The WinTesla Windows Port

The Elsewhen RP411 archive is a working Windows port of the MUNGA engine (VS2005, WinTesla.sln). It means the hardest infrastructure is already solved — the bt411 build sits on top of it. Full detail: docs/PROGRESS_LOG.md §5b, §8.

What WinTesla already did (do NOT rebuild)

Renderer bypass: DONE — MUNGA_L4/L4D3D.cpp + DXUtils replace libDPL / the IG board with Direct3D9. No libdpl.lib in the tree. (The early from-scratch D3D9 viewer was retired and removed.) [T1]
Device-loss: the #35 "Owens crash" — ROOT-CAUSED + FIXED 2026-07-29 [T2, bench-reproduced]. 8 byte-identical field stacks (one player, Surface Pro 9 / Iris Xe 128 MB — the only GPU in the fleet that ever actually loses the device): ParticleEngine::Destroy +0x11, access=0 target=0x0, from the per-frame render path. Not a weapons bug — the Owens two-trigger missile+laser combo is merely what provokes the GPU timeout on that iGPU. The broken protocol (both Present sites, inline copies): on D3DERR_DEVICELOST they called Reset() IMMEDIATELY — which always fails while still lost (V() only logged) — then ParticleEngine::Initialize() ran against the lost device, its creates failed and NULLed the out-params (verified: the bench repro faulted at target=0x0, not a dangling address), and the NEXT lost frame's blind mVertBuffer->Release() read a vtable at 0x0. FIX (L4VIDEO.cpp + L4PARTICLES.cpp): one shared DPLRenderer::BTResetLostDevice() — release POOL_DEFAULT resources every lost frame (idempotent, null-safe Destroy), gate Reset() on TestCooperativeLevel() != D3DERR_DEVICELOST, check the Reset HRESULT, and on success re-create via CreateDeviceObjects() — NOT Initialize(), whose memset(mInstalledEffects) silently killed every particle effect for the rest of the mission on each reset that DID succeed (the quieter sibling bug). Draw paths guard the NULL buffer (ExecuteParticles keeps draining particles while dormant). Bench: BT_DEVICELOST_TEST repro'd the field crash byte-for-byte pre-fix, and post-fix the same shape logs SURVIVED + 3 forced loss/reset cycles recover ([render] device reset OK). ⚠ Discovery from the verify: VIDEO\particles.png has NEVER existed — absent from the tree, the RES, and all of git history — so mParticleTexture has been NULL on every machine since the engine was written and all billboard particles draw as untextured quads (SetTexture(0,NULL) = the shipped look). Deliberately NOT gated on in RenderParticles — that would disable all particles everywhere. Filed on the tracker; a real texture (or the original particle sheet) is a content task. Possibly related to the field "missile artifact looks wrong" observation.
Audio SOURCE BUDGET — the dropout complaint was fixed 2026-07-23; the leftover is NOT the same bug [T2, 2026-07-28]. The reported bug is closed: a999e5c (per-source delete instead of the spec-atomic bulk alDeleteSources) stopped the pool leaking, and there have been no field complaints since. Night-5 logs still show 2017 ACQUIRE FAILED lines in a match, all at live=256 — but that is transient exhaustion, not a leak: the census sits at 45-60 live, spikes during firefights, and drains straight back (226→42), i.e. the engine's priority steal loop (AudioSourceStop/SuspendMaintenance → ReleaseChannels → ReleaseSourceSet → alDeleteSources) is working. A dropped voice there competes with ~256 already sounding, so it is very likely sub-perceptual — treat the log noise as a symptom to watch, not as the old bug. Mechanism if it ever does matter: alcCreateContext(device, NULL) passes no attribute list, so OpenAL Soft applies its DEFAULT 256 mono sources (an OpenAL default, unrelated to the original audio hardware), while a busy match logs ~7200 explosions × 3 DPLIndependantEffect voices. ⚠ Raising the cap is deliberately OPT-IN (BT_AUDIO_SOURCES=<n>, no rebuild needed) — the 256 ceiling doubles as a governor. The steal loop only steals when the incoming source outranks a running one, so a higher cap means far more voices mixing at once, and with EFX reverb + lowpass chains live that is real CPU spent exactly during heavy combat. Measure frame time before making it default. Granted values measured: 1024→1024, 2048→2048, 64→240 (OpenAL Soft's own floor, which is also why the NULL default lands on 256) — always read back what was granted, a request is not a promise.
