There's some bad behaviour introduced in r1274. When a server stops serving, all r1274+ clients will suffer a recursive error, a soft crash. Here's a technical description:

- ioq3, svn head, etc. is connected to a remote server
- the server decides to do a /disconnect or a /quit

- inside a normal VM_Call() to cgame...
- in the snapshot functions, CG_ExecuteNewServerCommands() handles every reliable server command
- for each, trap_GetServerCommand() is called
- which, through the syscall callback code, turns into a call to CL_GetServerCommand()
- that function recognizes the "disconnect" server command, and executes Com_Error() with ERR_SERVERDISCONNECT (note: com_errorEntered is set to true)
- in that sutiation, CL_FlushMemory() is called
- we don't have a server running, so we do a Hunk_Clear()
- there, VM_Clear() clears all the VMs via VM_Free()
- since the cgame is still running (vm->callLevel > 0), Com_Error() is called again, this time with ERR_FATAL
- that's our second error so far (com_errorEntered is true), so Sys_Error() shuts down the program