Commit graph

9 commits

Author SHA1 Message Date
Erik
fed838847b chore(cli): dump-font-atlas tool for headless font inspection
A `dump-font-atlas` subcommand renders a dat Font's fg/bg atlases (alpha as
luminance) plus a sample string composited exactly the way DrawStringDat does
it (outline + fill, integer-snapped). Used to reproduce the glyph-baseline
jitter offline (fractional-origin bug vs the fix) without launching the client.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-16 15:24:37 +02:00
Erik
26cb34f126 @
docs(D.2b): chat-window re-drive design spec + list-ui-layouts research tool

Plan-2 chat piece of the LayoutDesc importer. Identifies the chat window as
LayoutDesc 0x21000006 (gmMainChatUI, element class 0x10000041) and grounds a
faithful, data-driven re-drive in the named retail decomp (ChatInterface +
gmMainChatUI + UIElement_Text/_Scrollable/_Scrollbar/_Menu) plus a user-provided
retail screenshot.

Design (full-faithful scope, user-approved):
- transcript = UIElement_Text 0x10000011 (dat font, bottom-pinned, 10k behead cap,
  pixel scroll, 1 line/wheel-notch)
- scrollbar = right-side track 0x10000012 + thumb 0x1000048c + up/down
- input = editable UIElement_Text 0x10000016 (caret, 100-entry history, Enter/Send)
- channel menu = UIElement_Menu 0x10000014 ("Chat" selector -> active channel)
- shared ChatCommandRouter extracted from ChatPanel
- screenshot correction: the four 0x10000522-525 left-edge elements are the
  numbered CHAT TABS (1-4), not scroll buttons (a research-agent inference the
  retail screenshot refutes)
- deferred (need non-UI plumbing, each gets a divergence row): tab switching/
  filtering, squelch, clickable name-tags, in-element word-wrap, styled runs,
  font config, opacity transition

Tooling: AcDream.Cli `list-ui-layouts <datdir> [0xRootType]` — read-only index of
every UI LayoutDesc by root element class + size + element-Type histogram; how the
chat layout was located (root type 0x10000041). Reusable for future panel re-drives.

Spec: docs/superpowers/specs/2026-06-15-chat-window-redrive-design.md

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@
2026-06-15 19:38:27 +02:00
Erik
0f55599ba5 feat(D.2b): draw the window resize-grip overlay (gold ridges + corner studs)
The retail vitals window border is TWO layers, not one: the bevel chrome
(0x060074BF-C6) PLUS a resize-grip overlay on top — gold ridged edge strips
and a square corner stud at each corner. acdream only drew the bevel, so the
border looked plainer than retail and the corners lacked the little square
sprite the user spotted.

The overlay ids come from the vitals LayoutDesc 0x2100006C (elements
0x1000063B-0x10000642): corner stud 0x06006129 (same 5x5 at all four corners),
edge strips 0x0600612A/2C (top/bottom) and 0x0600612B/2D (left/right). They
have transparent gaps so the bevel shows through — both layers are drawn.
UiNineSlicePanel now draws the grip overlay (edges tiled via the existing
UV-repeat, corner studs 1:1) after the bevel, so every retail-chrome window
(vitals + chat) gets it.

Verified the grip sprites + the composited result headlessly: dump-sprite-sheet
(new CLI: composite arbitrary sprite ids magnified) showed 0x06006129 is a gold
stud and 0x0600612A-D are gold ridged strips; render-vitals-mockup now renders
the faithful default window with the overlay.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-15 11:05:18 +02:00
Erik
73468be02a fix(D.2b): tile the vital-bar middle instead of stretching it
Retail repeats the bar's "fill-tile" graphic at native width (verified:
the dat element 0x100000E9 is literally the fill-tile; the engine fills via
ImgTex::TileCSI; and a widened side-by-side shows retail tiling, not
stretching). acdream was stretching one copy of the middle slice across the
whole span, so the bevel/bead pattern smeared as the window widened.