Audio: BACKEND now REAL + soundbank loaded; only game-triggering remains [T2, updated 2026-07-15]. OpenAL replaces the HMI "SOS" engine and the renderer→device→buffer→source→play chain IS all implemented (L4AUDRND/L4AUDRES/L4AUDLVL; PatchLevelOfDetail::PlayNote really calls alSourcePlay). The "no sound, ever" root cause was DEEPER than a gate — both audio backend DLLs shipped in engine/lib/ were no-op STUBS: (a) libsndfile-1.dll exports 15 funcs BY ORDINAL ONLY (no names), sf_open() always returns NULL; (b) OpenAL32.dll (72KB) imported only KERNEL32 — no dsound/wasapi/winmm — so it returned fake handles (ctx=0x00000001, alGenSources→0) and never touched the hardware. So the chain ran clean and silent. FIXES (commit 5b46655): OpenAL32.dll → real OpenAL Soft 1.25.2 Win32 (drop-in: the exe imports the 25 AL funcs by NAME; alSourcePlay now reaches AL_PLAYING). libsndfile DROPPED — replaced by LoadWavPCM() in L4AUDRES (a tiny in-tree RIFF/WAVE fmt+data reader, no external dep). Gates fixed earlier: the AWE_FRONT/AWE_REAR env-var gate in MakeAudioRenderer was relaxed to default-on (BT_NO_AUDIO=1 disables). Soundbank recovered: 241 samples cracked from AUDIO1/2.RES (SF2 v1.0) by tools/sf2extract.py → content/AUDIO/*.wav + the allPresets[2][128] table (audiopresets.cpp, replacing the zero-init btstubs stub); all 241 load (alErr=0). In-game triggering now WORKS (commit 7b1e469): enabling audio unlocked L4AudioRenderer::CreateEntityAudioObjects, which binds AudioStateWatchers to entity/subsystem state attrs BY NAME (the attribute-pointer databinding, same as gauges) and calls AddAudioWatcher on the resolved StateIndicator. The reconstruction had pointed the Mech's state attrs at the scalar read-pad attrPad (fine for a gauge that READs a value, fatal for a state watcher that calls a METHOD → AV in SChainOf::Add on 0xCDCDCDCD). FIXED: the Mech's AnimationState/ ReplicantAnimationState (driven from SetBodyAnimation) and CollisionState (driven from ProcessCollision via collisionTemporaryState) are now real StateIndicator members — verified: walking fires SetupPatch+PlayNote/alSourcePlay (footstep/gait patches 78/80/116). So mech movement/animation/collision audio is live. Subsystem STATE audio DONE (commits 5ba541e/0e05a22/dcdd7dd): the audio watcher TYPE per attribute was recovered via a MakeObjectImplementation ClassID trace (class 28 = AudioStateTrigger/StateIndicator). KEY: the binary's 0x54-byte subsystem alarm (FUN_0041b9ec, reconstructed as GaugeAlarm54) IS a StateIndicator — its "three sub-indicators" @0x18/0x2c/0x40 are the audio/video/gauge watcher SChains, level@0x14 = currentState. The reconstruction had modeled that tail as an opaque _tail (never constructed → AddAudioWatcher AV'd on 0xCDCDCDCD). FIX: GaugeAlarm54 now has three REAL constructed sockets + AddAudioWatcher; SetLevel fires them on change (empty sockets = no-op, so other alarms/weapons unaffected; levelB/LevelCountB untouched; sizeof stays 0x54). Then registered each state attr → its subsystem's own alarm (own AttributePointers[] chained to the parent): Generator.GeneratorState→stateAlarm, Condenser.CondenserState→condenserAlarm, Reservoir.ReservoirState→reservoirAlarm, AmmoBin.AmmoState→ammoAlarm (+ chained AmmoBin's bare index so SimulationState resolves too). GeneratorOn→generatorOn. audiostate skips 85 → 0 — every StateIndicator audio attribute across the mech + all subsystems binds to a real indicator, and the alarms are already SetLevel'd by the sims, so generator/condenser/reservoir/ammo state sounds