BT412/docs/BT412-ROADMAP.md

# BT 4.12 Roadmap — Steam + Internet Multiplayer

Working plan for turning the 4.11 pod port into a consumer Steam title, following the
recipe proven end-to-end by **RP412** (Red Planet 4.11 → 4.12; see
`C:\vwe\RP412\docs\RP412-ROADMAP.md` and its git history — the reference implementation
for every workstream here). Status: plan authored 2026-07-14. Items marked
**[investigate]** need archaeology before they become concrete tasks.

## Where 4.11 stands

- **Game**: `btl4.exe`, 32-bit Win32, DirectX 9, CMake + VS2019 BuildTools (v142),
  DXSDK June 2010, `/FORCE:MULTIPLE`. Full single-player combat loop
  (drive → animate → target → fire → damage → destroy → respawn) on all 8 maps.
- **Engine**: the same WinTesla MUNGA/MUNGA_L4 engine as Red Planet — the MUNGA core is
  file-identical to RP's; BT's L4 layer adds bgfload/image + BT render work
  (L4D3D/L4VIDEO) and dev-gauge docking in L4VB16.
- **Networking**: same arcade stack — `MUNGA/NETWORK|HOSTMGR|INTEREST`, WinSock2 TCP in
  `MUNGA_L4/L4NET.CPP`. Two-instance loopback MP verified including cross-pod combat,
  kills, respawn, gait, and weapon visuals. Console feeder: `tools/btconsole.py`
  (egg msgID 3 in 1040-byte chunks, RunMission msgID 5 ×2, must stay connected).
- **Input/output**: pod hardware is a serial **RIO** board (stick X/Y, throttle, pedals,
  fire/config buttons 0x40/0x45/0x46/0x47 + MFD aux buttons) — `L4RIO`/`L4SERIAL`. Dev
  box uses keyboard debug controls (WASD etc.). Displays: **7 monitors** (main 800×600
  3D view, radar 640×480, five mono MFDs as one spanned 1280×480 surface) + the
  **128×32 plasma** (identical hardware to RP's; `L4PLASMA`). `BT_DEV_GAUGES` docks the
  gauge strip into the main window (dev mode).
- **Session control**: missions are **egg** files (NotationFile); the operator console
  configures and launches over TCP 1501. No in-game menu; no mission-end flow (mission
  review msgID 0x18 is a stub, `scoreAward=0`); live kill/death tally exists on the
  Comm MFD `pilotList`.

## Defaults (decided at planning, change only deliberately)

- Gates mirror RP412 with a BT412 prefix: `BT412_STEAM`, `BT412KEYLIGHT`, env
  `BT412STEAM=1`. Exe stays `btl4.exe`; CMake project is `bt412`.
- AppID: Spacewar **480** via hand-created `steam_appid.txt` (gitignored, never packed)
  until a real AppID exists.
- Mech catalog: base 8 (`avatar bhk1 loki madcat owens sunder thor vulture`) + variants
  (`ava1 lok1 lok2 mad1 mad2 own1 snd1 thr1 vul1 blkhawk`); trim any that fail to load.
- Results v1: read the live Comm-MFD `pilotList` tally (kills + observed deaths) at
  stop; EndMission wire scores when remote-pod marshaling needs them.
- `L4VB16.cpp` end-state: `MFDSplitView` is the successor to the dev-gauge dock; keep
  `BT_DEV_GAUGES` gated until parity, then retire.
- **This repo never touches the BT411 remote.** `origin` = `VWE/BT412.git` only.

## Phase 1 — Toolchain: VS2022 (v143)

RP412's step 1: unblocks everything, no design risk.

- Configure `-G "Visual Studio 17 2022" -A Win32` (drop the 2019
  `CMAKE_GENERATOR_INSTANCE`). **Stay Win32** — DXSDK June 2010 x86 libs are the hard
  constraint. BT's CMake has no `/Zp1`, so RP's packing-mismatch defines don't apply.
- Apply RP412 BUILD.md §3 source fixes as v143 errors demand: `MUNGA/TIME.h` Time
  copy-ctor; `MUNGA/NETWORK.h` inline `operator==`; `L4DINPUT.cpp` DIEnum callback
  renames; `CAMMGR.cpp` `std::ios::in`; `legacy_stdio_definitions.lib` for dxerr.
- [investigate] RP412's `L4D3D.cpp` NULL-texture AddRef guard — BT's L4D3D is heavily
  rebuilt; port the guard if the crash pattern survives.
- **Verify**: v143 build boots DEV.EGG; two-instance loopback + btconsole.py — drive,
  fire, cross-pod kill, respawn.

## Phase 2 — Input layer: RIOBase / PadRIO (Workstream A.1)

Rule for every touched engine file: 3-way check (BT copy vs RP411 baseline vs RP412);
identical-to-baseline → take RP412's file verbatim, diverged → hand-apply the diff.

