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>
The dungeon/house/outdoor lights read as hard-edged blown discs ("spotlights")
because our point/spot shader used `atten = 1.0` flat inside a hard `d < range`
cutoff. The mesh.frag comment claimed this was retail-faithful ("no attenuation
inside Range... the bubble-of-light look relies on crisp boundaries", citing
r13 10.2) — that was a misread and the literal cause of the symptom.
Verified against the decomp (not guessed): calc_point_light (0x0059c8b0, the
PER-VERTEX point-light path that lights static walls) scales each light's
contribution by (1 - dist/falloff_eff) — a LINEAR ramp that fades to exactly 0
at the edge, eliminating the hard disc. falloff_eff = Falloff * static_light_factor,
and static_light_factor = 1.3 (0x00820e24), NOT the 1.5 config_hardware_light
rangeAdjust (that 1.5 is the D3D-dynamic path for moving objects, a different
path). The Ghidra port (acclient.c:808639) is more garbled — BN pseudo-C is the
oracle here; the exact normalization factor + a half-Lambert wrap (0.5*dist+N*L)
are x87-obscured (same artifact class as GetPowerBarLevel) and left unported.
Changes:
- mesh_modern.frag + mesh.frag: replace flat atten with clamp(1 - d/range, 0, 1);
Range now carries falloff_eff so the ramp fades to 0 at the cutoff. Fix the
false "no attenuation / crisp bubble" comment in mesh.frag.
- LightInfoLoader: Range = Falloff * 1.3 (static_light_factor), was * 1.5.
- LightManager: correct the stale class doc comment (Tick is now nearest-8
allocation-free partial-select with NO viewer-range slack filter).
- divergence register: AP-16 updated (slack filter removed), AP-35 added
(per-pixel vs per-vertex Gouraud; dropped half-Lambert wrap + normalization).
- test: LightingHookSinkTests Range 8*1.3 = 10.4.
Build + 20 lighting tests green. Visual gate pending (game-wide lighting change:
dungeon torches, house candles, outdoor braziers).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The headless replay of the captured indoor frame proved the look-in flood ADMITS the porch 0x017A (Diagnostic_LookInFlood_AdmitsHallPorchFromCottage: 14 cells). So the portal (a SERVER object - the teleport proves it - with ParentCellId 0xA9B4017A) routes to partition.Dynamics and draws NOWHERE under an interior root: dynamics-last viewcone-culls it (the main cone has no look-in cells) and post-seal it would z-fail beyond the root's door plane (the #118 lesson). This is AP-33's own recorded deferral - 'look-in DYNAMICS are not drawn' - the deferred case was the most-stared-at object in town. Outdoors the merge path puts the porch in the main cone -> drawn -> 'appears when I walk out'.
Fix: DrawBuildingLookIns pass 2 draws look-in-cell dynamics with the statics (whole, AP-33 over-include) and their emitters ride the same DrawCellParticles call. No double-draw: dynamics-last keeps culling them; DrawDynamicsParticles only sees its cone survivors. #124 CLOSED by user gate same session. AP-33 row updated. Suites: App 261+1skip / Core 1439+2skip / UI 420 / Net 294 green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
User gate on 20d1730: the candle is FIXED indoors ("now the candle
light is visible when I'm in the house when it is in front of the
opening") and the OUTDOOR sibling surfaced exactly as AP-34 recorded
("when I go out it is not showing unless I turn so the angle doesn't
put it in front of the opening"): under an OUTDOOR root the merged
building interiors draw AFTER the landscape stage (DrawEnvCellShells),
so a slice-drawn flame is overpainted by a punched aperture's interior
behind it.
Fix: outdoor roots SKIP the late-slice Scene-particle draw; attached
outdoor-static scene emitters draw in the POST-FRAME pass alongside the
T3 unattached pass, where depth is complete and flames composite
correctly against interiors. The owner-id set carries over from the
late slice (single full-screen slice outdoors); cell-pass and
dynamics-pass emitters keep their own passes (their owners are never in
the outdoor-static id set - no double-draw). Interior roots keep the
late-slice draw (their stage ends with the clear + seal discipline).
