docs(D.2b): design spec — retail panel frame + live Vitals (Approach C)

Brainstormed design for the D.2b retail-look UI backend: our own KSML-style
markup + controls.ini stylesheet + retained-mode toolkit on Silk.NET (no
embedded browser, zero external deps — Approach C, chosen over Ultralight/CEF
and RmlUi for memory/dep-weight/faithfulness).

Spec 1 scope: an 8-piece dat-sprite window frame + live Vitals bars bound to
the existing VitalsVM, gated behind ACDREAM_RETAIL_UI=1, rendered via a reused
TextRenderer batch. Render-only (input/hit-test, AcFont glyphs, anchor solver,
LayoutDesc importer all deferred).

Grounded by a read-only research workflow (7 readers + gap-critic). The critic
corrected several stale memory/plan-doc facts now baked into the spec's
do-not-trust list: VitalsVM is a sealed class (not the old record); chrome
sprite IDs are unverified (Step-0 dat prove-out resolves them empirically);
controls.ini exists and #FFDBD6A8 is editbox text not a bg; DatCollection reads
are thread-safe; KSML is rich-text not the layout language (we mirror
ElementDesc).

Phase D.2b / Milestone M5 (parallelizable with M3/M4 — opened as a parallel
track while M1.5 stays the active critical-path milestone). Retires divergence
row TS-30 + adds one IA row when the chrome ships.

Also gitignores the /.superpowers/ visual-companion scratch dir.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Erik 2026-06-14 14:00:14 +02:00
parent f6a30f4aae
commit de9229eed5
2 changed files with 351 additions and 0 deletions

View file

@ -0,0 +1,349 @@
# D.2b — Retail panel frame + live Vitals (Approach C: KSML-style engine) — Design
**Date:** 2026-06-14
**Status:** Design approved (brainstorm), pending spec review → implementation plan
**Phase:** D.2b — Custom retail-look UI backend ([roadmap:427](../../../docs/plans/2026-04-11-roadmap.md))
**Milestone:** M5 "Looks like retail" — **explicitly PARALLELIZABLE with M3/M4** ([milestones:378](../../../docs/plans/2026-05-12-milestones.md)). Opened as a parallel track while M1.5 is the active critical-path milestone; the M5 parallelizable flag is the milestone-discipline carve-out.
**Grounding:** read-only research workflow `wf_39a90d37-e5a` (7 readers + gap-critic, 2026-06-14). Every binding fact below cites `file:line` in `src/` or a named-retail symbol; nothing rests on a memory note alone.
---
## 1. Context & goal
acdream needs a retail-faithful game UI. The shipped path (D.2a) is an ImGui
overlay gated on `ACDREAM_DEVTOOLS=1` — a debugger aesthetic, intentionally
temporary. D.2b replaces the *visual layer* with our own toolkit that draws
retail's actual dat assets and matches retail's look, while the stable
`AcDream.UI.Abstractions` contracts (ViewModels, Commands, `IPanel`) stay
unchanged underneath.
**The user's framing (2026-06-14):** AC's UI engine is Keystone — and Keystone
was *already* markup + stylesheet (KSML, an HTML-clone XML defined by `ksml.xsd`,
+ `controls.ini`, a CSS-like INI stylesheet). So "make it look + behave like
retail, but author it in a CSS/HTML-style way" is not a foreign graft — it
re-expresses AC's own design in its modern equivalent.
**Approach decision (Approach C).** Three integration families were weighed:
(A) embed a real web engine (Ultralight/CEF), (B) a native HTML/CSS-subset lib
(RmlUi), (C) our own KSML-style markup + stylesheet over a retained-mode toolkit
on Silk.NET. **C chosen** for: zero external deps (keeps the native-AOT
distribution goal intact), lowest memory (~310 MB vs CEF's 150300 MB), full
control, and maximal architectural faithfulness — it mirrors Keystone directly.
The cost (most code to write) is acceptable because the engine is ours forever
and the plugin API (a day-1 core constraint) gets a clean markup authoring
surface.
This spec covers **Spec 1**: the engine skeleton + the plugin-facing markup
contract, proven end-to-end on **one** real panel — the universal window frame
wrapping the live Vitals bars.
