- Mark 'con_drawnotify' FCVAR_MATERIAL_SYSTEM_THREAD
- Add assert for char* in CTextOverlay::AddLog()
- Make buffer stack based in CTextOverlay::DrawFormat()
- Make color var struct const in CTextOverlay::DrawCrosshairMaterial()
ImGui drawing code now takes place in the main thread, a snapshot of the render data is created in CMaterialSystem::SwapBuffers(), and is being rendered in the render thread right before SpinPresent().
The reason why this was necessary, is because ConVar::GetString() isn't thread safe if its not marked FCVAR_MATERIAL_SYSTEM_THREAD or FCVAR_ACCESSIBLE_FROM_THREADS, and we used it for the console suggestions window, which iterates over every ConVar, accessible from threads or not.
This is overall also a better code architecture.
This change also forced changes in the ImGui server browser window, which contained a bunch of threading bugs too.
Engine changes:
- g_pCVar is no longer exported, as this wasn't necessary in the end.
- g_ThreadMainThreadID and g_ThreadServerFrameThreadID are now exported from the game executable, as these are required as soon as the DLL is loaded.
Fix incorrect function prototype and add additional logic implemented in the engine. The code was crashing as the first parameter is the command target, but appears unused, the second parameter was actually the parameter that contained a pointer to the CCommand object which we needed.
Fully implemented ConVar class so we could statically construct all SDK convars, this avoids a level of indirection, and allows for creating ConVar's everywhere in the project.
This patch also removed the settings tab of the ImGui server browser, as it has threading issues, while it technically never caused a crash yet, it has been removed as there was no point keeping it vs the work required to make it thread save (it only managed 2 convars which are perfectly manageable through cfg's or the in-game console).
Also temporarily disabled the creation of ConVar's in the mod system due to a memory leak, we would allocate and register a convar based on details parsed out of a mod file definition, but never unregister and free it.
Properly implement the ConCommandBase and ConCommand classes so we could statically construct all ConCommand objects in the global scope of each translation unit, this way we don't need to put them in a global file and deal with preprocessor directives to compile then in/out for certain projects.
Add missing types to fully mapped out PakGlobals_s struct, which is a 13MiB+ structure! This covers the vast majority of the pakfile system which is why we had to change a bunch of stuff for this patch. This patch also comes with:
- Reversed 'JobFifoLock_s' structure
- Reversed 'PakTracker_s' structure
- Reversed 'PakAssetTracker_s' structure
Many globals have been dropped as they were covered by the large PakGlobals_s singleton.
The pak decoder logic has been changed up as well, we now use a decode mode enumerant which will make it easier to add in more decoders for the pak files in the future.
Mutex should be locked before IsConnected(), as IsConnected() accesses CSocketCreator which checks a vector (not thread safe), lock the mutex in the correct place.
- Upgraded hashing algorithm to SHA-512, and store the raw hash instead of a string copy, which is way cheaper to compute and compare.
- Only ever close sockets once in CRConServer::SetPassword().
- Made the game server & game client RCON singletons static.
- Added calls to gracefully shutdown RCON server and RCON client on Engine/SDK shutdown.
- Added more prints so RCON user knows when its shutdown, or when their password change is in effect, etc.
- Fixed bug where we could tokenize an empty string when we dispatch a console command.
- Removed 2 std::string copy constructions
- Removed 32 sprintf calls per token auth request.
- Fixed a bug where we format the NucleusID as s64 instead of u64.
- Added additional hardening for when token/sessionId stitching fails, this will now always reject the connection.
- Improved the macro to make sure we always free the JWT claims if it has been allocated.
* Move member vars to correct place, to match it with the engine.
* Added out-of-band network message ID's for the R5 engine.
Also implemented 'ServerDataBlockSender::SendDataBlock()' and 'ClientDataBlockReceiver::AcknowledgeTransmission()'. NOTE that these are currently not tested, and also not in use! The code uses the version stored in the vftable which is what the engine itself provides. These have been implemented for reference only. If they need to be used, they need to be thoroughly tested first!
All routines are now fully implemented. This was delayed on purpose as some bit buffer functions have changed due to the use of 64bit integers for sizes, and also the coord types. The porting of this has to be done carefully; all
reimplemented functions have been changed to feature 64bit integers (where necessary) for syze types, and all values in coordsize.h have been tweaked to reflect the changes in the R5 engine 1:1, allowing us to properly implement the coord bit buffer functions as well.
CCommand: use 'V_isdigit()' since that is faster, we don't need it to be locale aware.
CMemoryStack: move 'highest' var to string format directly to avoid compile warnings (unused loval variables) on more recent visual studio compilers when compiling the DLL in cert mode.
Rebuild CClientState::ProcessCreateStringTable() and add notes for multiple 'potential' exploit vectors that are currently in place. Also add wrappers for compressing/decompressing net packets.
This patch partially rebuilds the data block sender/receiver. The receiver leaks memory if the sender sends a bogus LZ4 packet, it would allocate memory to copy the encoded data into, from which it would decode it to the scratch buffer, but it would never deallocate this temporary buffer is the LZ4 decoder failed. This has been fixed. The second reason to rebuild these was to look into potential compression optimization. The data block rebuild now also features the latest LZ4 codec.
Fixed bug in CNetChan::SendNetMsg() where the return value of INetMessage::WriteToBuffer() was not checked, nor did we return if the stream we added the message in was overflowed. If the function gets to the stage of writing the msg in the buffer, we need to return true only if we didn't overflow AND if the msg was successfully written. Also added a method for sending bitbuf data into stream of choice.
Mixed MAX_BACKUP_COMMANDS with MAX_NEW_COMMANDS, added proper (confirmed) defines. Also fixed MAX_BACKUP_COMMANDS_PROCESS, which was 60 but was supposed to be 64, so we were 4 backup commands short in CPlayer::ProcessUsercmds().
This ensures that we can never allocate more memory than the maximum value that type could hold, without triggering a compile warning. There was a mention about shorts being slow on PowerPC platforms for utllinkedlist, but the index local type could be overridden (defaults to index storage type). In case it gets used for PowerPC, use 32bit ints as index local types; as the local type gets used for arithmetic operations while storage type is what's getting cached off in memory. This patch had no effect on generated output code, it was mostly implemented to avoid future problems.
Determine proper stack buf size for ssize_t += operator. If its 64bits then the buf should be doubled to accommodate for the maximum number of digits in a 64bit decimal number. Also properly determine stack buf size of double += operator. For FormatV, instead of a fixed stack buffer, perform a dry run to determine required buflen, reserve num char bytes of memory and write into it again on the next call.
Only update statistics, this code was mainly added for testing, but testing revealed no additional performance or smoothness improvements, even during >6 hour sessions. The clock drift already gets corrected each frame from Host_RunFrame(). Only update statistics to update CPU and frame times on the client's debug panels if enabled.
