Remove lots of unused logic in game server messaging #1463
Conversation
Yes, all the plNetClientMsgHandler::PeekMsg calls really did nothing, because plNetMessage::fNetCoreMsg was never set anywhere.
This simplifies the writing process especially, where the previous implementation actually had to write the message twice: first to calculate the required buffer size, then another time to actually fill in the buffer. Now the message is written directly into a hsRAMStream.
The channel number was ignored when sending messages to the server. The plNetTransportMember subscription lists were updated in various places, but also never actually used for anything.
Now there's only one single message handler, previously called the "default handler", which takes care of all message dispatching. The previous type-based message handling was already completely unused.
|
The channel/transport groups were one of the hairier removals, yeah. I'm relatively confident about that though - all the code that adds channel subscriptions is guarded behind I have to admit my multiplayer testing wasn't very in-depth - I mainly just checked that everything was still working at all. I suppose this could be a good thing to test on TrollLand with a small group of people. The voice chat filtering is particularly difficult to test though, because if all goes well, the filtering is done by both the sender (talk list) and the receiver (listen list). So even if I broke the filtering on send, I wouldn't notice, as long as the filtering on receive still works. Testing this properly would require looking at packet captures I guess... That said, the voice talk/listen lists are their own lists that are maintained separately from the channel group stuff. I didn't touch the voice code except for removing the P2P branches, so hopefully that works as before... |
Nothing used the option to send a plNetMessage to the auth server. DIRTSAND and MOSS don't support handling such messages either.
The parsed plNetMessage is now passed directly to the actual handler, making the code much easier to follow.
dgelessus
force-pushed
the
netclient_simplification
branch
from
bda752f
to
bba3838
Compare
Follow-up to a few things I mentioned in #1377. This simplifies the game server message send/receive code by removing a lot of unused logic. Specifically:
plNetMessagehad some incomplete code for reading/writing messages from/to memory buffers instead of streams - presumably to reduce copying? The actual message IO is still based onhsStreamthough, so the fixed-size buffers just complicated the API and logic for no reason. I changed it to usehsStreamall the way through.plNetMessages could theoretically be sent to the auth server instead of the game server, but this was never done - every single message was explicitly set to go to the game server, so I hardcoded that globally. (DIRTSAND and MOSS don't even implement the auth server PropagateBuffer messages necessary to support this.)switch. I removed all the dynamic message dispatching code and left only the single global message handler callback.I think all of these things will never be useful anymore (at least not without some serious reworking) so there's no reason to keep them - but if anybody disagrees or was planning to use any of this code, let me know 🙂