## 2. Scope
**In Spec 1:**
- A new retail-look backend in `src/AcDream.App/UI/Retail/` implementing
`IPanelHost` + `IPanelRenderer` from `AcDream.UI.Abstractions`.
- The 8-piece dat-sprite window frame (4 corners + 4 edges + center fill), a
title bar, and a *drawn* close button.
- Three live vital bars bound to the existing `VitalsVM`.
- The XML markup format (mirrors `ElementDesc`) + a minimal `controls.ini`
stylesheet loader.
- The plugin-facing contract: `IPanelRegistry` on `IPluginHost` + a `MarkupPanel`
shim, so the engine is plugin-ready by construction.
- A new `RuntimeOptions.RetailUi` toggle (`ACDREAM_RETAIL_UI=1`).
**Deferred to later sub-phases (explicitly OUT):**
- Input / hit-testing (window drag, working close-click). Spec 1 is **render-only**.
- The dat A8 glyph font loader (`AcFont`) → numeric overlays ("182/210").
- The full anchor solver (`StateDesc::UpdateSizeAndPosition` port).
- The `LayoutDesc` binary importer (sub-project 3).
- Reskinning Chat / Debug / Settings panels.
- Login / char-select / chargen screens (raw-JPEG backgrounds, sub-project 4).
- Extraction into a standalone `src/AcDream.UI.Retail/` project (see §4).
## 3. What the grounding corrected (do-not-trust list)
The research caught several load-bearing "facts" that were wrong or unverified.
These are binding:
| Claimed (memory / plan doc) | Reality (source-verified) |
|---|---|
| `VitalsVM` is `record VitalsVM(int HpCurrent, …)` | It is a **sealed class**: `HealthPercent` (float), `StaminaPercent`/`ManaPercent` (float?), `*Current`/`*Max` (uint?), ctor `(CombatState, LocalPlayerState?)`, `SetLocalPlayerGuid(uint)` — [VitalsVM.cs:35](../../../src/AcDream.UI.Abstractions/Panels/Vitals/VitalsVM.cs) |
| Chrome sprite IDs `0x06004CC2` / `0x21000040` / `0x060074BF..C6` are known | **Unverified + contradictory.** A second reader cited `0x06001125` etc. from a *non-existent* file; `0x06001125` is actually the char-select highlight. **No chrome ID is trusted — Step 0 dat prove-out resolves them empirically.** |
| `#FFDBD6A8` "parchment cream" is the panel background | It is the `[editbox]`/`[treeview]` **text** color. Real frame tokens: `[title]` bg `#FFFFFFFF`, font `Verdana-10-bold`, height 19; `[body]` bg `#00000000` (transparent), `color_border=#FF4F657D` |
| `DatCollection` is NOT thread-safe | Concurrent **reads are safe** since v2.1.7 (1.1M-read hammer test, [2026-06-09 investigation](../../../docs/research/2026-06-09-dat-reader-thread-safety-investigation.md)). No UI-specific lock. |
| KSML is the panel-layout language | KSML is **rich-text-in-text-regions**; the panel layout format is the binary `LayoutDesc`/`ElementDesc` tree ([acclient.h:33693](../../../docs/research/named-retail/acclient.h)). Our markup mirrors `ElementDesc`, not KSML. |
## 4. Architecture & placement
The docs name a future `src/AcDream.UI.Retail/` project, but the three pieces we
must reuse — `TextureCache`, `TextRenderer`, `Shader` — all live in
**`AcDream.App`** ([TextureCache.cs:70](../../../src/AcDream.App/Rendering/TextureCache.cs),
[TextRenderer.cs](../../../src/AcDream.App/Rendering/TextRenderer.cs),
[Shader.cs](../../../src/AcDream.App/Rendering/Shader.cs)). A separate project
cannot reach them without first extracting a shared rendering-primitives project
(a large, unrelated refactor). Unlike `AcDream.UI.ImGui` (which needs only the
ImGui packages), the retail backend needs dat sprites, which are App-resident.
