Network/MP additions&refactor, plus some additional bug fixes #2215
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…tor some code in doing so
…. Still need to implement a way to ensure the client receives the message before socket is closed.
… shutdown of server sockets
…, and then propertly dispose of the socket
…as the host from joining
…arly_send_buffer as it served no purpose.
…queue instead, by implementing client selectors
…time. Fixup some other code
…option for pre-execute&broadcast function for commands. Also add host_execution to allow for commands to be queued that the host does not execute themselves. Add function queue from update -> UI thread, so update thread can queue up ui-related things without breaking syncronisation. Changed the "busy" cursor so that only a special "normal_cancel_busy" cursor type can cancel it, so it isnt cancelled by random mouse clicks.
…nction instead. Experimental: allow host and client to fire off commands, even when some other clients are loading
…oad has been executed by the host, so that it is up to date. Added command_executed back to send_receive loop Minor optimizations
… refs to command_data, to reduce copying if there are many clients
Fix oos bug if joining while the game is running, by preventing normal commands from being added to send buffer for a client in handshake mode. Fix crash bug by adding a shared_mutex with the UI thread and cache thread
…f SPSC queue. Change logging to work on a diffrent thread with a concurrentqueue. Any thread can queue messages to be logged without blocking, and game_loop() spawns a thread at the start which will process any queued log messages.
…es of text instead of one
… from other headers. fixup tests
…eporting/handling if invalid command
…network performance gains
…around to avoid build errors
schombert
reviewed
Jan 14, 2026
| #define _WINSOCK_DEPRECATED_NO_WARNINGS 1 | ||
| #ifndef WINSOCK2_IMPORTED | ||
| #define WINSOCK2_IMPORTED | ||
| #include <winsock2.h> |
Owner
There was a problem hiding this comment.
by including these headers here, you are going to be pulling them into a huge number of places. If you are pulling them in just because you want SOCKET defined, SOCKET is unsigned int*
schombert
reviewed
Jan 14, 2026
| shutting_down = 2 | ||
| }; | ||
| #ifdef _WIN64 | ||
| typedef SOCKET socket_t; |
Owner
There was a problem hiding this comment.
also, using socket_t = ... is generally preferred these days over typedefs
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Import concurrentqueue.h from https://github.com/cameron314/concurrentqueue to be used as MPMC queue where needed.
Changes console_log to instead push to a concurrentqueue which is then procsessed by a seperate thread. This thread is initialized in the sys::state constructor, and closed in the destructor
Add SPSC queue for queueing up functions on the UI thread from the update thread. Eg changing cursors does not work properly unless executed on the same thread. Did a similar thing to for error windows being queued from the update thread
Made it so that only a special "normal_cancel_busy" cursor type can cancel the "busy" cursor, so that it isnt accidentually canceled by the player dragging their mouse or clicking.
Fixed a crash bug due to the cache-thread reading from uninitialized memory while the gamestate was reloading. It is fixed by using a shared_mutex, and having the readers (UI-thread, cache-thread) acquire a shared_lock, while the gamestate reset code acquires a unique_lock
Change loading a save to run as a command issued by the UI. This makes some other things easier, and also has the benefit of not freezing the UI in the process. It also means the command_lock can be removed fully as this was its only use
Add graceful client disconnects. Ie. the client will only be fully disconnected once all the pending data in the buffer has been flushed. This allows for the kicked/banned/disconnected client to know that they have been disconnected
Change server outgoing command queue and client outgoing command queue to be a concurrentqueue instead of SPSCqueue, as sometimes it is needed to queue commands directly from the update thread too, not just UI thread
Move the network client to dcon, with the exception of dynamic send/receive buffers. Those are a problem since the allocated memory will be freed on gamestate reload, but not recovered afterwards.
Change server -> client send buffers to be a queue of shared_ptr<command_data> intead of raw bytes. Increasing the amount of send() calls did not affect network performance much, and it significantly reduces the copying of large commands (eg save game data)
On the other hand, repeat recv() calls were a significant bottleneck in network performance in my tests. I changed the command-receive function to buffer the data in a 256KB buffer and then parse the commands from that. This significantly increased network performance when a large amount of commands were sent
Replaced the previous system of in-line adding network commands to client buffers directly to instead use the command queue, with a command_data wrapper that selects which clients should receive it. The new system is more stable as previously it relied upon there being no previous data in the send buffer to the client, but that was not a guarrentie. It was therefore possible to receive utility commands out-of-order sometimes
Change chat messages to reserve two lines of space per message by default. I was unable to find a solution to the memory corruption bug with the better variadic message length implementation. I may try to get it to work later
Add full-flegded private messages as present in other paradox games, in which you can select precisely which players should receive your message from the TAB-menu
add player limit to host settings
add player status for being OOS in the ui
And in the process of doing the above, refactoring a decent amount of the command/network code.