fire on transition. Still inert (read 0, silent, non-crashing — no modeled backing member in the size-locked layouts, a polish follow-up): ReportLeak (Logical), Torso.MotionState (Enum), Torso.SpeedOfTorsoHorizontal (Scalar). NOW REAL (fidelity Phase 1, commit fc7f311): ConfigureActivePress published at the MechSubsystem BASE (binary attr id 2, descriptor @0x50de5c → member @+0x110 — the field previously misnamed vitalSubsystemIndex; idle −1, driven 0/−1 by MechWeapon::ConfigureMappablesMessageHandler), AmmoBin.FireCountdownStarted → cookOffArmed@0x18C, L4MechControlsMapper.TargetRangeExponent → targetRangeExponent@0x1a4. The bring-up guards [T3, self-clearing] stay: a NULL subsystem attr → inert pad (was a fatal Fail); an AudioStateWatcher on an unconstructed indicator (chain@+0x18 null/0xCDCDCDCD) → skip. Both pass automatically once a subsystem is built. Diagnostics: BT_AUDIO_LOG (chain + skips/redirects), BT_ATTRBIND_LOG (every attr bind), BT_AUDIO_TEST (proof-of-life buffer0).
Audio FIDELITY pass (2026-07-16, commits fc7f311 + 691e885; the full audit ledger is docs/AUDIO_FIDELITY.md, 37 findings). Settled facts [T1 unless noted]:
- Patch numbering = SF2 wPreset. Any tooling must key presets on wPreset, never on phdr record order. AUDIO1/2.RES are SF2 v1.0 (16-byte shdr, NO name/rate/pitch field); pitch metadata rides generators: 55 = SBK samplePitch (root*100+cents, on every zone), 58 = overridingRootKey, 51/52 = coarse/fineTune. EMU8000 v1 base rate = 44100 Hz → a zone's true data rate is baked into its WAV: rate = round(44100·2^(((6000−rootCents)+tune)/1200)) (tools/sf2extract.py); the engine's note-60 pitch model (BTNotePitchFactor) then reproduces original playback at every note. SBK initialAttenuation(48) is INVERTED (127 = full volume; 0.375 dB/step [T3 exact curve], baked into the PCM).
- Full-zone table: 603 zones / 241 presets in audiopresets.cpp; the authored MIDI note SELECTS zones by keyRange in SetupPatch/PlayNote (key-splits: Loaded/Jam family low-36 clunk vs high-84 blip; Warnings01 8-way klaxon split), layers attach together, MAX_PRESET_SAMPLES=25 (AllExplosion). Authored loop REGIONS apply via AL_SOFT_loop_points; authored releaseVolEnv (1.1–3.9 s on ~20 looping presets) fades on Stop (PRESET_serviceReleaseFades, serviced from AudioHead::Execute).
- ⚠ THE LOOP FLAG IS NOT THE LOOP DECISION (Gitea #51, 2026-07-25). SetupPatch (L4AUDLVL.cpp) used to map AL_LOOPING = (sample != ForceStatic), which armed 224 of the 603 authored samples as OpenAL sources that never end — any whose note-off never arrives plays forever (SAURON: "coolant flush sound glitched and perma"). Three facts say the sample flag only expresses PERMISSION, and the SOURCE decides [T1]: (1) the enum's own comments (L4AUDLVL.h:14-19) — LoopAtWill = "will play once or loop as desired", ForceStatic = "plays only once even if looped" (a veto), LoopAlways = always; (2) LoopAtWill is enum value 0, i.e. the DEFAULT an unauthored sample receives (WTPresets.cpp:37) — it cannot mean "loop forever"; (3) LoopAlways, the real always-loop, is used by zero shipped samples. "As desired" is the source's authored AudioRenderType (Transient=one-shot vs Sustained=held), streamed from AUDIO*.RES (AUDLVL.cpp:28) and already consulted by the renderer (L4AUDRND.cpp:567). New rule: LoopAlways→1, ForceStatic→0, LoopAtWill→ the source's render type. Measured (BT_LOOP_AUDIT=1, scratchpad/loopaudit.py): LoopAtWill×Transient 1946 events flip 1→0, LoopAtWill×Sustained 65 events stay 1, ForceStatic×Transient 292 unchanged — the engine loops (EngineAccel07_z0-z2/EngineMotor01/EnginePower01) are all Sustained and preserved; all 24 reclassified samples are genuine one-shots (laser charge/fire/explosion, missile loading, coolant pressure inc/dec). A/B proof (scratchpad/loopab.py): under the old rule CoolantPresInc03_z2 + CoolantPresDcr03_z2 are still playing with loop=1 after every release; under the new rule zero coolant sources survive, engine loop intact. BT_LOOP_LEGACY=1 restores the old rule for a field A/B — listen to the beam-sustain layers (LaserA/CSustain* are LoopAtWill on a Transient source, so they now end with the sample). A stuck looping source also holds a pool slot for the rest of the match, so this plausibly contributed to #32's dropouts.
