Segfault in battle

A backtrace could help.

Do you have an autosave too?

Is this a regression? If this was in b18 already it should not go into b19.

> Am 27.10.2016 um 08:15 schrieb GunChleoc <email address hidden>:
>
> A backtrace could help.
>
> Do you have an autosave too?
>
> ** Tags added: crash
>
> ** Changed in: widelands
> Importance: Undecided => Critical
>
> ** Changed in: widelands
> Milestone: None => build19
>
> ** Changed in: widelands
> Assignee: (unassigned) => GunChleoc (gunchleoc)
>
> --
> You received this bug notification because you are subscribed to
> widelands.
> https://bugs.launchpad.net/bugs/1636966
>
> Title:
> Segfault in battle
>
> Status in widelands:
> New
>
> Bug description:
> Sorry to disturb the rc1 party but I found a bug. :-(
>
> When soldiers are fighting a segfault can occur when the battle only consists of one soldier.
> To be more specific:
>
> In src/logic/map_objects/tribes/soldier.cc :
> 1256 Soldier& opponent = *battle_->opponent(*this);
>
> The call might return nullptr when the other soldier is no longer part
> of the battle (in my cases it was the first_ soldier in battle_ who
> was missing / pointed to 0x0). The description of the method contains
> "Can return nullptr, but I have no idea what that means." which is
> what is happening here.
>
> Since someone was aware of (part of) the issue there might be a check for nullptr missing in soldier.cc. Otherwise there is probably some cancel-battle-code somewhere which is missing the part of removing the other soldier. Does anyone has an idea which of the cases applies?
> Sadly I can not provide any more information about what was happening since I wasn't observing the game. However, GDB is still running if that is any use.
>
> Edit: Happens in build-19-rc1
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/widelands/+bug/1636966/+subscriptions

I have no idea where the bug comes from or how to trigger it, so I don't know if it's a regression or not.

Sadly I have no savegame. I was running multiple games in parallel so they were overwritten. Since I wasn't participating in the games (7 AIs + passiv player) I don't know what was happening. But at least I can offer a backtrace:

Forcing flag at (64, 104)
Cmd_EnemyFlagAction::execute player(7): flag->owner(8) number=11
Cmd_EnemyFlagAction::execute player(8): flag->owner(7) number=1

Thread 1 "widelands" received signal SIGSEGV, Segmentation fault.
0x000000000119091e in Widelands::Coords::operator== (this=0x48, other=...) at ../src/logic/widelands_geometry.cc:31
31 return x == other.x && y == other.y;

(gdb) bt
#0 0x000000000119091e in Widelands::Coords::operator== (this=0x48, other=...) at ../src/logic/widelands_geometry.cc:31
#1 0x0000000001190973 in Widelands::Coords::operator!= (this=0x48, other=...) at ../src/logic/widelands_geometry.cc:35
#2 0x0000000000dc155a in Widelands::Soldier::battle_update (this=0x99e3ef0, game=...) at ../src/logic/map_objects/tribes/soldier.cc:1257
#3 0x0000000000d58e90 in Widelands::Bob::do_act (this=0x99e3ef0, game=...) at ../src/logic/map_objects/bob.cc:193
#4 0x0000000000d58d5d in Widelands::Bob::act (this=0x99e3ef0, game=..., data=3472) at ../src/logic/map_objects/bob.cc:179
#5 0x0000000000d7736c in Widelands::CmdAct::execute (this=0xc209220, game=...) at ../src/logic/map_objects/map_object.cc:98
#6 0x0000000000e204f9 in Widelands::CmdQueue::run_queue (this=0x7fffffffa090, interval=67, game_time_var=@0x7fffffff9ec8: 19081447) at ../src/logic/cmd_queue.cc:123
#7 0x0000000000d137eb in Widelands::Game::think (this=0x7fffffff9ec0) at ../src/logic/game.cc:555
#8 0x0000000000f34fb5 in InteractiveBase::think (this=0x311fa40) at ../src/wui/interactive_base.cc:306
#9 0x0000000000f48a54 in InteractivePlayer::think (this=0x311fa40) at ../src/wui/interactive_player.cc:169
#10 0x0000000000e9ad13 in UI::Panel::do_think (this=0x311fa40) at ../src/ui_basic/panel.cc:441
#11 0x0000000000e9a5dd in UI::Panel::do_run (this=0x311fa40) at ../src/ui_basic/panel.cc:183
#12 0x0000000000c0c6b6 in UI::Panel::run<UI::Panel::Returncodes> (this=0x311fa40) at ../src/ui_basic/panel.h:96
#13 0x0000000000d134dc in Widelands::Game::run (this=0x7fffffff9ec0, loader_ui=0x7fffffff9e30, start_game_type=Widelands::Game::NewNonScenario, script_to_run="", replay=false,
    prefix_for_replays="single_player") at ../src/logic/game.cc:525
#14 0x0000000000c06ebe in WLApplication::new_game (this=0x18c7690) at ../src/wlapplication.cc:1228
#15 0x0000000000c064b2 in WLApplication::mainmenu_singleplayer (this=0x18c7690) at ../src/wlapplication.cc:1088
#16 0x0000000000c05eb6 in WLApplication::mainmenu (this=0x18c7690) at ../src/wlapplication.cc:994
#17 0x0000000000c02a1e in WLApplication::run (this=0x18c7690) at ../src/wlapplication.cc:441
#18 0x0000000000c010c1 in main (argc=1, argv=0x7fffffffd738) at ../src/main.cc:49

(gdb) display &opponent
9: &opponent = (Widelands::Soldier *) 0x0
(gdb) display battle_
10: battle_ = (Widelands::Battle *) 0xe02cb50
(gdb) display *battle_
11: *battle_ = {<Widelands::MapObject> = {_vptr.MapObject = 0x125b368 <vtable for Widelands::Battle+16>, removed = warning: RTTI symbol not found for class 'bo...

It would be interesting to find out why battle_->first_ is nullptr, and if that's a bug. I think for now checking for nullptr should at least prevent the segfault.

GUn: I tried a Multiplayer game with
bzr8872[fri-portraits] on OSX
and
_widelands_dev_widelands_fri_portraits-3917_Debug_x64 on windows

Hosting on Windows (Barbarians, pass thorugh te mountains, Endless agme)
client OSX (Imperial) crashed immediately, actual reason masked by this bug

Playing the other way round, worked.

I will try some other Version (without this bug) and / or try to debug on OSX,

Fixed in build20-rc1