**Decision:** Spec 1 builds the backend in **`src/AcDream.App/UI/Retail/`** as
dedicated classes. This honors Code-Structure Rule 1 (nothing substantial added
to `GameWindow.cs`'s body — only a few wiring lines), Rule 2 (Core stays
GL-free), and Rule 3 (panels still target `AcDream.UI.Abstractions`; the backend
*implements* the host/renderer contracts). The clean `AcDream.UI.Retail` project
extraction is a follow-up, gated on a rendering-primitives home existing.
```
┌──────────────────────────────────────────────────────────┐
│ retail dat (read-only fidelity source) │
│ controls.ini → style tokens · RenderSurface 0x06xxxxxx │
│ → sprites · Font 0x40xxxxxx → glyphs (deferred) │
└───────────────┬──────────────────────────────────────────┘
│ assets via TextureCache.GetOrUpload
┌───────────────▼──────────────────────────────────────────┐
│ NEW: src/AcDream.App/UI/Retail/ │
│ RetailPanelHost : IPanelHost │
│ RetailPanelRenderer : IPanelRenderer (+ chrome) │
│ UiSpriteBatch (wraps TextRenderer + UV-rect quads) │
│ NineSlice (8 pieces + center) · ControlsIni (parser) │
│ MarkupDocument (XML → ElementDesc-shaped tree) │
└───────────────┬──────────────────────────────────────────┘
│ VMs out / Commands in (unchanged)
┌───────────────▼──────────────────────────────────────────┐
│ AcDream.UI.Abstractions (exists) — IPanel/IPanelHost/ │
│ IPanelRenderer/ICommandBus/PanelContext/VitalsVM │
└───────────────┬──────────────────────────────────────────┘
┌───────────────▼──────────────────────────────────────────┐
│ game state (unchanged) — CombatState, LocalPlayerState │
└──────────────────────────────────────────────────────────┘
```
**Coexistence with ImGui.** The retail pass renders in the same post-3D slot as
ImGui's `Render()` ([GameWindow.cs:8232](../../../src/AcDream.App/Rendering/GameWindow.cs)),
with deterministic ordering. `ACDREAM_RETAIL_UI=1` activates the retail Vitals
panel; `ACDREAM_DEVTOOLS=1` keeps the ImGui overlay (Chat/Debug/Settings) working
with **no regression**. Both may be on at once during development.
## 5. Render foundation — reuse, don't rebuild
`IPanelRenderer` is a 34-method, ImGui-shaped immediate-mode API; `Begin(string
title)` carries no position/size/sprite/style ([IPanelRenderer.cs:23](../../../src/AcDream.UI.Abstractions/IPanelRenderer.cs)).
It is **structurally incompatible** with positioned, chrome-decorated retail
windows, so the markup engine does **not** route chrome through it. Instead:
- **`UiSpriteBatch` wraps `TextRenderer`** — which already provides pixel→NDC
conversion ([ui_text.vert:12](../../../src/AcDream.App/Rendering/Shaders/ui_text.vert)),
dynamic VBO growth, and a save/restore pattern. We add a **source-UV-rect
parameter** to its quad path so one sprite can be sliced into 8 border pieces.
- **Extend `ui_text.frag`** with `uUseTexture=2` (RGBA sampling) for dat sprites;
it currently does only solid-color (`0`) and R8 coverage (`1`)
([ui_text.frag:9](../../../src/AcDream.App/Rendering/Shaders/ui_text.frag)). ~3-line edit.
- Use the simple `Shader` class, **not** `GLSLShader` — no bindless promotion,
uniform cache, or `QueueGLAction` teardown is needed for a synchronous
main-thread 2D pass.
- **Self-contained GL state** (project rule [feedback_render_self_contained_gl_state]):
the pass explicitly sets blend (`SrcAlpha/OneMinusSrcAlpha`), `DepthTest` off,
**`DepthMask(false)`** (TextRenderer omits this today —
[TextRenderer.cs:171](../../../src/AcDream.App/Rendering/TextRenderer.cs)),
`CullFace` off, scissor — and restores them. It must not inherit state from the
3D pass or ImGui.
## 6. Dat assets & the Step-0 prove-out gate
`TextureCache.GetOrUpload(uint surfaceId)` returns a conventional `Texture2D`
handle (not the bindless `Texture2DArray` the world MDI path uses) — exactly
right for the UI batch ([TextureCache.cs:70](../../../src/AcDream.App/Rendering/TextureCache.cs)).
