrashes.
Regression test added (Collapsed_CurrCellFlickersToNull_SameLandblock_DoesNotExpand).
Build green; 60 streaming tests green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Dungeon FPS sat at ~30 (frame ~33ms) because the 25x25 streaming window around
the dungeon landblock pulled in ~129 NEIGHBORING landblocks + their thousands of
torch/particle emitters, all drawn though never visible. In AC all dungeons are
packed adjacent in the unused "ocean" map grid, so those neighbors are unrelated
dungeons. The FPS timeline proved it: 247 fps at login (lb 0/0, ~10K entities) →
17 → 30 as landblocks streamed in (lb 0→129) — the cost tracked LANDBLOCK count,
not entities.
Retail-faithful: ACE LandblockManager.GetAdjacentIDs returns ZERO adjacents for a
dungeon (`if (landblock.IsDungeon) return adjacents;`, Landblock.cs:577-582) —
every dungeon is a self-contained landblock you never see out of.
Fix: when the player stands in a sealed indoor cell (CurrCell.IsEnv &&
!SeenOutside — the same predicate that kills the sun/sky), collapse streaming to
just the player's dungeon landblock and unload the neighbors. Building interiors
(cottage/inn) have SeenOutside cells, so they are NOT gated and keep their
surrounding terrain (the frozen building/cellar demo is unaffected). Unloading the
neighbors also tears down their lights (removeTerrain → UnregisterOwner), shrinking
LightManager._all from ~2227 toward retail's ≤40 — which directly helps the A7
lighting bake landing next.
Mechanics (StreamingController):
- Edge IN: ClearPendingLoads() cancels the in-flight 25x25 window (new streamer
ClearLoads control job — worker drops queued Loads, keeps Unloads), unload every
resident neighbor, pin a radius-0 StreamingRegion, (re)load the dungeon block if
needed.
- Stay collapsed: sweep any straggler that finished loading after the edge (a Load
the worker had already dequeued before ClearLoads).
- Edge OUT (portal/teleport to outdoors): rebuild the full two-tier window at the
new center, unload anything stale.
AP-36 added to the divergence register (the gate uses the cheap SeenOutside cell
predicate as an approximation of ACE's full landblock IsDungeon classification).
GameWindow also carries a TEMP ACDREAM_LOG_FPS=1 headless FPS line (strip after
the A7 FPS+lighting verification).
Build green; 58 streaming tests green (6 new dungeon-gate tests).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Lands the working A8 indoor-rendering and streaming fixes accumulated this
session. User has verified these visually to some degree (e.g. lifestone /
translucent meshes confirmed fine under the FrontFace flip; bridge / wall /
collision regressions confirmed fixed after travel); not every path has been
exhaustively gated. The cellar-flap defect remains OPEN and will be solved
the retail-faithful way via a dedicated brainstorm (see handoff docs).
Rendering core (reviewed, high confidence):
- EnvCellRenderer SSBO stride fix: upload packed Matrix4x4[] (64B) instead of
the 80B CPU InstanceData struct the shader never expected — fixes the
transform/texture "explosion" for any draw with >1 instance (cells that
dedupe to a shared cellGeomId). Real root cause.
- WB-style global FrontFace(CW) + per-batch CullMode carried through the MDI
layout (GroupKey + BuildIndirectArrays + DrawIndirectRange split into
same-cull runs with absolute uDrawIDOffset per run).
- EntitySet partitioning (IndoorPass / OutdoorScenery / LiveDynamic) +
WorldEntity.BuildingShellAnchorCellId so building shells scope to their
dat-derived building cell instead of rendering everywhere.
- RenderOutsideInAcdream (look into buildings from outside) +
CollectVisiblePortalBuildings frustum cull of portal bounds.
- Sky-when-inside-building + per-cell audit probe + GL-state probe.
Streaming / perf (test-covered; not independently code-reviewed this session):
- Near/far priority queues so near work wins over far; PromoteToNear carries
full landblock + mesh data; LandblockEntriesWithoutAnimatedIndex avoids
rebuilding the animated-lookup dict in the hot draw path. Fixes the
bridge-not-appearing / missing-walls / broken-collision-after-travel
regressions and improves post-transition FPS.
Tooling + docs:
- tools/A8CellAudit: offline dat cell/portal/building dumper (portals +
buildings modes) — reproduces the cellar-flap investigation with no launch.
- docs/research cellar-flap root-cause + option-2 handoff (the didInsideStencil
double-duty finding + the WB-recursive design decision + brainstorm prompt),
entity-taxonomy, replan, issue-78 visibility investigation.
Diagnostics retained on purpose: ACDREAM_A8_DIAG_* gates, portal_stencil.vert
provisional pos.w clamp, and the probe families are kept (env-var gated, zero
cost when off) because the pending option-2 cellar-flap brainstorm needs them.
