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>
Keep AttackDone control statuses out of chat, preserve dead motion across zero-velocity position updates, render the authored power meter from right to left, and retain the rotatable viewer offset during target tracking.
Co-Authored-By: Codex <codex@openai.com>
Port ClientCombatSystem's Keep in View consumer through CameraSet's pivot-to-target boom heading, including the transformed half-metre target offset and melee/missile validity gates. Preserve the existing stateful chase damping and collision architecture, with conformance coverage and named-retail pseudocode.
Co-Authored-By: Codex <codex@openai.com>
The camera-collision sweep strobed the eye 0.27m every ~5-10 frames while
the compressed chase boom moved along corridor walls (pulledIn 0.27<->0.53
on ~1.4mm input drift): RetailChaseCamera re-demanded the FULL-length ideal
boom from scratch each frame, so the pivot->eye ray re-rolled the same
knife-edge r+-eps graze on the double-faced slabs every frame, and its two
first-contact solutions tear-interleaved at ~1700fps into the #176
"stripes/triangles".
Retail never re-rolls that ray. CameraManager::UpdateCamera (0x00456660)
interpolates FROM THE CURRENT SWEPT VIEWER toward the desired pose
(interpolate_origin/rotation, stiffness 0.45 x dt x 10, clamped) and the
result becomes viewer_sought_position (SmartBox::PlayerPhysicsUpdatedCallback
0x00452d60); update_viewer (0x00453ce0) sweeps pivot->SOUGHT. Pressed against
a wall the sweep ray extends one interpolation step past the contact
(sub-mm at high fps), so a bistable graze can move the eye by at most that
step - the strobe is structurally impossible. A 0.4mm/2e-4 dead-band parks
the sought exactly on the viewer when converged (0x00456fcd-0x00457035).
- RetailChaseCamera: _dampedEye -> _soughtEye + _publishedEye (retail's two
Positions); lerp base = the published (swept) viewer; sweep targets the
sought; total-fallback (ViewerCellId==0) resets the sought like
set_viewer(player_pos, 1). The old "collision must NOT feed back into the
damped state" comment had the coupling backwards - what stays clean is the
transient desired pose, not the sought.
- SweepEye untouched (faithful update_viewer port, exonerated by the #180
investigation).
- Tests: the old pin asserting instant full re-extension after a clamp
(the divergence itself) replaced with four retail pins: gradual
re-extension, sweep-target-converges-onto-contact, total-fallback
re-extends from the player, wall-press glide stability.
- Pseudocode doc: docs/research/2026-07-06-camera-sought-position-pseudocode.md
(UpdateCamera tail incl. the sought derivation + set_viewer reset semantics
+ Frame interpolate/close_rotation).
- Register: AD-37 (forward-vector nlerp vs quaternion slerp), AD-38
(init-at-full-extension vs retail re-extend-from-player) - both
pre-existing, identified during the decomp reading.
Suites green (Core 2599+2skip / App 729+2skip / UI 425 / Net 385).
Pending: autonomous visual verify + user gate (#180 + the #176 re-gate).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The 137b4f2 payload, re-landed now that #110 is resolved: the missing-indoor-
textures correlation was the pre-existing #105 staged-texture-flush drop
(fixed in c787201), not a near-plane mechanism. znear=0.1 merely raised #105's
trigger probability — a closer near plane makes close-up geometry newly
visible, inflating per-frame prepare/upload pressure indoors and growing the
never-flushed tail. Exactly the handoff's only-credible-link hypothesis,
verified instead of assumed.
Retail: Render::SetFOVRad sets znear=0.1 flat (decomp :342173, initializer
:1101867). 0.1 < the 0.3m camera-collision sphere, so a wall the collided eye
presses against no longer falls inside the near plane — the §4 corner
see-through-wall closes.
Verification on the 0.1 arm (the arm that struck 2-of-3 on 2026-06-10):
nearplane-reland-1.log — [tex-flush] after=0 on all 45 lines, 68,291 [shell]
lines with zero zh>0 batches, all four dat tripwires silent, no [wb-error].
ISSUES.md: #105 + #110 moved to Recently closed with root cause + evidence.
