Adversarially-verified review findings 1 and 10:
(1) The bake pipeline no longer accumulates every decoded ObjectMeshData
in one ConcurrentBag before writing (multi-GB OOM risk on the full
bake), and no longer writes blobs in thread-completion order (which
violated the plan's "bakes must be byte-reproducible run-to-run").
New shape in BakeRunner: build the FULL id list, sort by PakKey, chunk
into 512-id batches; Parallel.ForEach WITHIN each batch; sort each
batch's results by key and AddBlob sequentially; release the batch.
Batches are contiguous key ranges, so the blob region lands in global
key order regardless of thread scheduling, and peak memory is one
batch's output. Side-staged particle-preload meshes drain per batch
into a key-deduped map (first instance wins — per-id extraction output
is deterministic, so instance choice cannot affect bytes) and are
written after all batches, sorted by key, skipping keys already
written.
Program.cs is now a thin arg-parsing shell over the public BakeRunner
so the new dat-gated byte-reproducibility test can drive the REAL
pipeline: tests/AcDream.Bake.Tests (new project, rule 6; registered in
slnx; no Silk.NET in its resolved dependency graph — verified) bakes
the same 9-id mixed fixture twice with DIFFERENT thread counts (8 vs
3 — thread scheduling was the nondeterminism source) and asserts the
two pak files are byte-identical. Ran for real against the dats on
this machine: green.
(10) The isSetup argument for EnvCell extraction now matches at both
call sites (BakeRunner and PakEquivalenceTests both pass false) and is
documented at each: the runtime's own request sites
(WbMeshAdapter.IncrementRefCount / EnsureLoaded) pass isSetup: false
for every MeshRef id including cell-geometry ids. The parameter is
currently dead in MeshExtractor.PrepareMeshData (dispatch is on the
resolved dat type), but two disagreeing call sites were a latent trap.
Header FormatVersion is no longer set by the bake (PakWriter stamps it
per review finding 2, previous commit).
New offline console tool driving AcDream.Content.MeshExtractor (the SAME
extraction code the live client runs, per MP1a) to produce a versioned
pak file. Arguments: --dat-dir (required), --out (default acdream.pak
next to the dats), --ids/--landblocks (dev filters), --threads (default
Environment.ProcessorCount).
Enumeration mirrors existing idioms rather than inventing new ones:
GfxObj/Setup ids via DatCollection.GetAllIdsOfType<T>() (src/AcDream.Cli/
Program.cs); EnvCell ids by walking dats.Cell.Tree for LandBlockInfo
entries (low 16 bits == 0xFFFE) and then the firstCellId+NumCells range
per landblock (the same idiom as GameWindow.BuildPhysicsDatBundle /
BuildInteriorEntitiesForStreaming). GetAllIdsOfType<T>() does not cover
cell-dat range-based types, hence the manual walk.
BakeDatCollectionAdapter is a from-scratch copy of AcDream.App.Rendering.
Wb.DatCollectionAdapter (which is `internal` to AcDream.App and, more
importantly, referencing AcDream.App would drag Silk.NET/GL into the
bake tool — a hard violation of "no Silk.NET anywhere"). It's plain dat-
access glue, not an AC-specific algorithm, so a small duplicate is the
right call over inventing a shared-but-App-rooted package.
Particle-preload GfxObjs MeshExtractor side-stages mid-extraction
(sideStagedSink, thread-safe ConcurrentQueue per MP1a's documented
contract) are deduped against the primary GfxObj set and against each
other before being written as their own GfxObjMesh entries.
ConsoleErrorLogger is a minimal hand-rolled ILogger (stderr, Warning+)
rather than NullLogger — MeshExtractor's LogError/LogWarning calls on
malformed dat entries must stay visible per the project's "logger
injection for silent catches" lesson.
Progress line every 5s (baked/total, failures, elapsed, ETA); a
malformed dat entry is caught per-id and counted as a failure, never
fatal — matches the runtime's own per-id try/catch behavior. Failures
report is printed at the end, capped at 200 lines.
