|
OpenBF1942-Plus
Refractor 1, rebuilt from scratch
|
| Item | Kind | Notes |
|---|---|---|
| RFA unk directory field | open question | constant per archive; meaning unknown |
| Patch archive override order | unverified | assumed _NNN archives override base files |
| rfa.py extract to disk on large archives | untested | full-install check was in-memory only |
| Pure-Python LZO speed | accepted debt | ~2 min for non-sound archives; fine for tools |
| CON level load sequence and object file order | unverified | assumed, see con.md |
| Loose files vs archives priority | unverified | con.py assumes loose files win |
| CON method case-insensitivity | unverified | subsystems verified, methods assumed |
| SM collision tree, footer, v10 byte, alpha field | open question | triangles and materials decoded; the tree after them is opaque, see sm.md |
| C++ reads Terrain.con by line scan | shortcut | replace with a C++ CON interpreter when objects are placed |
| C++ VFS mounts archives only, not loose files | shortcut | add when a mod needs it |
| Missing colour tiles drawn grey | unverified | original behaviour unknown; only over open sea so far |
| Audio under WSLg | external | the device stalls for seconds after a minute or so, also in a bare SDL program with no window (see sound.md); to check with a native Windows build |
| Effects | unverified | field readings fitted by eye (effects.md); mesh particle orientation and scale read by eye; part effects at rest positions; impacts face up; wrecks burn 30 s; 400 particles per emitter |
| Bots ignore vehicles their weapon barely hurts | shortcut | a soldier on foot does not target a vehicle that its weapon would need more than 10 hits to destroy (guessed); the original AI's rule is unread |
| Outside the world | not implemented | past the combat area or on material 7 the original damages soldiers and vehicles; default damage and time allowed unknown (measure in D), see terrain.md |
| texOffsetX/Y, terrain LOD; material map for driving and outside-world rules | not implemented (the material map is read for impact effects only) | see terrain.md |
| Water drawn on every map | unverified | Berlin has waterLevel but no Water.* settings |
| Little-endian hosts only | accepted | memcpy of file integers in C++ |
| Texture alternativePath lookup | unverified | tried as <alt>/<file name> before the given path; every level's init.con names its own Texture folder this way, and 76–146 mesh textures per mod resolve only through it |
| Tree sprite orientation | unverified | corner.y sign chosen visually |
| Viewer draws LOD 0 / first LodObject child only, distance cull 1.5 km | shortcut | per-mesh setLodDistance exists in geometry templates |
| Effects not drawn | not implemented | later steps |
| Spawner team in the viewer | shortcut | instance setTeam, else first setObjectTemplate; no game state |
| Animation rotations and held weapons | unverified | .baf quaternions read conjugated and biped roots turned from Max space, chosen from pose checks; held weapon meshes turned 180° on their main bone, fitted; first-person set1pFov read as vertical radians, fire / reload states chosen by the game, not the state machine's transitions; no switch, sway or recoil animation; see anim.md |
| Animation frame rate and state machine | shortcut | 25 fps guessed; first animation of a state, looping; no transitions; weapon states by prefix |
| animations/animationstates.con loading | unverified | the viewer runs it after the object folders |
| Object lightmap blend | unverified | 2 × lightmap replaces the sun term on lightmapped meshes |
| Sun direction | measured | sky.sunLightDirectionVec is used with x and z negated. That doubled El Alamein's tanks to within 15 % of the original screenshot and puts the sun opposite the shadows baked into the colour tiles; Berlin's tanks are unchanged (fog dominates). Checked on two levels |
| Fog, sky, light scale | shortcut | fog uses renderer.fogStart / fogEnd where the level sets them (15 of 26 checked; otherwise half the view distance to the view distance). The sky is fogged as geometry 250 m away (our rule), which matches both original references: Berlin's sky a uniform pale sepia, El Alamein's clear blue. Mesh light ×2; terrain drawn unlit (baked lighting). First-person arms fitted by eye to one original screenshot (x offset dropped, 0.75 rad field of view) |
| Look against the original | measured | side by side on Berlin with the original client (2026-11-10). The ground lacked the terrain detail texture; it is now applied, with anisotropic filtering. Terrain brightness is still open. In a second test the camera was matched to the same spot (Berlin, Soviet base, by the tank stoppers and the two tanks). There the original's road measures 1.85 times ours, while buildings (≈1.0) and haze (≈1.1) already match, and the field of view matched ours (vertical ≈1.0 rad). The El Alamein reference (same spot, the Allied base flag facing the factory) settles that terrain brightness already matches there: sand 225 against the original's 213, factory 112 against 123. So no ×2. The Berlin road gap (original about 2.4 times our tile there) is still open. Ruled out: a ×2 on the terrain (wrong in the desert), denser fog (matches the haze at ×0.6 but leaves the road at 60 against 105 and over-brightens buildings), a shifted colour map (no offset within ±40 m lifts the road above 36), and patched tiles (Berlin has one version, in Berlin.rfa). Untested lead: Terrain.ShadowAmbient differs by level (Berlin 80, El Alamein 50, Stalingrad 140). Our buildings are grey-blue and darker, the original's warm and sepia. The original's haze starts much closer. Skinned meshes were lit per face (faceted, flickering); now smooth normals are recomputed from the posed mesh |