- Port `L4RIO.h/.cpp` (RIOBase split), new `L4PADRIO.*`, `L4PADBINDINGS.*`; the
  `L4CTRL.cpp` hook (`new PadRIO()` for token `PAD`). BT never customized these —
  expected clean.
- CMake: explicit source lines (no globs); link `xinput9_1_0.lib`.
- Confirm `L4CONTROLS=PAD` yields `primaryControlType=PrimaryRIO` so the stock
  `MechRIOMapper` engages unchanged (`btl4app.cpp MakeViewpointEntity`) — RP precedent:
  stock VTVRIOMapper ran against PadRIO unmodified.
- Author the default BT `bindings.txt` (vRIO grammar): stick/throttle/pedal axes, fire
  buttons, MFD aux banks.
- Fix desktop-unreachable controls: secondary-MFD view cycling (mapper cases
  0x13d/0x13e are dead DOS F3/F4 codes). Weapon regrouping comes alive via the RIO
  path for free.
- **Verify**: full mech drive+fight from pad+keyboard, no COM ports; pad hot-plug;
  `L4CONTROLS=RIO` pod regression still initializes.

## Phase 3 — Cockpit on the desktop (Workstream A.2)

- **3a PlasmaScreen** (low risk, land first): `L4PLASMASCREEN.*` verbatim + the
  `L4GREND.cpp` selector diff. `L4PLASMA=SCREEN`.
- **3b Single-window MFD split** (`L4MFDSPLIT=1`) — **the HIGH-risk hand-merge**:
  RP412 rebuilt `L4VB16.cpp` ~3846–4046 (splitViews, fixed 1920×1080 canvas,
  3D Present into the viewscreen pane via `hDestWindowOverride`); BT rebuilt the same
  region for dev-gauge docking (`BT_DEV_GAUGES*`, `BT_GAUGE_*`, `BT_ALL` — 15 sites).
  Write 3-way diff notes into `context/steamification.md` **before** editing.
  `L4MFDVIEW.*` copied as skeleton; replace RP's pod tables with BT's display set
  (viewscreen, radar, five mono MFDs sliced from the 1280×480 span, plasma glass).
- **[investigate]** BT cockpit button-bank geometry: no vRIO `CockpitLayout` exists for
  the BT pod. Sources: `content/GAUGE/l4gauge.cfg`, pod photos, ask Nick. Land
  panes-without-buttons first; clicks inject `PadRIO::SetScreenButton`, lamps light
  from `GetLampState` once geometry lands.
- **Verify**: one cockpit window, 3D presents into its pane (smoke BT's rebuilt L4D3D
  present path EARLY), live MFDs/radar/plasma; pod multi-surface path (`L4GAUGE`)
  unbroken.

> **Status (2026-07-14):** 3a landed. **3b DEFERRED** — blocked on (1) the missing BT
> cockpit geometry and (2) BT's own MFD dev-composite being an unfinished upstream
> beachhead (`docs/GAUGE_COMPOSITE.md`: only radar/secondary composites; the five mono
> MFDs don't render yet). Prereq order when resumed: finish the MFD dev-composite →
> obtain BT cockpit geometry → hand-merge `MFDSplitView`. Details in
> `context/steamification.md`. Phase 4 does not depend on 3b.

## Phase 4 — NetTransport seam + Steam transport (Workstream C.1)

- `L4NETTRANSPORT.*` verbatim (abstract seam + `WinsockNetTransport` default;
  SOCKADDR_IN-shaped addresses keep egg `[pilots] ip[:port]` working in both worlds).
- `L4NET.CPP`: apply RP412's refactor (~22–26 Winsock sites → `NetTransport_Get()`);
  **re-site the 5 `BT_NET_TRACE` blocks** onto the seam calls.
- `L4STEAMTRANSPORT.*` near-verbatim, gate `BT412_STEAM`; vendor
  `extern/steamworks_sdk_164/` (copy from RP412).
- CMake: seam always compiled; `option(BT412_STEAM)` gates the transport TU + define +
  SDK include + `steam_api.lib` + post-build dll copy.
- **Verify**: (1) loopback + btconsole.py full MP parity through the seam (TCP);
  (2) Steam smoke — transport installs, FakeIP allocated, graceful TCP degrade when
  Steam is absent; (3) real Steam-sockets session (may defer to Phase 6).

> **Status (2026-07-14): DONE.** Seam landed; `L4NET.CPP` from RP412 post-seam with
> BT's 3 `BT_NET_TRACE` blocks re-sited; `gConsoleLossEndsMission` engine global
> ported. `option(BT412_STEAM)` gates the Steam TU + SDK. Verified: default TCP build
> passes full loopback MP through the seam (console → egg → mesh → both tick, traces
> fire through `NetTransport_Get()`); `BT412_STEAM=ON` compiles + links against the
> vendored SDK + boots solo. Live Steam session deferred to Phase 6 (needs the lobby +
> 3 machines). Details in `context/steamification.md`.