UiMeter.DrawHBar now UV-repeats each slice at its NATIVE width: caps span one
native width (a single 1:1 copy), the wide middle spans many (it tiles, last
copy UV-cropped). This works because the UI textures are already GL_REPEAT-
wrapped (TextureCache.UploadRgba8) — the exact mechanism UiNineSlicePanel's
chrome border already uses, so the border edges were ALREADY tiling and need
no change. One draw call per slice; composes with the existing fill-fraction
clip (the partial last tile shows a partial bead).

render-vitals-mockup now renders a widened window twice (stretch vs tile) so
the difference is verifiable headless. Confirmed the tile repeats seamlessly
(no seams).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-15 10:39:56 +02:00
Erik
ff29787f12 fix(D.2b): vitals from the real stacked-window LayoutDesc (0x2100006C)
The vitals bars were rendered from the WRONG layout. The ids in vitals.xml
(0x0600113x) belong to LayoutDesc 0x21000014 -- the 800x28 floaty side-vitals
ROW. The stacked vitals window the user sees is LayoutDesc 0x2100006C
(160x58), which uses a different sprite set and geometry. Dumped the real
tree (new dump-vitals-layout CLI, reflective) and ported it:

- Sprites (#2): the stacked-window set 0x0600747E-0x0600748F (health/stamina/
  mana, each back+front 3-slice; caps 10px, mid 130px).
- Right cap (#1) + fill model: retail UIElement_Meter::DrawChildren draws the
  back 3-slice full then the front 3-slice CLIPPED to the fill fraction (its
  own right-cap shows at 100%, the back's shows through when partial). UiMeter
  now clips the front per-slice (UV-crop) instead of growing a capless slice.
- Spacing (#5): three flush 150x16 bars at y=5/21/37 in a 160x58 window
  (16px pitch, zero gap), per the dat rects -- not the old 20px-apart guess.
- Border (#3): the window is the 8-piece chrome frame (corners 0x060074C3-C6,
  edges 0x060074BF-C2, 5px) -- dat-confirmed identical to RetailChromeSprites.

The headless render-vitals-mockup now composites this exact window
(0x2100006C) from the real sprites with the same clipped-fill model, so the
look was verified before launch. Font (#4, dat Font 0x40000000) is the next
commit.

Decomp refs: gmVitalsUI::PostInit @0x4bfce0; UIElement_Meter::DrawChildren
@0x46fbd0 (scissor-fill); geometry from LayoutDesc 0x2100006C.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-14 22:50:17 +02:00
Erik
1453ff7da2 feat(D.2b): retail 3-slice vital bars + headless mockup verifier
Render each vital bar as a horizontal 3-slice from the real retail
RenderSurface sprites (authoritative ids from the vitals LayoutDesc
0x21000014 via dump-vitals-bars): a fixed-width bevelled left-cap, a
stretched glassy-gradient middle, and a fixed-width right-cap. The
empty back track draws full width; the coloured front fill grows from
the left to the value (the track owns the right end, so the fill omits
its own right-cap). Replaces the flat single-sprite Alphablend overlay
that read as the old UI - this is the bordered gradient look from the
retail screenshot (red HP / gold stamina / blue mana).

UiMeter gains the six 9-slice ids (BackLeft/Tile/Right +
FrontLeft/Tile/Right) and a DrawHBar helper; MarkupDocument parses the
backleft/backtile/backright/frontleft/fronttile/frontright attrs;
vitals.xml carries the 18 per-vital ids. The temporary
ACDREAM_BAR_PROVEOUT component grid is removed.

Adds AcDream.Cli render-vitals-mockup: a headless ImageSharp composite
that assembles the bars with the SAME DrawHBar logic, so the sprite
assembly can be verified by eye (Read the PNG) without launching the
client + server - the fast UI-iteration loop the user asked for.
export-ui-sprite dumps a single RenderSurface to PNG for HTML mockups.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-14 21:40:11 +02:00
Erik
56ee5eff60 chore(D.2b): CLI dump-vitals-bars — read vitals LayoutDesc meter sprites
Adds `AcDream.Cli dump-vitals-bars <datDir>` subcommand that:
- Scans all 101 LayoutDesc objects in client_local_English.dat
- Finds the vitals window layout (0x21000014) by locating the Health
  meter element id 0x100000E6 (from gmVitalsUI::PostInit decomp)
- Walks each meter's sub-element tree (typed access via ElementDesc.Children,
  ElementDesc.States, ElementDesc.StateDesc, StateDesc.Media, MediaDescImage.File)
- Prints every RenderSurface DataId (0x06xxxxxx) per vital