Smoke-tested against the real dats on this machine: --ids
0x01000001,0x02000001 baked 1 GfxObj (40 vertices) + 1 Setup (34 parts)
in 1.4s with 0 failures; a follow-up run adding EnvCell 0xA9B40100
(Holtburg's first interior cell) baked all three asset types
successfully. PakReader opened both scratch paks and read back
deep-correct ObjectMeshData; scratch files deleted after verification.
*.pak added to .gitignore in this commit per the plan.
New xunit test project for the Content layer (rule 6: tests live in the
project matching the layer under test). ProjectReference to
AcDream.Content; TreatWarningsAsErrors + LangVersion latest per MP1b's
csproj requirements. Registered in AcDream.slnx. Placeholder PakKeyTests
smoke test becomes the real Task 2 suite next.
Lifts 13 startup-time environment variables out of GameWindow.cs into a
single typed AcDream.App.RuntimeOptions record read once in Program.cs.
Behavior-preservation only — no live behavior change, no visual change.
Verified end-to-end against ACE on 127.0.0.1:9000: full M1 demo loop
(walk Holtburg, click door, click NPC, portal entry) plus DEVTOOLS
ImGui panels load cleanly.
Why: GameWindow.cs is 10,304 LOC and scattered Environment.GetEnvironmentVariable
calls were one of the structural smells called out in the new
"Code Structure Rules" doc. Typed options is the safest cut to make
first because the substitution is mechanical and parsing semantics
get pinned by unit tests.
What lands:
- CLAUDE.md: removed stale R1→R8 execution-phases line, replaced with
pointers to the milestones doc + strategic roadmap (the actual
source of truth). Tightened the "check ALL FOUR references"
section to describe WB as the production rendering base, not
just a reference. New "Code Structure Rules" section (6 rules)
captures the discipline we're committing to.
- docs/architecture/acdream-architecture.md: removed dangling link
to the deleted memory/project_ui_architecture.md.
- docs/architecture/code-structure.md (NEW, 376 LOC): rationale for
the 6 rules + 6-step extraction sequence
(RuntimeOptions → LiveSessionController → LiveEntityRuntime →
SelectionInteractionController → RenderFrameOrchestrator →
GameEntity aggregation). This PR is Step 1.
- src/AcDream.App/RuntimeOptions.cs (NEW, 100 LOC): typed record
with FromEnvironment(string) factory and Parse(datDir, env)
overload for testability. Covers ACDREAM_LIVE, _TEST_HOST/PORT/
USER/PASS, _DEVTOOLS, _DUMP_MOVE_TRUTH, _NO_AUDIO,
_ENABLE_SKY_PES, _HIDE_PART, _RETAIL_CLOSE_DEGRADES,
_DUMP_SCENERY_Z, _STREAM_RADIUS.
- src/AcDream.App/Program.cs: builds RuntimeOptions once, passes
to GameWindow.
- src/AcDream.App/Rendering/GameWindow.cs: ctor takes RuntimeOptions;
7 startup-cached env-var fields become expression-bodied
properties or direct _options.X reads; TryStartLiveSession,
audio init, legacy stream-radius branch all route through
_options.
- tests/AcDream.App.Tests/ (NEW project, 10 unit tests + csproj):
pins parser semantics — default-off bools, the literal "0"
gate for RETAIL_CLOSE_DEGRADES, the >=0 guard for
STREAM_RADIUS, null-vs-empty for user/pass, exact-"1" check
for diagnostic flags. Registered in AcDream.slnx.
Out of scope (per code-structure.md §4):
- Per-call-site ACDREAM_DUMP_* / _REMOTE_VEL_DIAG diagnostic reads
sprinkled through GameWindow (~40 sites). Rule 5 in CLAUDE.md
commits us to migrating these opportunistically as larger
extractions land, not in a bulk pass.