## Phase 5 — Front end + LocalConsole marshal (Workstream B — bulk of new code)

- **5a `game/reconstructed/btl4fe.cpp`** (skeleton: RP412 `RP_L4/RPL4FE.cpp`): no
  `-egg`/`-net` → setup menu. Catalog: 8 maps
  (`cavern grass rav polar3 polar4 arena1 arena2 dbase`), mechs per defaults,
  scenario=freeforall, time/weather/temperature/length, pilot name.
  `BTMission::ToEggString` against the verified egg format (`content/MP.EGG`,
  `DEV.EGG`) incl. GDI-rendered 128×32 `[ordinals]`/name plasma bitmaps → the standard
  egg-load path.
- **5b `game/reconstructed/btl4console.cpp`** (skeleton: `RPL4CONSOLE.cpp`): in-process
  console on its own thread over NetTransport — egg chunks + ACK, RunMission ×2
  (mirror btconsole.py), stays connected (fixes the console-loss listener bug for
  solo), owns the mission clock, StopMission at `length=` expiry, launcher-role loop,
  `InstallNetworkRace` marshal for remote pods.
- **5c Score intake — genuinely new for BT** (no mission-end flow exists): a
  `gConsoleScoreSink` equivalent through `BTPlayer`; v1 results screen = kills/deaths
  per pilot from the Comm-MFD `pilotList` tally at stop.
- **5d `game/btl4main.cpp` WinMain wiring** (skeleton: RP412 `RPL4.CPP` diff):
  front-end-mode detection, single-binary `for(;;)` loop (results → menu → launch),
  host/member branch, self-documenting `environ.ini` parsing, minidump crash filter,
  `SC_KEYMENU` swallow, `BT412STEAM` env → `SteamNetTransport_Install()`.
- **Verify**: no-args boot → menu → LAUNCH → mission → stop at set length → results →
  fresh menu; `-egg DEV.EGG` stays unmarshaled (dev shortcut); btconsole.py external
  feed still works; marshaled 2-instance loopback collects both score lines.

## Phase 6 — Steam lobby (Workstream C.2)

- **`game/reconstructed/btl4lobby.cpp`** (skeleton: `RPL4LOBBY.cpp` — mechanics
  generic): lobby owner = console; member data = FakeIP + fake console/game ports +
  persona + loadout (**mech**/color/badge); nonced launch roster → `RegisterPeer` all →
  owner hosts via egg+marshal, members join as network pods; Push/PullRaceResults
  distribute the score sheet. FE gains HOST/JOIN screens.
- Author `docs/STEAM-3-MACHINE-TEST.md` for BT (adapt RP412's; keep its gotchas:
  disable Steam Input on 480, one Steam account per machine, abort key).
- **Verify**: 3-machine Steam test — lobby → mesh → marshaled timed mission →
  kills/deaths → stop → results everywhere → rematch. Known MP fidelity gaps (peer
  warp spheres, missile clusters, PNAME/PLACE billboards) do not block.

## Phase 7 — Polish, packaging, release

- KeyLight RGB mirror: `L4KEYLIGHT.*` verbatim (C++/WinRT TU — per-file C++17 props),
  gate `BT412KEYLIGHT`.
- `pack-dist.ps1` adapted: `btl4.exe`+pdb, `BTL4.RES`, `BTDPL.INI`,
  `VIDEO/ GAUGE/ AUDIO/`, eggs, `steam_api.dll`, OpenAL runtime, generated commented
  `environ.ini`, `start-windowed.bat`, README.txt, refuse-while-running.
- `tools/two-pod-test.ps1` adapted (feeder = btconsole.py). Version banner `4.12.x`;
  tag the release on gitea.
- **Verify**: dist zip runs on a clean machine (no DXSDK); solo loop; packaged
  2-machine Steam session.

## Copy matrix (from RP412)

| Treatment | Files |
|-----------|-------|
| **Verbatim** (after 3-way check) | `L4PADRIO.*`, `L4PADBINDINGS.*`, `L4PLASMASCREEN.*`, `L4NETTRANSPORT.*`, `L4STEAMTRANSPORT.*` (gate rename), `L4KEYLIGHT.*` (gate rename), `extern/steamworks_sdk_164/` |
| **Diff-apply / hand-merge** | `L4RIO.*`, `L4CTRL.cpp`, `L4GREND.cpp` (expected clean); `L4NET.CPP` (medium — BT_NET_TRACE); `L4VB16.cpp` (**high** — dev-gauge collision); `L4VIDEO.*`; v143 source fixes |
| **Adapt** (RP skeleton, BT tables/fields) | `L4MFDVIEW.*`, `btl4lobby`, `pack-dist.ps1`, `two-pod-test.ps1`, `environ.ini` template, STEAM-3-MACHINE-TEST doc |
| **New** | `btl4fe`, `btl4console`, score sink + results screen, BT `bindings.txt`, CMake Steam gating, `context/steamification.md` |

## Standing regressions (after every engine-file merge)

1. Solo: `run\run.cmd` boots DEV.EGG; drive → fire → kill → respawn works.
2. Loopback MP via btconsole.py keeps entity/combat replication.
3. Pod paths (`L4CONTROLS=RIO`, `L4GAUGE`, `L4PLASMA=com2`) still compile and
   initialize.