- Distance/volume/doppler: OpenAL distance model is AL_NONE — the AUTHORED GetDistanceVolumeScale() curve multiplies in CalculateSourceVolumeScale (Dynamic3D + Static3D); per-source AL_GAIN = volume_scale² (GM CC7 squared law); alDopplerFactor(0) — the authored GetDopplerCents() (AUDIO.INI model, consumer decomp part_008.c:7466) adds into the Dynamic3D pitch chain.
- Audio clock: FUN_0044e19c is ApplicationManager::GetFrameRate — the original audio "frames" were SYSTEM CLOCK TICKS; MakeAudioRenderer calibrates at SystemClock::GetTicksPerSecond() (fallback 1000), never the old garbage divisionParameters+0x10 read.
- Published localVelocity smoothing is authentic [T1]: Mech::Execute (FUN_004ab430) runs the localVelocity components through AverageOf running-average filters (part_012.c:15169-15179); the port's ~0.25 s exponential filter in mech4.cpp reconstructs it.
- Phases 3+4 SHIPPED (2026-07-16, commits 0ca7d26..0e2401f). Phase 3 EFX: L4AUDEFX bridge — per-source lowpass (Dyn3D = highFreqCutoffScale×brightness UNGATED, Static/Direct = brightness; AWE 100–8000 Hz → 5 kHz-ref gainhf [T3 curve]), one EAXReverb slot at global_reverb_scale 0.3 (3D wet / cockpit Direct dry), Direct position-enum placement. Phase 4 attributes (all [T1] from binary attribute tables, walked as 16-byte {id,name,off+1,pad} rows): HeatSink table @0x50e438 (ids 3..12) confirms every heat binding + adds ReportLeak id 12 @0x138 = coolantActive (serves all 19 leak watchers via PoweredSubsystem⊂HeatSink); Mech table @0x50c0xx (ids 21..56): IncomingLock@0x3fc / DistanceToMissile@0x400 (real members, Missile::MoveAndCollide reports, PerformAndWatch latches; old maxSpeed@0x400 was a misread of the FLT_MAX far default), CollisionSpeed@0x4B4 (=|impact velocity| at contact-arm), ReduceButton@0x340, FootStep@0x394, RearFiring@0x410 (old stateFlags misread); Torso table (ids 3..15): full publish onto existing members (SpeedOfTorsoHorizontal=twistVelocity@0x1E8 abs'd, MotionState=statusFlags@0x20C, ==2 = limit-hit frame). F5 footStep = CONTACT LEVEL: (jointlocal.y <= *ANI hdr[2] threshold) per frame, states 0/1 retain (SelectSequence now captures hdr[2]; the 150 ms pulse is retired). F19: the authored footstep volume = LocalAcceleration |lin| [0,10]→ctl100 + LocalVelocity |lin| [0,0.6]→ctl101 (ALL motion watchers extract |linearMotion|, motionValue=3); localAcceleration.linear = d(published velocity)/dt (binary part_012.c:15186-15195); the mech2.cpp step broadcast is REMOVED. attrnull count = 0 — every authored audio attribute binds a real member.
- Post-Phase-4 closures (2026-07-16): UnstablePercentage is LIVE (the instability model @0x3F0 byte-decoded + reconstructed — accel-sway² + gun-the-engine + turn-in-place terms, feeds gyro swayBias + the audio alarm; commit c51d64a). The F14 per-zone static resonant low-pass is BAKED (RBJ biquad at the AWE 100–8000 Hz curve, float pipeline + peak normalization; 232 zones incl the LaserLoaded charge hum fc=57 Q=87; c581553). The footstep 10–20 s warm-up bug is fixed (a11a697): an idle source's watcher chain executes only at Start attempts and each hop is audio-frame-gated, so the authored fill-0 N=30 smoothers starved behind the 0.3 transient drop gate — fixes: AudioScaleOf sends EVERY poll (the bitwise change-gate froze on our deterministic gait floats), recursive gate-free Component::PrimeWatchers(30) on every transient Start request (AUDREND), and localAcceleration via the binary's exact 15-sample ring buffers. LESSON: trace caps saturate early and fake "per-stride" rates — always timestamp before inferring cadence.
