Autosave after desync cannot be used to resume game

I noticed that clients can try to join a still running game again after a desync, resulting in a server crash. We don't support late joins yet, so the client shouldn't join at all.

I can confirm the desync with the attached savegame. Separate bug report for #2: https://bugs.launchpad.net/widelands/+bug/1805325

Server log:

 MapSaver::save() for 'Astoria 2.R' took 1899ms
Game: Writing Map Data took 1900ms
Game: Writing Player Economies Info ... took 0ms
Game: Writing ai persistent data ... took 4ms
Game: Writing Command Queue Data ... took 2533ms
Game: Writing Interactive Player Data ... took 1ms
GameSaver::save() took 4816ms
SaveHandler::save_game() took 4816ms
[Host]: disconnect_player_controller(1, Player 2)
[Host]: disconnect_client(0, CLIENT_DESYNCED, )

Client log:

MO(7012,wheat): cancel_moving
MO(6492,marble): cancel_moving
MO(6514,log): cancel_moving
MO(6943,fish): cancel_moving
MO(4471,empire_builder): [transfer]: starting task [movepath] and setting location to road 8677
MO(3198,barbarians_soldier): [transfer]: starting task [movepath] and setting location to road 8933
MO(3229,empire_soldier): [transfer]: starting task [movepath] and setting location to road 8806
MO(9894,barbarians_battlearena): Checking soldier (3) level 1)
MO(9894,barbarians_battlearena): okay
MO(9894,barbarians_battlearena): Check done!
MO(6630,wheat): cancel_moving
MO(7305,granite): cancel_moving
MO(6667,log): cancel_moving
MO(5485,empire_donkey): [transfer]: starting task [movepath] and setting location to road 8750
MO(3519,atlanteans_stonecutter): [transfer]: starting task [movepath] and setting location to road 8504
MO(7200,log): cancel_moving
MO(7075,meal): cancel_moving
MO(3402,atlanteans_builder): [transfer]: starting task [movepath] and setting location to road 9625
MO(6390,fish): cancel_moving
[NetClient] Error when trying to receive some data: End of file.
[NetClient] Closing network socket connected to fe80::ce81:da84:109d:caa5%wlp12s0:7396.
[Client] Pong!
[Client] received NETCMD_INFO_DESYNC. Trying to salvage some information for debugging.
[Client]: disconnect(CLIENT_DESYNCED, )

Current trunk will segfault - we probably have some savegame compatibility issues here from when we ripped out the anti-congestion algorithm.

Thread 1 "widelands" received signal SIGSEGV, Segmentation fault.
0x000055555702bed6 in Widelands::Flag::fetch_pending_ware (this=0xa56474000004e1, game=..., dest=...)
    at src/economy/flag.cc:519
519 for (int32_t i = 0; i < ware_filled_; ++i) {
(gdb) backtrace
#0 0x000055555702bed6 in Widelands::Flag::fetch_pending_ware (this=0xa56474000004e1, game=..., dest=...)
    at src/economy/flag.cc:519
#1 0x000055555689b6f7 in Widelands::Carrier::pickup_from_flag (this=0x612000740d40, game=..., state=...)
    at src/logic/map_objects/tribes/carrier.cc:256
#2 0x000055555689af71 in Widelands::Carrier::transport_update (this=0x612000740d40, game=..., state=...)
    at src/logic/map_objects/tribes/carrier.cc:169
#3 0x000055555669699e in Widelands::Bob::do_act (this=0x612000740d40, game=...)
    at src/logic/map_objects/bob.cc:193
#4 0x0000555556696706 in Widelands::Bob::act (this=0x612000740d40, game=..., data=5166)
    at src/logic/map_objects/bob.cc:179
#5 0x00005555566d4692 in Widelands::CmdAct::execute (this=0x603001c44580, game=...)
    at src/logic/map_objects/map_object.cc:100
#6 0x0000555556885c...

Since the map Astoria allows seafaring, this bug is likely fixed with:

https://code.launchpad.net/~widelands-dev/widelands/bug-1811030-desync-ai/+merge/361689

I compiled r8934 with the mentioned fix for bug 1811030 and could no longer reproduce the desync with the above savegame. So I think this bug is indeed fixed now.

Fixed in build20-rc1