| Skinned normals | shortcut | face normals from screen derivatives; posed normals need a bind pose |
| Soldier parts | shortcut | viewer hides Parachute and 1p* children; the engine switches parts by camera and state |
| C++ vs Python edge differences | accepted | C++ pose loops and interpolates (Python: whole frames, clamped); C++ splits CON tokens and lower-cases ASCII only (Python also splits on \xa0, \x85); init.con mod paths are a line scan without if |
| .baf block second byte | open question | always present, unused by the decoder |
| TM collision block size | open question | same layout as .sm blocks, but its tree length is unknown; geometry is found from the end of the file |
| No model viewer in the engine | deliberate | step 3 exports OBJ for Blender; rendering starts with the map viewer |
| Networking protocol | risk | undocumented; roadmap step 8 |
| Vehicle handling units | unverified | spring/damping per wheel mass; drive and drag scales are placeholders; see vehicles.md |
| Grip, rollover, idle hold | shortcut | friction 1, grip at 80 % of centre-of-mass height, centre of mass at wheel-centre height, no c_PGF* or material friction |
| Tank turn rate | unverified | engine yaw setMaxSpeed × 8 °/s |
| Physics scope | not implemented | no gears, jets, gear retraction, submarine diving, track animation; trees, animated meshes and stationary spawner objects have no collision; M3A1 and Sexton hulls touch the ground; C-47 does not lift off; big ships turn very slowly |
| Weapons | shortcut | explosion damage falls off linearly to the radius, measured to vehicle centres; bouncing (grenades, mines, packs) with guessed restitution, resting shots do not ride moving vehicles; mines trigger on vehicle centres; no switch time; no drag, no effects; bot rules for pistols, mines and packs guessed (pistol only once the primary is dry; mines near held flags, packs by enemy-crewed vehicles); vehicle guns: only a seat's first weapon fires and counts rounds, no overheat; reloading drops the rounds left and starts by itself on empty; supply depot rates (addAmmoType, setHealth) read as per second, depots at their root object's position, no vehicle repair; armour does not turn with turrets; body part by hit height |
| Bots and AI | shortcut | sight, reaction, aim error, speed, respawn times guessed; routes on the 2 m infantry map, weak in cities (Berlin, Stalingrad); straight runs on the 13 levels without level-0 maps; no vehicles; none on ship spawn points |
| HUD | shortcut | our own layout and colours, not the original HUD textures (setAmmoBar images unread); space width guessed; .dif fonts unused; no minimap, radio |
| Audio | shortcut | .ssc semantics inferred; randomPlay variants all play (M1 Garand peaks at 46 voices); volumes above 1 capped; six-param ramps use four; no Doppler, pitch randomisation, triggers; engine load from speed; only shots and the driven engine sound; files load inside the audio lock (a first shot may stutter) |
| Collision blocks | unverified | static collision and armour use the last (detailed) block, which RefractorForge found stops movement in the original (not checked by us in game); what the first block is for is open, see sm.md |
| Conquest rules | unverified | capture, ticket bleed and win rules read from field names; conquest mode only |
| Parked vehicles | shortcut | lifted onto the terrain; LCVPs on davits drop into the sea; two Salerno Willys fall 50–70 m (unexplained); column hulls approximate concave parts |
| Soldier movement | shortcut | speeds, jump, swim depth guessed; entry by distance, not EntryPoint, into the first free seat; half-body seats drawn whole, seat offsets measured on the Willys pose only; bot crews: land and landing craft only (no pilots, no large ships), boarding range, stops, beaching and jam rules guessed; crouch and prone speeds and capsules guessed, prone capsule upright, bot postures and grenade range / rate guessed; parachute key, drag reading and fall damage guessed, bots take none; ladder climb speed and reach guessed; one looping climb / swim cycle (no start, end or exit animations) |
| Moving parts | shortcut | propeller rpm guessed; drawn wheel droop capped at 0.2 m; tracks, hands, gear retraction, engine rotation not drawn |
| Boat model | unverified | water drag rates, ship thrust = Σ torque / 4; float rest heights match retail spawns, see vehicles.md |
| Aircraft model | unverified | wing force constants, regulator as 1 g shared, thrust = Σ torque / 4; see vehicles.md |
| Jolt fetched at configure | accepted | network on first configure; default AVX2 build (ADR 0004) |