- Still deferred: F21 attack velocity/time (all captured events were default 1), F22 HRTF (user has 2 speakers; enable via alsoft.ini hrtf=true for headphones).
- Energy-weapon post-beam buzz = AUTHENTIC (Gitea #5 verdict, 2026-07-20) [T2, measured]: the reported "buzz/electrical sustaining 1–2 s past the beam" is the AUTHORED charge loop (LaserACharge01-04, looped, release 1.309 s), started by the WeaponState==3 (Loading) watcher at beam-off, volume authored to track PercentDone (crescendo to ready), stopped on 3→2 with the authored 1.309 s release fade. Measured live (BT_AUD_TAIL=1 + BT_FIRE_PULSE single shots): beam 203–207 ms (authored DischargeTime 0.2 s); sustain loops cut ≤2 ms after leave-Firing (authored release 0); charge StopNote ≤2 ms after Loading→Loaded; fades 1.311–1.326 s vs authored 1.309 s (≤ 1 frame over); laser recharge 2.3–3.8 s, PPC 8.8–16.3 s under multi-weapon generator load (authentic electrical model). NOT the #12 source overflow — no energy preset exceeds 5 zones (max 3; >5 is only the death/UI families: AllExplosion 25, MechExplosion01/ Death01 16, AllButtons 14, AllCoolant/AllProjectile 9, Warnings01/AuxExplosion01/AllHits 8, AllWarning 7). Caveat: the reporter's 2026-07-18 build predated 2edde29 (Emitter::SetDirty DelayWatchersFlag poison retirement — PerformAndWatch skips ExecuteWatchers while set, SIMULATE.cpp:461; cleared only by the encore path, UPDATE.cpp:215), which made emitter audio watcher timing encore-erratic that session. Harness kept (env-gated, additive): [aud-tail] logging in L4AUDLVL.cpp (PLAY/STOP/fade) + emitter.cpp (state transitions), and BT_FIRE_PULSE="on,off" single-shot trigger pulsing in mech4.cpp.
Audio-dropout fix (field 2026-07-23: "cutting in and out toward the end of the match") [T2, code-verified vs the AL spec; awaiting live playtest confirm]: three structural defects in the port's OpenAL source-pool lifecycle (L4AUDRND.cpp / L4AUDIO.cpp / L4AUDHDW.h): (1) SourceSet.sources[] was uninitialized heap garbage until first acquisition — AL names are small ints, so a recycled-heap slot could alias a LIVE source owned by another sound; alIsSource() then skipped generation, two sounds shared one source, and whichever released first deleted the other's source mid-play (random sound deaths worsening with heap churn). Fix: SourceSet ctor zero-inits (0 is never a valid name). (2) ReleaseSourceSet used bulk alDeleteSources(count, sources), which is ATOMIC on invalid names — one empty/-1 slot (partial set, double release) and NOTHING deleted; after the first pool exhaustion every partial release leaked and the pool never recovered. Fix: per-slot guarded delete, slots parked at 0. (3) Failed acquisitions stranded partial sets forever — dropped transients are never released by anything (the dtor didn't release; maintenance only touches running sources). Fix: handback via ReleaseChannels() at both failure sites (StartRequest + dormant-resume) + a dtor backstop. Context: the port's RequestAudioChannels has NO channel budget (the original per-card budget is the commented-out block) — it alGenSources until OpenAL Soft's ~256 cap; on failure a transient is silently dropped (StartRequest early-return), a sustained goes dormant — which is exactly the "cutting in and out" presentation. Priority-steal (RequestAudioResources) only engages at cap. PERMANENT diagnostics: 30 s [audio] source census: live/acquireFails + an always-on [audio] ACQUIRE FAILED line. Baseline (2-node soak): live ≈ 21 peak under fire, 0 at mission end; exhaustion needs 7-player scale or the leak spiral, aliasing needs neither.
L4 HAL on Windows + the VS build system: DONE. [T1]
Red Planet game logic: COMPLETE & buildable — VTV/VTVMPPR/WEAPSYS/.... RP is at/near playable. [T1]
Multiplayer: the DOS NETNUB driver is already replaced by a ~3.5k-line WinSock2 TCP reimplementation (MUNGA_L4/L4NET.CPP) + the master/replicant distributed-sim core. See multiplayer. [T1]