The decode chain (Surface → SurfaceTexture → RenderSurface → RGBA8) and the
`PFID_*` formats (incl. `PFID_A8`) already work ([SurfaceDecoder.cs:39](../../../src/AcDream.Core/Textures/SurfaceDecoder.cs)).
**Step 0 is empirical and comes first.** Because no chrome sprite ID is verified,
the first implementation task loads each candidate ID (`0x06004CC2`,
`0x060074BF..C6`, `0x0600129C`, …) via `GetOrUpload`, draws each as a raw quad,
and visually confirms which decode to frame-shaped art vs magenta vs the wrong
sprite. The confirmed IDs are recorded in the spec's follow-up and in code
comments before any layout code is written. **No ID is hardcoded on faith.**
The frame is **8 quads + a center fill**, not one stretched 9-slice texture: 4
corner sprites, 4 edge sprites (tiled or stretched along their axis), and a
separate center-fill sprite. Slice/edge metrics are a **documented stopgap
constant** (with a divergence row) until the `LayoutDesc` tree is parsed
(sub-project 3) to supply the real insets.
## 7. Markup + stylesheet model
**Markup** mirrors `ElementDesc` 1:1 ([acclient.h:33693](../../../docs/research/named-retail/acclient.h)):
an element has a type, id, `x/y/w/h/z`, the four anchor edge-codes, a
`defaultState`, and a media list (sprite DataIDs). Example authoring shape:
```xml
<panel id="acdream.vitals" x="10" y="30" w="220" h="96" title="Vitals">
<meter id="health" x="8" y="24" w="200" h="13" fill="{HealthPercent}" color="#FF0000"/>
<meter id="stamina" x="8" y="44" w="200" h="13" fill="{StaminaPercent}" color="#10F0F0"/>
<meter id="mana" x="8" y="64" w="200" h="13" fill="{ManaPercent}" color="#0000FF"/>
</panel>
```
This is deliberately the shape the future `LayoutDesc` importer will *emit*, so
the authoring format and the imported format converge. It is **not** KSML —
KSML is reserved for rich-text content inside text regions (deferred).
**Anchor codes** are *defined* now (0=fixed, 1=stretch, 2=proportional-translate,
3=center, 4=proportional-scale — from `StateDesc::UpdateSizeAndPosition` @`0x0069BF20`)
but the **solver is deferred**: the Vitals window is fixed-size, so Spec 1 places
it at fixed pixel coords. Building the full solver now would be gold-plating
(gap-critic risk #7).
**Stylesheet.** A small INI loader parses `controls.ini` keyed by element-type
section, honoring the `#AARRGGBB` color format (alpha-first) and the
`font://Face-Pt[-style]` font URI. The cascade is: element-type defaults from
ini → per-element `class=` section → inline attributes. `controls.ini` is
**optional** (see §10): if the AC install is absent, the real `[title]`/`[body]`
token values are baked as fallback.
## 8. VM binding (the Vitals slice)
Bind to the **real** `VitalsVM``HealthPercent` / `StaminaPercent` /
`ManaPercent` ([VitalsVM.cs:67](../../../src/AcDream.UI.Abstractions/Panels/Vitals/VitalsVM.cs)).
The VM already does all server plumbing (CombatState + LocalPlayerState, updated
from the wire), so we do **not** re-derive vitals from the retail
`gmVitalsUI`/`CACQualities` decomp.
Each bar uses the retail **scissor-fill** technique: draw the empty background
rect, set scissor to the bottom `pct * height` pixels, draw the filled rect.
Colors Health `#FF0000`, Stamina `#10F0F0`, Mana `#0000FF`. This uses only the
solid-color shader path (`uUseTexture=0`) — **no dat font needed**. The
`StaminaPercent`/`ManaPercent` nullable case (null until `PlayerDescription`
arrives) renders an empty bar.
The Vitals panel is constructed and registered the same way as today — built in
the live-session path and given the player GUID at EnterWorld via
`SetLocalPlayerGuid` ([GameWindow.cs:1984](../../../src/AcDream.App/Rendering/GameWindow.cs)) —
but registered into `RetailPanelHost` instead of `ImGuiPanelHost` when
`ACDREAM_RETAIL_UI=1`.