AP-34 row updated (the outdoor residual is now covered; the remaining
residual is translucent MESH batches within stage draw calls).
Portal swirl (#131): the user's "same results" on 20d1730 KILLS the
look-in-erasure hypothesis for the portal - the mesh now draws after
the look-ins and is still missing indoors. No further speculative fix;
the [outstage] probe now prints each outside-stage dynamic's
SourceGfxObjOrSetupId (portals have distinctive setups) and
[outstage-pt] lists up to 12 distinct UNMATCHED attached emitter owner
ids - the next capture identifies whether the portal entity reaches the
through-door draw at all, and where its emitters point.
Suites: App 259+1skip / Core 1439+2skip / UI 420 / Net 294 green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The user's screenshot pair re-attributed both reports to ONE mechanism -
a compositing gap in the #124 look-in sub-pass:
- #131: the portal swirl (a TRANSLUCENT MESH, not only particles) stood
exactly in front of the hall's doorway. The slice drew it BEFORE the
look-in sub-pass; translucents write no depth, so the hall's interior
- drawn into its far-Z-punched aperture - overpainted the swirl.
Outdoors the look-ins are the post-stage merge path, so the swirl
survives ("stepping out it pops into existence").
- #132: the candle/lantern flame is an attached emitter in the slice's
Scene-particle pass - same pre-look-in placement, same erasure
whenever "the opening through a house" sat behind it; against a wall
nothing overdraws it. Background-dependence explained exactly.
Retail cannot exhibit this class: every alpha draw of the landscape
stage is collected and flushed ONCE after LScape::draw
(D3DPolyRender::FlushAlphaList, PView::DrawCells pc:432722) - i.e.
after all building look-ins.
Port (the two-phase split): DrawLandscapeThroughOutsideView now runs
EARLY per slice (sky, terrain, outdoor STATIC meshes - the look-in
punches need their depth to mark against, the #117 lesson), then the
#124 look-ins, then LATE per slice (outside-stage dynamics' meshes +
ALL attached scene particles + weather + SkyPostScene), then the #131
unattached pass. New RetailPViewLandscapeLateSliceContext carries the
dynamics survivors + the particle-owner set (statics + dynamics cone
survivors). GameWindow's slice handler split accordingly. Outdoor
roots: no look-ins live in the stage, so the net order is unchanged
(zero behavior change outdoors).
Register: AP-34 added - the two-phase split vs retail's single
deferred flush, with the residuals recorded (outdoor-root slice
particles still draw before merged building interiors - the unreported
outdoor sibling; building exteriors' own translucent batches draw
early).
The earlier #131 unattached-emitter pass (1d3f9a8) remains - it fixes
an independent hole (that class had NO indoor pass at all) - and now
runs at the end of the late phase.
Suites: App 259+1skip / Core 1439+2skip / UI 420 / Net 294 green.
Awaiting the user gate: swirl through the doorway, candle flame with
the opening behind it, far-building interiors (#124).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
From inside a building, looking out at ANOTHER building with an opening
showed its back walls missing (see-through to the world): per-building
look-in floods only ran for outdoor roots; under an interior root the
far building's interior never flooded.
Decomp anchor (named-retail, this session's read): retail runs the
look-in INSIDE the landscape stage for ANY root - LScape::draw is the
FIRST call of PView::DrawCells' outside-view branch (pc:432719),
strictly BEFORE the depth clear (pc:432732) and the exit-portal seals
(pc:432785). ConstructView(CBldPortal) (0x005a59a0) clips each aperture
via GetClip against the INSTALLED view - the accumulated doorway region
when looked into from inside - and build_draw_portals_only pass 1
far-Z punches ALL apertures before pass 2 floods + draws any interior
cell. The nested DrawCells has an empty outside view (PView ctor
draw_landscape=0): no recursive landscape/clear/seal.
Port:
- GameWindow's per-building gather (frustum pre-gate on
Building.PortalBounds) now runs for interior roots too; the root's
own doorway self-excludes via the seed eye-side test (the eye is on
its interior side).
