Netcode hotfix
Various netcode fixes are included in this branch. Most important of all of course is a fix for the Net_CloseConnection invalid node -32769 detected message thing, or rather what actually caused that to occur.
It's a merge to master since everything here should be compatible with 2.1.18: the only main changes are related to recieving packets (whether you're the host of a netgame or a client), which AFAIK shouldn't cause desyncs if you played with these fixes on a 2.1.18-hosted netgame (or hosted a netgame with the fixes and 2.1.18-using players join you).
See merge request !185
* if you sent it to a client rather than the server, game over, your connection is closed
* if files that don't exist or are too large are requested are listed, game over, your connection is closed (they should have been checked on YOUR side beforehand, silly)
* if the server has downloading disabled anyway, ...yeah, you get the idea
Don't worry, I made sure Got_RequestFilePak cleaned up the full file request list for the node in case of failure
Exit your server desync fix
This branch fixes this bug: https://mb.srb2.org/showthread.php?t=41811
Here's an explanation of how this bug is actually happening:
* Pressing Enter with "Start" selected calls `D_MapChange` to make a map change command. (Note that part of it is actually delayed because it's called from a menu, but this isn't relevant to the bug)
* `D_MapChange` itself calls `SV_StartServer` to start up the server, if you're supposed to be the server player.
* Pressing Esc while starting up a server like shown in the gif causes the server to be cancelled, and go back to the title screen.
* HOWEVER, since we're still in `D_MapChange`, the map change command is made anyway.
* Since we're not in a netgame as expected, no random seed is added, and the map change command has a total size of 9 bytes (1 for the id + 8 for the actual data). This is then stored in the net command buffer. The random seed would add another 4 bytes if it WAS added, note.
* Next time you start a server, and let it work as normal, two more net commands are added: an "add player" command (size 3), and a map change command (size 13, including random seed this time). Your net command buffer now has a total size of **25** (9 + 3 + 13).
* This net command buffer is then sent as a `PT_TEXTCMD` packet to yourself.
* SRB2 recieves the `PT_TEXTCMD` you sent to yourself, and some time later, tries to run the net commands stored in it.
* The first map command is read as if it's size 13 rather than size 9 (it isn't), mistakeningly thinking a 4-byte random seed is stored (since you ARE in a netgame now). This means your buffer reads what it thinks is the next net command starting from position **[14]** in the buffer.
* Bad luck, it gets an id of **0**, which doesn't correspond with any net command!
* This message is then printed in the console:
> WARNING: Got unknown net command [14]=0 (max 25)
* The game subsequently kicks you out the server with a "desynch" message.
tl;dr The game doesn't correctly cancel the map change when you cancel the server with Esc like that. This is not a good thing.
See merge request !188
Dedicated bonuses fix
Fixes the issue reported here: https://mb.srb2.org/showthread.php?t=42530
What it turns out is going on is that making the game bail out in the middle of Y_StartIntermission if you're a dedicated server's host prevents the game from awarding the players bonuses in coop mode. Therefore it's not just special stage bonuses, but ANY kind of bonuses that can cause desyncs if someone loses all their lives.
This can be merged to master since it's a change only for dedicated server hosts, and should otherwise be compatible with 2.1.18.
See merge request !186
SDL fixes and cleanup
Mostly cleanup tbh. I noticed a few things that bugged me when examining SDL's code recently, thought I might as well fix them up a bit.
These changes definitely does not affect netplay, so this can be merged to master just fine.
See merge request !183
Jet Jaw crash fix
Turns out not having MF_SHOOTABLE can cause the Jet Jaw to endlessly loop between the two states, until somehow LUA_CallAction inexplicably causes Z_StrDup to crash anyway (that one's a mystery to me, I'm not going to look into it right now). I tweaked A_JetJawChomp so this endless loop can't happen anymore.
See merge request !176
obj folders fix
Partial backtrack of changes I made in !174. Apparently removing all the `.gitignore` files from `/objs`'s subfolders removed the only reason Git was keeping the folders alive (Git doesn't care about empty folders as it turns out). So now they have `.gitignore`s again, with a warning in each not to remove them.
See merge request !177
.gitignore stuff
The following changes have been made to `.gitignore` files in the objs/ and bin/ folders:
* All the `.gitignore`s living in subfolders of objs/ have been done away with and replaced with a single `objs/.gitignore`, covering all the things that the subfolders ignored before (including depend.ped, if anyone is like me and has to remove that file manually)
* All the `.gitignore`s living in subfolders of `bin/Mingw` ignore .exes of any name, not just srb2win/srb2dd/srb2sdl. en.mo is also ignored now... or rather any .mo file (just in case). This is mostly for my own sanity and that of anyone else who uses EXENAME=[name.exe] when compiling with MinGW.
If you're not sure what `.gitignore` does exactly, as far as I'm aware it literally just tells Git what to "ignore" (or rather, not track), so anyone using Git GUI or some other Git program etc doesn't have to see changes certain files and can't accidentally commit the files or whatever.
See merge request !174
Admin ban fix
Okay, THIS time admin bans should work properly. Turns out the relevant code for banning did not properly consider the case of admins doing the banning, at least until now.
Unlike my last attempt at fixing admin bans (!165), this one would require both host and admin to have the fix exe assuming everyone else would be using 2.1.17, so I'm merging to next instead of master.
See merge request !167
Alam Ed Arias