What BT still needs (the only remaining work below the game layer is NONE)

Everything below the game layer — engine, renderer, audio, HAL, build — exists and is shared. The ONLY thing BT needs is its game-logic source, which is gone (see source-completeness) → it must be reconstructed, with the three anchors (BT headers + the WinTesla engine to compile against + RP's parallel code + the binary oracle). [T2]

Build implications

WinTesla.sln is VS2005 and hard-depends on the legacy DirectX SDK (June 2010) (d3dx9, dinput, dxerr — all removed from the modern Windows SDK). The DXSDK is installed at C:\Program Files (x86)\Microsoft DirectX SDK (June 2010)\. [T2]
The shared engine builds GREEN to munga_engine.lib via CMake/MSVC Win32 (see build-and-run). Recipe gotchas (include-path shadowing, the CAMMGR/time.h fixes, the ATL shim) are in docs/PROGRESS_LOG.md §8. [T2]

Roadmap position

Old plan ("build a libDPL shim + port the engine") is superseded — WinTesla did it. New plan: reuse the WinTesla engine (shared) + reconstruct the BT game library (BT410_L4 + game logic). The from-scratch libDPL shim is no longer needed. Phases 4-8 (BT game layer + pod bring-up) still apply. [T2]

Key Relationships

Solves: the infrastructure that source-completeness would otherwise require.
Feeds: build-and-run, rendering, multiplayer.

22 KiB Raw Blame History Unescape Escape