- PortalVisibilityBuilder.BuildFromExterior/ConstructViewBuilding gain
seedRegion - the installed-view clip: interior-root look-ins seed
against the OutsideView polygons (a building not visible through the
doorway never floods); null = full screen (outdoor roots unchanged).
- RetailPViewRenderer.DrawBuildingLookIns: a landscape-stage sub-pass
(before ClearDepthForInterior + seals) - per building, punch ALL
apertures (new DrawLookInPortalPunch callback, always forceFarZ=true,
closing the ISSUES "forceFarZ keys on root kind, under-punches" gap),
then draw the flooded cells' shells + statics far->near. Look-in
frames are NEVER merged into the main frame: a merged cell would draw
post-clear and z-fail against the root's seal (the old ledger
portShape sketch was wrong on this point).
- Look-in cells join the Prepare + partition set so shells have batches
and statics route to ByCell (consumed only by the sub-pass; the main
cell-object pass iterates the main flood's cells).
Register: AP-33 added in the same commit - look-in statics draw WHOLE
(no per-part viewcone; over-include is the safe direction) and look-in
DYNAMICS are deferred (an NPC inside a far building stays invisible -
retail draws objects per overlapped cell in the landscape stage).
Pins: Issue124LookInSeedRegionTests on the real corner-building door -
a seed region containing the aperture floods (and never more than the
full-screen seed), a disjoint region floods NOTHING, and an
interior-side eye never seeds its own exit portal.
Suites: App 259+1skip / Core 1439+2skip / UI 420 / Net 294 green.
Awaiting the user gate: far-building interiors visible through their
apertures from inside; #130 re-gate (top-edge strip) rides the same
launch.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The user's re-gate refuted the scissor fix as THE strip (6c4b6d6 was a
real but sub-pixel under-coverage): the strip survived, screenshot at a
doorway, full width of the opening, top edge only, "very subtle".
Root cause (pinned by Issue130DoorwayStripTests.UnliftedGate_*): the
+0.02 m shell render lift. Cell shells DRAW 2 cm above the dat origin
(z-fight vs coplanar terrain); f35cb8b (the #119-residual fix,
2026-06-11) deliberately reverted the VISIBILITY graph to the physics
(unlifted) transform - but the OutsideView color gate (terrain/sky/
scissor through the doorway) and the seal/punch depth fans are
DRAW-space consumers and kept projecting the unlifted polygons. The
drawn lintel therefore sits one lift-projection above the gate's top
edge - measured 6.7 px at a 2.4 m doorway - and that band never
receives terrain/sky color while the seal also stamps 2 cm low.
A regression from f35cb8b, NOT from the W=0 clip port (987313a stays
exonerated). Vertical aperture edges are immune (the lift slides them
along themselves) - top edge only, exactly as reported; explains the
"also NOW" timing precisely.
Fix - draw space draws lifted, visibility stays physics (the f35cb8b
invariant, now symmetric):
- PortalVisibilityBuilder.Build gains drawLiftZ: the exit-portal branch
projects the OutsideView region with the lifted transform; flood
admission, side tests, and CellViews are untouched (default 0 keeps
every existing visibility test bit-identical).
- The seal/punch fans (DrawRetailPViewPortalDepthWrite) lift their
world verts to the drawn shell's space.
- One shared constant PortalVisibilityBuilder.ShellDrawLiftZ feeds the
shell registration (GameWindow:5604), the gate, and the fans.
Register: AP-32 ADDED - the +0.02 lift had NO row (a pre-register
deviation the 2026-06-12 sweep missed). The row records the split
invariant both ways: a draw-space consumer that forgets the lift
re-opens the #130 strip; a visibility consumer that picks the lifted
transform re-opens the #119-residual side-cull.
Pins: the lifted gate covers the drawn (lifted) aperture to 0.00 px
across the 147-combo sweep; the unlifted gate shows the 6.7 px strip
(sensitivity proof - if the lift is ever removed, this test says the
drawLiftZ plumbing can go too).
Suites: App 257+1skip / Core 1439+2skip / UI 420 / Net 294 green.