Pending user re-gate: corner press (wall stays solid) + distance scan for
z-shimmer (none expected; retail ships 0.1 with D24).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Bisect (user-gated): two consecutive runs on 0.1 lost indoor textures; the 1.0 bisect run rendered clean. #105 tripwires silent on the bad runs (GL-side). No known mechanism links the near plane to texturing - #110 filed to investigate (RenderDoc / flip-testing) before re-landing retail's znear=0.1, which the corner see-through fix depends on. Comments on all four cameras point at #110 so the retail value is not re-landed blind.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The collided camera eye sits 0.3m from walls (viewer_sphere radius); a 1.0m near plane clipped the wall face away, so pressing the camera into a corner showed the clear color through the wall (gate result: unchanged by the flood fix - it was never a flood bug). Retail sets Render::znear = 0.1 flat in SetFOVRad (decomp :342173, initializer :1101867). All four cameras aligned. Also files #107 (indoor spawn wedge, 3-for-3), #108 (cellar-up terrain sweep across door opening), #109 (exit-door texture/background oscillation) from the 2026-06-10 visual gate; gate confirms the dac8f6a flood fix: room-room + indoor-outdoor transitions clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Port retail CameraManager::UpdateCamera's convergence snap (0x00456fcd):
once the per-frame lerp step is below 0.0004 m AND the rotation within
0.000199999995, freeze the damped eye at an exact fixed point instead of
Vector3.Lerp's endless sub-mm asymptote. The drift was walking the 3rd-person
eye across the vestibule/room portal plane at rest, flipping the per-frame
viewer-cell resolve 0170<->0171 -> the indoor grey/texture flicker. The
collided-eye firewall (separate publishedEye local) is already present.
Adds ApplyConvergenceSnap static (TDD: 3 unit tests + 1 integration freeze
test) + SnapEpsilon/RotCloseEpsilon. App suite 183 -> 187, all green.
Plan: docs/superpowers/plans/2026-06-05-indoor-viewer-cell-flicker-fix.md
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Complete Render Residual A's faithful port: PhysicsCameraCollisionProbe.SweepEye
now mirrors SmartBox::update_viewer (acclient_2013_pseudo_c.txt:92761) end-to-end:
- Start cell (pc:92824-92844): indoor (>=0x100) seats the sweep at the head-PIVOT
via PhysicsEngine.AdjustPosition (the cellar-lip case — feet in the low connector,
head up at floor level); outdoor keeps the player cell.
- Sweep pivot -> sought-eye from the seated start cell (unchanged 0x5c viewer flags).
- Success (pc:92870): set_viewer(curr_pos), viewer_cell = curr_cell.
- Fallback 1 (pc:92878): AdjustPosition(sought_eye).
- Fallback 2 / no-cell (pc:92775, 92886): snap to player, viewer_cell = null. This
also makes cellId==0 faithful (was returning the desired eye; retail snaps to
player_pos) and adds the playerPos arg to ICameraCollisionProbe.SweepEye.
Supporting: ResolveResult.Ok surfaces FindTransitionalPosition's return (retail
find_valid_position != 0, pc:273898) so SweepEye knows when to fall back.
TDD: 11 new tests (FindVisibleChildCell 4, AdjustPosition 3, ResolveResult.Ok 2,
SweepEye orchestration 2). The seating test's RED proved the sweep does NOT auto-
advance feet->room, so the pivot-seated start cell is genuinely decisive. Core
1326 pass / 4 documented-fail / 1 skip; App 179 pass / 0 fail. No regression.
Per the live-capture finding, the visible payoff is the cellar-corner (point 3);
the cottage-room bluish void stays for residual C. Spec:
docs/superpowers/specs/2026-06-05-residual-a-camera-collision-design.md.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Update() now always sets ViewerCellId: the camera-collision sweep's swept cell when collision
is on (retail viewer_cell = sphere_path.curr_cell), else the passed player cell. This is the
robust, per-frame, graph-tracked 'which cell is the camera in?' answer that V1 roots the render
on — no AABB, no grace frames (the U.4c flap source). 176 App tests green (2 new).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The camera spring-arm sweep already resolves the collided eye's cell (ResolveResult.CellId
= sp.CurCellId = retail viewer_cell = sphere_path.curr_cell, update_viewer pc:92871).
Return it from SweepEye so the render can root on the viewer cell (Phase W single-viewpoint
V1, Task 1). Pure plumbing — behavior unchanged; callers extract .Eye. 174 App tests green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Visual verification showed the camera vibrating/bouncing when pressed against a
wall. Cause: the sweep wrote its clamped result back into _dampedEye, so the
next frame's damping lerped from the wall toward the target and the sweep
re-clamped it — a per-frame feedback loop. Retail keeps viewer_sought_position
(damped, uncollided) separate from viewer (the published collided eye). Fix:
collide into a separate publishedEye for Position/View/fade and leave _dampedEye
as the clean sought position. New regression test
Update_CollisionDoesNotCorruptDampedState (clamp-then-release → full recovery).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Add ICameraCollisionProbe? CollisionProbe { get; init; } to RetailChaseCamera.
Extend Update() with optional cellId/selfEntityId params (default 0) so all
existing callers compile unchanged. After the exponential-damping block (step 5)
and before publishing Position/View (step 6), sweep _dampedEye through the
probe when CameraDiagnostics.CollideCamera is true and a probe is wired in
(step 5b). The fade computation in step 7 then naturally uses the collided eye.
