This is a more reliable solution as pretty much all code patch in CMDLCache checks the ptr for NULL, but none of them do for 0xDEADFEEDDEADFEED (probably a debug only thing that somehow made it into this build). Patched out so the engine could deal with it properly. Anything interesting already gets logged from the SDK so we won't miss out on anything patching this out.
Remove dev parameters from retail configs, disable net_usesocketsforloopback by default now. This was a leftover from before we got dedicated servers working, and was thus required to get other people to connect to the listen server. This is no longer needed, it was causing a problem with authentication on the local server as the input address from the client would differ from what the master server would bake into the JWT token (localhost translates into an actual ip address on the master server side).
- Add configuration support for each depot
- Add configurable vendor for each depot
- Add size field for each depot (used for installer's progress callback)
- Compute depot package checksum on the spot
- Take manifest version input as string instead (used as tags)
Increase to 2MiB, so all 120 players could be rendered on the client without overflowing the bone memory (which causes an error in 'CModelRenderSystem::DrawModels').
The bug was only reproduced on the compiled implementation in the game executable. The CUtlBuffer::ParseToken implementation in the SDK did not bug on the same input string. More research is required, and a possible good fix would be to just hook and replace the game's implementation with that of the SDK.
This ConVar's name was set to its helpstring. This patch solves it by assigning the name field to an existing string that makes sense, and moving the helpstring into the actual helpstring field. An additional instruction has been inserted to load the NULL string address into RAX, which required light instruction shifting. After performing this patch, the ConVar now could be ser in the console.
Function 'CL_Move' has been fully rebuild in the SDK. Originally, the game checked if the delta time exceeded an amount defined by an immediate value, and dropped usercmd's if that was the case. This logic has been replaced with a more dynamic solution, and the console variable regulating this ('fps_input_max') is set to 200.0 by default (the same as the fix applied in the Season 9.1 Genesis update). This function also has been slightly optimized by removing duplicate operations that were performed in the original function. A second fix has been applied to 'CInput::JoyStickApplyMovement' that was also found changed in the Season 9.1 Genesis executable. In that function, an extraneous clamp was performed on the frame time causing viewstick problems when usercmd's get dropped in CL_Move.
Save 2 bytes from segment padding by performing zero test at original address followed by the long jump to patch, there were 5 bytes, zero test only requires 2, thus only 3 bytes were nopped with these changes.
This commit adds support for loading cubemap lumps from the disk, several patches and hooks had to be implemented to make this work. Since the game was trying to read the pakfile lump from the BSP file (even when absent), it would read EOF (heavily reduces down map loading performance).
* Run autoexec(_dev).cfg before anything else.
* Always initialize RCON (only gets ran if an actual password is set).
* Removed 'Cbuf_Execute()' call from 'CHostState::LoadConfig', let it run in the frame loop instead.
* Removed hardcoded set call for 'net_usesocketsforloopback', will be set in autoexec from now on.
This patch prevents 'fps_max' from being reset in 'CEngine::Frame' if 'sv_cheats' isn't set; this cvar is now bound to the server's tickrate in sdk code to save CPU time.
Also removed these from the CFG files.
* 'sq_showvmoutput' has been renamed to 'script_show_output'.
* 'sq_showvmwarning' has been renamed to 'script_show_warning'.
Added 2 new patches:
* Patch0: Unnecessary loop called from 'CModelLoader::UnloadModel'. This calls a 'nullsub' CMaterialGlue vtable method.
* Patch1: CClientState constructor; Removed due to high mem allocation during init (100MiB+). 'CClientState' is useless on dedicated, and all its feature are thereof disabled.
Path has to be changed due to project include changes. The r5dev directory is no longer a system include, however, the compiler expects includes from system when compiling resources.