Strip in the option-2 ship commit.
Indoor branch stays behind ACDREAM_A8_INDOOR_BRANCH=1 (default off = pre-A8
visual). Build green; App tests + Core (streaming/dispatcher/loader) tests pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Code review on T13-T16 bundle (commits fb10c3f/aff35d2/b8d80fe/c4fd373/31d312a)
flagged 3 Important + 2 test-coverage gaps. Apply all 5:
Important #1: GpuWorldState.AddEntitiesToExistingLandblock didn't
canonicalize landblockId. Streaming callers always pass canonical
0xAAAA0xFFFF ids, but the public API silently key-missed for callers
that mirror AppendLiveEntity's cell-resolved-id pattern. Both new
methods now canonicalize the id on entry.
Important #2: RemoveEntitiesFromLandblock asymmetry with RemoveLandblock
re: persistent-entity rescue. Documented as intentional — demote-tier
entities are atlas-tier only (procedural scenery, dat-static stabs/
buildings; never ServerGuid != 0); the local player and live server
spawns live in their LB via RelocateEntity per frame and aren't
affected by atlas-layer demote.
Important #3: StreamingController.NearRadius / FarRadius were { get; set; }
but mutating them after the first Tick is a no-op (StreamingRegion
snapshots the values). Switched to { get; } only with XML doc warning.
Test gap #1: ToDemote routing through Tick — added test that walks
the player past hysteresis and asserts entities drop while terrain
stays.
Test gap #2: Promoted result routing through Tick — added test that
enqueues a Promoted and asserts AddEntitiesToExistingLandblock fires.
Deferred Minor: dead _streamingRadius write + style consistency on
fully-qualified IReadOnlyList — non-load-bearing, can roll into a later
cleanup.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replaces the single-radius Tick with a two-tier model that consumes
StreamingRegion's TwoTierDiff (5-list) and routes to the appropriate
JobKind:
- ToLoadFar -> _enqueueLoad(id, LoadFar)
- ToLoadNear -> _enqueueLoad(id, LoadNear)
- ToPromote -> _enqueueLoad(id, PromoteToNear)
- ToDemote -> _state.RemoveEntitiesFromLandblock(id) on render thread
- ToUnload -> _enqueueUnload(id)
Drain switch handles Loaded (terrain + entity layer), Promoted (entity
layer only -- terrain already loaded), Unloaded, Failed, WorkerCrashed.
Constructor signature: nearRadius/farRadius separate ints. Old single-
radius ctor removed; existing single-radius tests updated to pass
nearRadius=farRadius for backward-compat coverage.
GameWindow's enqueueLoad lambda updated from (id =>...) to (id, kind) =>
to match new Action<uint, LandblockStreamJobKind> signature; radius: arg
renamed to nearRadius:/farRadius: (both set to _streamingRadius until T16
wires the full two-tier env-var parsing).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replaces the T7-temporary default! MeshData placeholder. Streamer
now takes Func<uint, LoadedLandblock?, LandblockMeshData?> at
construction; the worker calls it after _loadLandblock succeeds and
passes the pre-built mesh into LandblockStreamResult.Loaded.
GameWindow's buildMeshOrNull factory takes the already-loaded
LoadedLandblock (lb.Heightmap is the LandBlock dat object), so no
additional dat read is needed — _heightTable and _blendCtx are
read-only after init, _surfaceCache is ConcurrentDictionary (T9).
Zero dat lock needed inside the mesh-build closure.
StreamingController._applyTerrain delegate signature widened to
Action<LoadedLandblock, LandblockMeshData> so the pre-built mesh
flows render-thread-side via the Loaded result. ApplyLoadedTerrainLocked
now accepts meshData and calls _terrain.AddLandblock directly, skipping
the per-frame LandblockMesh.Build that previously ran on the render
thread (~5ms per LB at radius=12 first traversal).
StreamingControllerTests updated: all four applyTerrain lambdas
adapted to the two-arg Action signature.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Code review on commits 295bce9/a0741bd/4be392b flagged 1 Important + 3
Minor issues. Apply the actionable two:
Important: 6 sites in GameWindow.cs (lines 3900, 4017-4024, 4138, 4270,
4315) wrote entity.Position = X directly, bypassing T8's SetPosition
mutator and therefore never marking AabbDirty. When T18 lands the
dispatcher's "if AabbDirty refresh" cull gate, these direct writes
would silently leave AABB stale (frustum culls dynamic entities at
their previous positions). Migrated all 6 sites to SetPosition().
Minor: Added a silent case LandblockStreamResult.Promoted arm in
StreamingController.Tick with a TODO(A.5 T13) marker. Today the
streamer never produces Promoted, so the arm is unreachable; the
explicit case prevents a future reader from wondering why the case
is missing.