- AcDream.Core's project-reference to Chorizite.OpenGLSDLBackend.
Only the stateless .Lib namespace is used; tightening the project
reference is documented as future work in code-structure.md §2.
Build: green.
Tests: AcDream.App.Tests 10/10 ✓, Core.Net.Tests 294/294 ✓,
UI.Abstractions.Tests 419/419 ✓,
AcDream.Core.Tests 1073/1081 (8 pre-existing failures verified
against pre-refactor baseline by stash-and-rerun).
Visual verification: full M1 demo loop against ACE +Acdream login
including DEVTOOLS panel host load.
Next: Step 2 — extract LiveSessionController per code-structure.md §4.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Second piece of Phase D.2a: the ImGui-specific backend that implements
AcDream.UI.Abstractions' IPanelRenderer / IPanelHost. No GameWindow
hookup yet — compiles standalone for clean review before integration.
Packages:
* Hexa.NET.ImGui 2.2.9 (auto-generated from cimgui 1.92.2b)
* Hexa.NET.ImGui.Backends 1.0.18 (consolidated — OpenGL3 is here)
* Silk.NET.Input 2.23.0 + Silk.NET.OpenGL 2.23.0 (matches AcDream.App)
Files:
ImGuiBootstrapper.cs
One-shot static Initialize(glslVersion) / Shutdown() pair. Creates
the ImGui context, applies dark style, enables NavEnableKeyboard,
and boots ImGuiImplOpenGL3. Re-init is a no-op.
SilkInputBridge.cs
Event-driven Silk.NET -> ImGui IO bridge. Subscribes on construction;
Dispose() unsubscribes. Covers:
- KeyDown/Up -> ImGui.AddKeyEvent with modifier latching
(Ctrl/Shift/Alt/Super routed via both ModXxx flags AND named
key events so both IsKeyPressed checks and ImGui shortcut
matching work)
- KeyChar -> AddInputCharacter for text fields
- MouseMove -> AddMousePosEvent
- MouseDown/Up -> AddMouseButtonEvent (L=0, R=1, M=2)
- Scroll -> AddMouseWheelEvent (both axes)
Silk.NET.Input.Key -> ImGuiKey map covers WASD, arrows, modifiers,
letters, digits, function keys. Unmapped keys silently ignored.
BeginFrame(displaySize, dt) sets IO.DisplaySize + IO.DeltaTime.
ImGuiPanelRenderer.cs
IPanelRenderer impl — one-line wrappers on ImGui.Begin/End,
TextUnformatted, SameLine, Separator, ProgressBar. The ONLY place
Hexa.NET.ImGui types appear outside bootstrap/input plumbing. Panels
still never import ImGui.
ImGuiPanelHost.cs
IPanelHost impl. Dictionary keyed by IPanel.Id for idempotent
Register. RenderAll iterates visible panels and calls their Render.
Does NOT call ImGui.NewFrame / ImGui.Render — ownership belongs to
the caller (GameWindow) so GL state is explicit. Diagnostic `Count`
property.
No behavior change yet; next commit wires this into GameWindow behind
ACDREAM_DEVTOOLS=1 and ships the first visible VitalsPanel.
Covers the pure-logic surface of commit 1:
VitalsVMTests
* HealthPercent reads from CombatState.GetHealthPercent
* safe-default 1.0 when GUID unknown / not yet set / never updated
* SetLocalPlayerGuid reroutes lookup to the new guid (no stale cache)
* StaminaPercent / ManaPercent are null for D.2a scope
* ctor throws ArgumentNullException on null CombatState
PanelContextTests
* record-struct fields round-trip
* value-based equality
NullCommandBusTests
* Publish accepts any record type without throwing
* Instance is a true singleton
csproj template mirrors AcDream.Core.Tests (xUnit 2.9.3, Test.Sdk 17.14.1,
runner 3.1.4, coverlet 6.0.4, implicit Using Xunit). References only
AcDream.UI.Abstractions — no runtime / GL dependency, tests run fast.