Authoritative output:
  HEALTH  (0x100000E6): front-bar fill 0x06005F3D / track fill 0x06005F3C
                        E8/E9/EA pieces: 0x06001131/32/33, 0x06001141/40/3F
  STAMINA (0x100000EC): front-bar fill 0x06005F3F / track fill 0x06005F3E
                        E8/E9/EA pieces: 0x06001137/38/39, 0x06001147/46/45
  MANA    (0x100000EE): front-bar fill 0x06005F41 / track fill 0x06005F40
                        E8/E9/EA pieces: 0x06001134/35/36, 0x06001144/43/42

LayoutDesc shape discovered: Fields Width, Height, Elements (HashTable<uint,ElementDesc>).
ElementDesc shape: ElementId, Type, BaseElement, BaseLayoutId, DefaultState,
  X/Y/Width/Height/ZLevel, LeftEdge/TopEdge/RightEdge/BottomEdge,
  States (Dictionary<UIStateId,StateDesc>), Children (Dictionary<uint,ElementDesc>),
  StateDesc (direct single state).
StateDesc shape: StateId, PassToChildren, IncorporationFlags,
  Properties (Dictionary<uint,BaseProperty>), Media (List<MediaDesc>).
MediaDescImage shape: File (uint DataId), DrawMode.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-14 19:39:33 +02:00
Erik
d3165f99d7 feat(vfx): Phase E.3 particle system + hook wiring + registry
Full runtime particle pipeline consuming Phase E.1's animation hooks.
13 motion integrators, per-emitter particle pools with overwrite-oldest
eviction, colour / scale / alpha interpolation over life, and a
ParticleHookSink routing CreateParticle / DestroyParticle / StopParticle /
CreateBlockingParticle hooks from the animation-hook router.

Core layer:
- ParticleSystem: handle-based emitter pool, per-tick emission
  accumulator (retail Birthrate = time-between-spawns → our emit rate
  via 1/B), 13 integrators covering the full ParticleType enum:
  Still, LocalVelocity, GlobalVelocity, 7 Parabolic variants (all
  apply Gravity * dt to velocity), Swarm (orbital drift),
  Explode (outward from anchor), Implode (inward to anchor, dies at
  convergence).
- EmitterDescRegistry: id-keyed EmitterDesc cache with fallback-to-
  default for unknown ids. Replaces the dat-loaded path until
  Chorizite.DatReaderWriter exposes ParticleEmitterInfo (v2.1.7 does
  not; upgraded from 2.1.4 anyway for future types).
- ParticleHookSink: wires the full hook family:
  - CreateParticleHook → SpawnEmitterById at entity pose + hook offset
  - CreateBlockingParticleHook → marker only (blocking semantics live
    in the sequencer not here)
  - DestroyParticleHook → StopEmitter(handle, fadeOut=false)
  - StopParticleHook   → StopEmitter(handle, fadeOut=true)
  - (Default/CallPES deferred until PhysicsScript dat is loadable)

GameWindow integration:
- ParticleSystem created eagerly (no driver dep), sink registered with
  hook router, Tick advanced per OnRender frame after animation tick so
  hooks fired this frame get integrated.

Tests (11 new): spawn-handle, emit-over-time steady state, lifetime
death curve, LocalVelocity movement, Parabolic gravity arc, Explode
outward trajectory, StopEmitter instant kill vs fadeOut, MaxParticles
cap enforcement, registry default fallback, registry custom
registration.

Upgraded Chorizite.DatReaderWriter 2.1.4 → 2.1.7 across Core + Cli.

Build green, 508 tests pass (up from 497).

Ref: r04 §2 (CParticleManager), §3 (13 integrators), §6 (PhysicsScript).
Renderer (instanced billboarded quads in translucent pass) ships next
commit; this one covers the data / logic / wiring layer in full.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-18 16:48:17 +02:00
Erik
020ec2a35d chore: phase 0 — skeleton + dat asset inventory
Brand-new solution targeting .NET 10, using Chorizite.DatReaderWriter 2.1.4
to walk a retail AC dat directory and print how many of each asset type live
in client_portal / client_cell_1 / client_highres / client_local_English.

Opens the four dats in ~16 ms and counts 887,381 indexed assets across 40+
tracked DBObj types. Cell-database terrain (LandBlock, LandBlockInfo, EnvCell)
uses mask-based IDs that DatReaderWriter 2.1.4's GetAllIdsOfType<T> does not
support; worked around with a manual b-tree walk in CountCellByLow16.

Sanity check: LandBlock count is 65,025 = 255 x 255, exactly the AC world grid.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-10 09:02:56 +02:00