## 9. Plugin contract (designed now, first consumer first-party)
`IPluginHost` exposes only `Log`/`State`/`Events` today — no UI surface
([IPluginHost.cs:9](../../../src/AcDream.Plugin.Abstractions/IPluginHost.cs)). Spec 1
adds:
- `IPanelRegistry Panels { get; }` on `IPluginHost` — a one-method
`void Register(IPanel)` wrapper over `IPanelHost.Register` (does **not** expose
`RenderAll` to plugins).
- A `MarkupPanel(string id, string title, string markupPath, object binding)`
`IPanel` implementation: owns a parsed `MarkupDocument` + a binding object whose
properties the `{Binding}` expressions resolve against.
- ALC note: if `AcDream.UI.Abstractions` types cross the plugin boundary, add it
to the host-shared exclusion set alongside `AcDream.Plugin.Abstractions`
([PluginAssemblyLoadContext.cs:13](../../../src/AcDream.Core/Plugins/PluginAssemblyLoadContext.cs)).
- Registrations from `IAcDreamPlugin.Enable()` (main thread, before the GL window
opens) buffer into a list the host drains into `RetailPanelHost` after init —
the threading concern lives in the host, the plugin call is unconditional.
The first consumer is the first-party Vitals panel, but the contract lands here
so the markup format is designed against a real plugin path rather than
retrofitted. Wiring an actual plugin-supplied panel end-to-end is a thin
follow-up.
## 10. Confirmed decisions (approved 2026-06-14)
1. **Render-only first slice.** Frame + live (updating) bars; the close button is
drawn-not-clickable and the window is not draggable. Input/hit-testing is its
own sub-phase — neither `IPanelHost` nor `IPanelRenderer` carries a hit-test or
bounds contract today, and building it up front is scope creep.
2. **`controls.ini` optional.** Assume `C:\Turbine\Asheron's Call\` may or may not
exist. Add an `ACDREAM_AC_DIR` `RuntimeOptions` field for when it's present;
when absent, fall back to the source-verified `[title]`/`[body]` token values.
The build never fails on a missing AC install. (Chrome is sprite-based, so
`controls.ini` is barely load-bearing for Spec 1 anyway.)
## 11. Build sequence
| Step | Deliverable | Proves |
|---|---|---|
| 0 | Dat prove-out: load candidate chrome IDs, render raw quads, confirm real IDs | Resolves the chrome-ID contradiction empirically |
| 1 | One decoded dat sprite drawn at fixed coords (shader `uUseTexture=2`, self-contained GL state) | A dat sprite composites correctly over the 3D scene |
| 2 | 8-piece border + center → an empty titled frame (UV-rect quads, stopgap insets) | The frame renders |
| 3 | Three scissor-fill bars bound to real `VitalsVM` (solid-color path) | End-to-end data binding, no font needed |
| 4 | `RetailPanelHost` wired into the frame loop, gated by `RuntimeOptions.RetailUi`; ImGui unaffected | Backend slots under the seam; no `ACDREAM_DEVTOOLS` regression |
| 5 *(deferred)* | `AcFont` dat-glyph loader → numeric overlays | Only if numbers are wanted in this slice |
## 12. Error handling & edge cases
- **Missing/undecodable sprite**`GetOrUpload` magenta fallback is visible by
design; Step 0 catches it. A null/zero DataID in markup logs a warning and
draws nothing (no throw).
- **AC install absent**`controls.ini` load is skipped, baked fallback tokens
used (no throw).
- **Vitals null percents** (pre-`PlayerDescription`) → empty bar.
- **Window resize** → fixed-coord placement re-clamps to stay on-screen via the
existing `OnFramebufferResize` panel-layout reset
([GameWindow.cs:10375](../../../src/AcDream.App/Rendering/GameWindow.cs)). No DPI
scaling (a known, out-of-scope gap — `_window.Size` is treated as framebuffer
size).
- **Both toggles on** → both UIs render; the retail Vitals and the ImGui Vitals
may both show (acceptable in dev).