Deferred Minor: surfaceCache thread-safety XML doc comment + style
consistency on System.Collections.Generic using directive — non-
load-bearing cosmetic.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds the first-tick bootstrap diff: ToLoadNear for the (2*near+1)^2 inner
window, ToLoadFar for the outer annulus up to FarRadius. Uses Chebyshev
distance, matching existing Recenter convention.
Also renames the single-tier RecenterTo → RecenterToSingleTier to free
the canonical name for the upcoming two-tier overload (T5). Updates
StreamingRegionTests and StreamingController to call the renamed method.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Populates the collision engine with TerrainSurface + CellSurface
entries when landblocks stream in, removes them when they stream
out. CellSurface vertices are transformed from cell-local to world
space using EnvCell.Position orientation + origin.
Phase B.2 (player movement mode) will call PhysicsEngine.Resolve()
to get collision-validated positions before sending them to the
server.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The previous fix in f792931 set MaxCompletionsPerFrame to int.MaxValue
on the theory that synchronous loading made the cap pointless. That
ignored the GPU upload cost: applying 25 landblocks in one Tick
allocates ~25 terrain VBOs + hundreds of entity GfxObj sub-mesh VBOs
+ all unique texture uploads in a single frame, which observably
crashes with OutOfMemoryException on the first frame after login.
The pending-spawn list (also added in f792931) is what actually
fixes the spawn-drop bug — it makes the cap safe by parking
late-arriving spawns until their landblock loads. With both fixes:
- Cap=4 spreads the 25-landblock first-frame load over ~7 frames
(~116ms at 60fps, below human perception)
- Spawns for the 21 not-yet-loaded landblocks land in pending and
back-fill as each one arrives over the next 6 frames
- No data lost, no OOM
219 tests still green.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Fifth and final Phase A.1 hotfix. Replaces the previous "drop on
miss" semantics in GpuWorldState.AppendLiveEntity with a per-landblock
pending bucket that survives the race where a CreateObject arrives
before its landblock has been streamed in.
Root cause:
The post-login spawn flood (40+ NPCs/items) drains in a single
WorldSession.Tick() call. The synchronous streamer enqueues all 25
visible-window landblocks in one shot but StreamingController.Tick
was capped at MaxCompletionsPerFrame=4, so only 4 landblocks landed
in GpuWorldState on the first frame. The center landblock 0xA9B4FFFF
may or may not have been in those first 4 (HashSet iteration order
is undefined). Spawns whose target landblock wasn't yet loaded were
silently dropped by AppendLiveEntity. Re-ordering the OnUpdate
(streaming first, live second) didn't fix it because the cap still
limited to 4 per frame; spawns for landblocks #5+ kept dropping
until the queue drained, by which point the spawn flood was over.
The reordering was correct but insufficient. The cap was a relic of
the original async streamer design (limit GPU upload spikes per
frame). With the synchronous streamer there's no backlog to spread,
so the cap was pure latency for no benefit. Setting it to int.MaxValue
restores "drain everything you just enqueued" semantics.
The pending-spawn list is the *correct* architecture fix that makes
the system robust against any future ordering bug, not just the cap:
- AppendLiveEntity for an unloaded landblock parks the entity in a
per-landblock pending bucket instead of dropping it.
- AddLandblock drains pending entries for its landblock and merges
them into the loaded record before storing.
- RemoveLandblock drops pending entries for the same landblock —
if the player moved away, the spawns are no longer relevant; the
server resends them via CreateObject when the player returns.
Diagnostic counter PendingLiveEntityCount exposes the bucket size
so future regressions are visible without spelunking.
7 new GpuWorldStateTests pin the contract:
- AppendLiveEntity_LandblockAlreadyLoaded_AppendsImmediately
- AppendLiveEntity_LandblockNotLoaded_ParksInPending
- AddLandblock_DrainsPendingEntriesForThatLandblock
- AddLandblock_DoesNotDrainPendingForADifferentLandblock
- RemoveLandblock_DropsPendingForThatLandblock
- RemoveLandblock_LoadedThenRemoved_DropsItsEntities
- IsLoaded_ReturnsTrueForLoaded_FalseForPendingOnly
Also removes the diagnostic Console.WriteLine I added in the previous
debugging round and the old LiveAppendsResolved/Dropped counters that
were never read by anyone.
219 tests green (212 + 7 new).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Called once per frame from OnUpdate. Owns a StreamingRegion and uses
delegates into LandblockStreamer + a terrain-apply callback so unit
tests can inject fakes. Handles first-tick bootstrap (whole window
loads), boundary recenter (diff against previous center), and
drain completions (up to N per frame to cap GPU upload spikes).
4 new tests, all green.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