Adds the backend-agnostic UI contract layer called for by the 2026-04-24
staged UI strategy (docs/plans/2026-04-24-ui-framework.md). This is the
stable layer both the Phase D.2a Hexa.NET.ImGui backend and the later
D.2b custom retail-look backend implement.
New module `src/AcDream.UI.Abstractions/`:
* IPanel — a drawable panel (id/title/visible/Render)
* IPanelHost — owns the panel list, drives per-frame dispatch
* IPanelRenderer — drawing primitives (Begin/End/Text/SameLine/
Separator/ProgressBar). Kept small + retail-friendly
on purpose — if a widget can't be expressed with
dat-sourced sprites+fonts later, don't add it here.
* ICommandBus — user-intent publisher; NullCommandBus is D.2a default
* PanelContext — per-frame record struct (DeltaSeconds + Commands)
* Panels/Vitals/
VitalsVM — reads CombatState.GetHealthPercent for the local
player. Stamina/Mana return null in D.2a; they await
a LocalPlayerState cache of PlayerDescription (0x0013)
which is filed as a follow-up issue.
VitalsPanel — first real panel. HP bar always drawn; Stam/Mana
appear automatically when the VM returns non-null.
Invariant documented in IPanel's XML doc: no `using Hexa.NET.ImGui` in
panel files, ever. If a widget needs something IPanelRenderer can't
express, the interface grows; panels never reach through.
References AcDream.Core for CombatState. Zero runtime/UI dependencies
— this project compiles headless, perfect for unit testing the
ViewModels.
No visible change yet. Next commits: (2) tests, (3) ImGui backend,
(4) GameWindow hookup + visible panel behind ACDREAM_DEVTOOLS=1.
Ran a live memory probe against retail acclient.exe (new tool:
tools/RetailTimeProbe/) to read the TimeOfDay struct at
DAT_008ee9c8 and compare against our computed values. The decompile
agent's identification of TimeOfDay+0x10 as "SecondsPerDay (int
copy)" turned out to be WRONG — the live value is **360**, which is
GameTime.DaysPerYear.
The retail FUN_00501990 LCG seed is:
seed = Year × (*+0x10) + DayOfYear
= Year × DaysPerYear + DayOfYear
= flat "total days since epoch" day-index
Our previous Phase 3c port passed 7620 (DayLength in ticks) as the
multiplier, producing seed=883,967 against retail's seed=41,807 —
completely different LCG outputs, completely different DayGroup
picks. That's why the user's retail kept showing stormy/rainy while
acdream showed sunny/clear (or vice versa) even after Phases 3c.1
and 3f aligned Year and DayOfYear.
Also confirmed by the probe:
- EpochBase / ZeroTimeOfYear = 3600 ✓ Phase 3f already correct
- BaseYear / ZeroYear = 10 ✓ DerethDateTime.ZeroYear
- Year=116, DayOfYear=47 ✓ our AbsoluteYear / DayOfYear
- SecondsPerDay float (+0x0C) = 7620 ✓ DayTicks
- SecondsPerYear = 2,743,200 ✓ YearTicks
One "finding that's not a fix": retail's +0x48 DayFraction is a
sub-period fraction (fraction through current day/night window)
NOT a full-day fraction. CurDayEnd - CurDayStart = 2857.5 = 0.375
of a day = 6 Dereth hours = night duration. Not relevant for our
keyframe bracket interpolation, which correctly uses a full-day
0..1 scale matching the SkyTime.Begin values. Documented in the
probe research doc so future work doesn't trip on it.
Changes:
- tools/RetailTimeProbe/ — new P/Invoke tool. Forced x86 target to
match retail's bitness so hardcoded DAT_xxxxxxxx addresses are
pointer-width-correct. Handles ASLR relocation via
Process.MainModule.BaseAddress.