## 13. Testing
- **`ControlsIni` parser** (pure, no GL) — unit tests for `#AARRGGBB` parsing,
`font://` URI parsing, the cascade order. Since the parsers live in
`src/AcDream.App/UI/Retail/` (per §4), their tests go in
`tests/AcDream.App.Tests/` (App-layer, Rule 6). If/when the backend is
extracted to a standalone `AcDream.UI.Retail` project, the tests move with it
to `tests/AcDream.UI.Retail.Tests/` (registered in `AcDream.slnx`).
- **`MarkupDocument` parser** — unit tests for the XML → element-tree mapping and
`{Binding}` resolution against a fake binding object.
- **`NineSlice` geometry** — unit test that 8 pieces + center tile to the right
rects for a given frame size + insets.
- **Visual acceptance** (user) — the Vitals frame renders retail-shaped with live
bars in `ACDREAM_RETAIL_UI=1`; ImGui panels unaffected in `ACDREAM_DEVTOOLS=1`.
- `dotnet build` + `dotnet test` green.
## 14. Bookkeeping
- **Phase D.2b**, Milestone **M5** (parallelizable; NOT on the M1.5 critical
path). The CLAUDE.md "Current state" line stays on M1.5 — this is a parallel
track, not a milestone flip.
- **Divergence register:** in the commit that ships the first real dat-sprite
chrome, **delete row TS-30** ([retail-divergence-register.md:166](../../../docs/architecture/retail-divergence-register.md))
and **add one** new IA-row (Intentional Architecture — keystone.dll has no
PDB/decomp, a byte-port is impossible by definition) for the markup/serialization
layer. Assign the next sequential IA number at commit time. Retail oracle:
"LayoutDesc 0x21xxxxxx; controls.ini panel-property vocabulary; keystone.dll
layout evaluation (no PDB)". Do **not** duplicate IA-12 (which already covers the
UI toolkit's *behavioral* approximation). A second row for the stopgap slice
insets is added if/when they ship.
- **Spec file:** this document, `docs/superpowers/specs/2026-06-14-d2b-retail-panel-frame-design.md`.
## 15. Open gaps & deferred sub-projects
- **Input/hit-testing contract** — neither `IPanelHost` nor `IPanel` reports
bounds; required before drag/close-click. Next sub-phase.
- **`AcFont`** — dat A8 glyph loader (Font `0x40000xxx``ForegroundSurfaceDataId`
→ RenderSurface, upload as **R8** so `ui_text.frag`'s `.r`-coverage branch works
unchanged). Sub-phase for numeric overlays.
- **Anchor solver** — port `StateDesc::UpdateSizeAndPosition`. With the
`LayoutDesc` importer.
- **`LayoutDesc` binary importer** (sub-project 3) — bulk-transpile retail's
layouts → our markup, supplying real slice insets + coords. Resolver symbols:
`LayoutDesc::InqFullDesc` @`0x0069A520`, `ElementDesc::Incorporate` @`0x0069B5A0`
(algorithm captured in [2026-05-08 pseudocode](../../../docs/research/2026-05-08-retail-ui-layout-resolution-pseudocode.md)).
- **Standalone `AcDream.UI.Retail` project** — after a rendering-primitives home
is extracted.
- **`PFID_CUSTOM_RAW_JPEG`** decode + login/char-select/chargen (sub-project 4).
## 16. Acceptance criteria
- [ ] Step 0 prove-out done; real chrome sprite IDs confirmed + recorded.
- [ ] In `ACDREAM_RETAIL_UI=1`: a retail-shaped Vitals window renders with an
8-piece dat-sprite border + title bar + drawn close button, and three
scissor-fill bars that track HP/Stam/Mana live as the character takes
damage / regens.
- [ ] In `ACDREAM_DEVTOOLS=1`: ImGui Vitals/Chat/Debug/Settings unchanged (no
regression).
- [ ] `controls.ini` loads when present, falls back cleanly when absent.
- [ ] `IPanelRegistry` on `IPluginHost`; a `MarkupPanel` exists and is unit-tested
against a fake binding.
- [ ] TS-30 deleted + one new IA-row added, same commit as the chrome.
- [ ] `dotnet build` green, `dotnet test` green.
- [ ] Visual verification by the user.