Null probe and cellId=0 both short-circuit cleanly. Three new xUnit tests
cover: probe-wired+flag-on publishes collided eye, flag-off skips probe,
null probe doesn't throw. All 30 RetailChaseCameraTests pass.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Original symptom: jumping made the camera swing around the player
vertically — the basis tilted up/down with the player's Z velocity.
Root cause: ComputeHeading used the raw 3D velocity vector as the
heading direction. During a jump, velocity has a substantial Z
component (vy ≈ jump speed), and `normalize((vx, vy, vz))` produced
a heading pointing up. The basis tilted accordingly and the camera
went under/over the player.
Retail's actual ALIGN_WITH_PLANE algorithm (decomp at
acclient_2013_pseudo_c.txt:95644-95795) is different:
1. Velocity is only used as a gate. If |vx| AND |vy| > epsilon
(player is moving in XY), proceed; otherwise fall back to the
LOOK_IN_DIRECTION path (player's facing direction unchanged).
2. The base heading is `localtoglobalvec(player, (0, 1, 0))` —
the player's local +Y axis in world space, which in our
convention is `(cos yaw, sin yaw, 0)`.
3. Pick a surface normal:
grounded: contact_plane.N
airborne: (0, 0, 1) [world up]
4. Project the base heading onto the plane perpendicular to that
normal: projected = forward - normal * dot(forward, normal).
5. Normalize. Fall back to the base if projection collapses.
Behaviorally:
* Standing jump (vx≈0, vy≈0): gate fails → base heading. Camera
doesn't move with the jump.
* Running jump (vx, vy, vz all nonzero, airborne): projects onto
world up → no-op since base is already horizontal. Camera basis
stays horizontal; player visibly rises in frame.
* Walking uphill (grounded, slope normal tilted): projection
adds a Z component matching the slope angle. Camera basis tilts
with the terrain.
* Walking on flat ground: projection is a no-op. Camera basis
horizontal.
Surface changes:
* RetailChaseCamera.ComputeHeading gains `isOnGround` and
`contactPlaneNormal` parameters.
* RetailChaseCamera.Update gains the same two parameters and
threads them through.
* GameWindow's two Update call sites pass `result.IsOnGround` and
`_playerController.ContactPlane.Normal` (already exposed on
PlayerMovementController — no plumbing change there).
* Tests: 2 existing heading tests reshaped (Moving* and Uphill);
2 new tests added (AirborneJumping straight-up + running-jump);
1 renamed (SlopeAlignDisabled). Net 25 → 27 tests in
RetailChaseCameraTests; full AcDream.App.Tests: 39 → 41.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
EnterChaseMode now takes (ChaseCamera, RetailChaseCamera); Active
consults CameraDiagnostics.UseRetailChaseCamera to pick which to
expose. Flag flip at runtime swaps cameras instantly (both are kept
warm). GameWindow's two EnterChaseMode call sites get a temporary
stub RetailChaseCamera; Task 7 wires proper construction +
per-frame updates.
Also folds in two minor cleanups from the Task 3 code review:
- Update() discards the unused `right` axis from BuildBasis (no
caller in the chase-cam math; viewer_offset.X is always 0)
- The three CameraDiagnostics-mutating integration tests now
save and restore the static state in try/finally to avoid
ordering-dependent contamination
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds the per-frame Update(playerPos, yaw, velocity, dt) entrypoint
that composes the math primitives into a renderable View matrix +
PlayerTranslucency. State: 5-frame velocity ring, damped eye + forward
unit vector, first-frame snap flag, mouse-filter shared state.
Public surface: Distance/Pitch/YawOffset/PivotHeight tunables,
AdjustDistance/Pitch (with clamps), FilterMouseDelta entry, View +
Position + PlayerTranslucency outputs. 5 new integration tests, all
pass; total RetailChaseCamera test count 25.
Also folds in two minor cleanups from the Task 2 code review:
- AverageVelocity uses ring.Length instead of hardcoded 5
- Basis_NearVerticalHeading test asserts orthogonality of right & up
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Seven pure-math helpers in the new RetailChaseCamera class:
ComputeHeading (slope-align with flat fallback), BuildBasis (heading
→ orthonormal frame, near-vertical fallback), PushVelocity +
AverageVelocity (5-entry FIFO ring), ComputeDampingAlpha (retail's
stiffness*dt*10), FilterMouseAxis (0.25s low-pass), ComputeTranslucency
(linear ramp 0.20..0.45 m). 20 tests, all pass. State machine + Update()
land in the next commit.
Per spec docs/superpowers/specs/2026-05-18-retail-chase-camera-design.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
