| Risk | Impact | Notes |
| The script bridge module has no ABI asserts | a script crossing whose type changes size between ABIs (long double, time_t on the MSVC runtime) would miscall at run time; none crosses today | only server and client_server have test/<module>_abi*.txt; hand-writing a script ABI file in that format would add them |
| Web build not run | its page may not reach a game: 65 symbols link undefined (test/web_undefined.txt, CG_Init among them) | build_check.py web builds it; nothing runs it in a browser |
| x64 clients not fully checked against the 32-bit ones | on Windows the view was closer and bluer (before the 2026-11-03 fixes; not rechecked) | Linux x64 draws effects and lights mp_toujane as 32-bit (fx_test.py, eye-level screenshots, llvmpipe); Windows only under a 64-bit Wine, which keeps data below 4 GB. Translucent effects sort by material address, so overlaps can blend in another order. The save-game Archive code still uses x86 offsets (unused in MP). The Windows client builds without libcod: src/libcod/gsc_memory.cpp keeps pointers in long, 4 bytes on Win64 (to fix in OpenCoD2-Plus Server) (ADR 0010) |
| Possible overruns in OpenCoD2-Plus Server's copies | unconfirmed; fixes go there first | src/qcommon/msg_mp.cpp reads a playerstate's field count unchecked, src/stringed/stringed_hooks.cpp may copy past its buffer as the engine's did; src/server, src/game and src/libcod were not audited |
| A cut HTTP download on Windows | checked on Linux only | CL_WWWDownload compares the bytes received with the server's size; not run against WinINet |
| A Windows dedicated server on IPX loses a byte of its address | replies to the wrong node | altPort overlays ipx[9] to keep netadr_t at 20 bytes; win_net.cpp writes it for every protocol |
| Client and server movement differ by 1 ULP on some frames | none measurable | OpenCoD2-Plus Server's bgame moved literals to float (0.001f, 0.4f, 0.65f), src/PC/bgame keeps doubles; far below prediction_test.py's 0.1 miss |
| r_gfxApi d3d12 is full screen only | a windowed d3d12 cannot be had | windowed D3D9On12 presents nothing on an RTX 5080 (discard, copy and flip; the back buffer holds the frame): d3d12 always asks for exclusive full screen at the desktop's refresh, which shows it. The console logs a Present result other than S_OK. Not tried on other GPUs; if none of the first 120 presents returns S_OK the client falls back to d3d9, but a black frame that returns S_OK is not caught |
| prediction_test.py --maxfps 60 gives about 500 misses of 1 unit in some runs | unknown | also at 20 Hz on the vendored server, and 0 in other runs of the same setup; 2026-11-02 against main's server: 0 in two runs, 1.3's client 1 and 0; Windows sessions show few, so it may be the Linux client under llvmpipe |
| Water has no automated check | its waves could stop again unseen | checked by screenshots in sps_island, a custom map. mp_harbor shows its water from the spectator spot turned 180° left (com_maxfps 20, cl_yawspeed 90, +left 40 frames), but two frames of it differ little more with the waves than without: its snow (not behind fx_enable) moves too, r_drawWater is a cheat dvar the command line does not set, and the turn lands a little differently each run |
| Real Direct3D 9 path tested mostly under Wine | a driver may differ | users test natively on Windows |
| Item | Why accepted |
| The GL Linux client (-DCOD2_LINUX_DXVK=OFF) keeps stand-in shaders | the default cod2_linux draws the game's shaders on DXVK; the GL one is for machines without Vulkan |
| fxRestart takes hunk memory each time until the next map load | developer command; many restarts on one map end in a drop error |
| Server browser's find-player list indexes 1-based (ui_main_mp.c) | nothing starts that search, in the original either |
| Windows dedicated server prints only to its window, shows a message box on fatal errors | OpenCoD2-Plus Server's src/win32; scripts read the console log |
| Gap | Where | Notes |
| Snow's particle_cloud_outdoor shader does not compile under Wine | Wine's d3dx9 cannot flatten an if | Microsoft's d3dx9 compiles it |
| Reverb tuned by numbers only | src/snd_sdl/snd_reverb.c | public EAX presets at a measured gain; not compared by ear with Miles |
| MP3 lengths are estimates | src/snd_sdl/snd_stream.c | from the first frame's bitrate: stock 073.mp3, which is VBR, is off by up to 51%. Miles estimates from a bitrate too |
| Particle clouds look faint at low resolutions | flakes are 1.5 units, the GPU samples the 1×1 mip of their 4×4 texture | matches the data; not compared with the original at runtime |