Awaiting the user re-gate at a doorway with the lintel on screen.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The user's "doors/doorways leak through terrain and houses over a
landblock" is the #117 mark-pass bias evaluated in the wrong space.
Mechanism (confirmed analytically, Issue129PunchBiasTests): the punch's
pass-A stencil mark biased the aperture fan toward the viewer by a
CONSTANT 0.0005 NDC. NDC depth is non-linear - a constant NDC bias b
spans ~= b*d^2*(f-n)/(f*n) meters of eye depth at eye distance d. With
retail's znear 0.1 (d4b5c71) that is 0.125 m at 5 m but ~190 m at one
landblock: every hill/house in front of a distant aperture passed the
LEQUAL mark and was far-Z punched -> door-shaped leak through the
occluder. This is exactly the risk AD-18's register row recorded
("an occluder within ~bias in front of a distant aperture gets punched
through") - the symptom-scan rule found it before instrumentation.
Fix: cap the bias's EYE-SPACE span at 0.5 m -
biasNdc(d) = min(0.0005, capMeters * near / d^2)
in the mark-pass vertex shader (clipPos.w = eye depth), CPU-mirrored as
PortalDepthMaskRenderer.MarkBiasNdc for tests. Below the ~10 m
crossover the constant-NDC term is smaller and wins - bit-identical to
the T5-validated close-range behavior, so the #108 grass coverage that
justified the bias is untouched. Beyond it the punch can never reach an
occluder more than 0.5 m in front of the aperture plane.
Pins (Issue129PunchBiasTests): the old form spans >100 m of eye depth
at a landblock (the leak, kept as documentation of the refuted shape);
the capped form stays <= 0.5 m at every distance 1-400 m and matches
the validated constant bit-for-bit below 10 m.
AD-18 register row updated in the same commit (bias description + the
#129 closure + the residual risk note: door-hugging geometry beyond the
0.5 m cap at >10 m viewing range re-occludes - the cap constant is the
tuning knob if the gate shows residue).
Suites: App 256+1skip / Core 1439+2skip / UI 420 / Net 294 green.
Awaiting the user visual gate at the original spot (+ #108 cellar
re-check up close).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The register's UN-2 row recorded a contradiction: the GetMaxSpeed XML doc
claimed the bare run rate was retail-correct (~5.9 m/s catch-up, calling
the xRunAnimSpeed multiply a misread), while the implementation multiplied
by RunAnimSpeed citing ACE. Settled against the binary, not the pseudo-C:
- BN pseudo-C (acclient_2013_pseudo_c.txt:305127) renders get_max_speed as
void with a bare `this->my_run_rate;` because it DROPS x87 instructions.
- Disassembling the PDB-matched v11.4186 binary at VA 0x00527cb0: all THREE
return paths end `fld <rate>; fmul dword ptr [0x007C8918]; ret`, and the
.rdata dword at 0x007C8918 is 4.0f. Sibling get_adjusted_max_speed
(0x00527d00) carries the same trailing fmul. Verifier committed at
tools/verify_un2_fmul.py (PE parse + byte decode, rerunnable).
- Retail paths: weenie null -> 1.0 x4; InqRunRate ok -> queried x4;
InqRunRate failed -> my_run_rate x4. ACE MotionInterp.cs:665-676 matches.
Changes:
- Doc-comment rewritten: the implementation is retail-correct; the catch-up
speed 2 x get_max_speed ~= 23.5 m/s at run 200 IS retail. The 1-Hz
remote-blip symptom the old comment attributed to this multiply is
therefore UNEXPLAINED by it (if it recurs: #41 family, not this).
- Weenie-null path aligned to retail's LITERAL 1.0 default (was MyRunRate).
- Tests re-pinned to the three retail paths (the old NoWeenie test pinned
the non-retail fallback).
- Register: UN-2 row deleted per the retire rule (6 -> 5 UN rows);
shortlist renumbered.
This is the 2nd confirmed instance of the BN x87-dropout artifact class
(memory: feedback_bn_decomp_field_names) deciding a register row.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