- src/AcDream.App/Rendering/GameWindow.cs: RefreshSkyForCurrentDay
passes 360 (DaysInAMonth × MonthsInAYear) not 7620.
- src/AcDream.Core/World/SkyDescLoader.cs: ActiveDayGroup(ticks)
and DefaultDayGroup same.
- docs/research/2026-04-23-retail-memory-probe.md — full probe
results + decompile-agent correction.
- AcDream.slnx — add tools/ folder.
Build + 733 tests green.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
First step of Phase 4 (networking). Adds a new AcDream.Core.Net project
for the AC UDP protocol implementation and a matching AcDream.Core.Net.Tests
project. Keeps networking isolated from rendering and the dat layer,
which also keeps the AGPL-reference-material hygiene cleaner.
AcDream.Core.Net/NOTICE.md documents the attribution policy: we read
ACE's AGPL network code (and holtburger's Rust ac-protocol crate) to
understand AC's wire format, but we reimplement everything in acdream's
own style. Wire-format facts aren't copyrightable; specific code is.
This commit adds one component: IsaacRandom — AC's variant of Bob
Jenkins' ISAAC PRNG, used to XOR a keystream into the CRC field of
every outbound packet for authentication. Clean-room reimplementation
based on reading:
- references/ACE/Source/ACE.Common/Cryptography/ISAAC.cs (AGPL oracle)
- Bob Jenkins' public ISAAC algorithm description
Implementation notes:
- 256 uint32 mm[] state, 256 uint32 rsl[] output buffer, a/b/c regs
- Initialize() runs 4 golden-ratio Mix() warmup rounds then two fold-in
passes over rsl[] and mm[] (fresh instance → both start as zeroes)
- AC variant: seed is exactly 4 bytes, interpreted as little-endian
uint32 assigned to a = b = c before the first Scramble()
- Scramble() produces 256 output words in one pass; Next() consumes
them backwards from offset 255 → 0, re-scrambling at offset -1
- Test seed 0x12345678 matches ACE's reference output byte-for-byte
across the first 16 values (golden vectors transcribed from a
throwaway oracle harness that compiled ACE's ISAAC.cs and printed
its output; the harness was deleted after extracting the values)
Tests (5, all passing):
- Next_Seed12345678_MatchesAceGoldenVectors: 16 golden uint32 values
- Next_TwoInstancesSameSeed_ProduceIdenticalSequence: 1000 outputs
- Next_DifferentSeeds_ProduceDifferentFirstOutput
- Next_512Calls_SpansTwoScrambleBatches: >400 distinct values in 512
outputs (catches all-zero / stuck-at-one bugs at scramble boundary)
- Ctor_ShortSeed_Throws
Both test projects still green: 77 core + 5 net = 82/82.
Phase 4.2 (packet framing + checksum) next.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Phase 1 MVP end-to-end. Program.cs initializes Serilog, builds an
AppPluginHost that hands plugins a SerilogAdapter (IPluginLogger),
discovers plugins from the App's output plugins/ dir, loads each via
PluginLoader, calls Enable on all of them before opening the GameWindow,
and calls Disable in a finally block on shutdown.
AcDream.Plugins.Smoke is a new first-party plugin that logs through
the host during Initialize / Enable / Disable. Its csproj references
the abstractions with Private=false + ExcludeAssets=runtime to avoid
shipping a second copy of AcDream.Plugin.Abstractions.dll (which would
break ALC type identity). An MSBuild Target on the App project copies
the plugin DLL into plugins/AcDream.Plugins.Smoke/ and writes the
plugin.json manifest next to it.
Smoke verified against real dats. Console output observed:
[INF] scanning plugins in ...\plugins
[INF] smoke plugin initialized
[INF] loaded plugin acdream.smoke (Smoke Plugin)
[INF] smoke plugin enabled
loaded landblock 0xA9B4FFFF
<window renders terrain>
[INF] smoke plugin disabled (on shutdown)
Phase 1 done.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
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>
