segmentation fault - AI building enhancement

[TiborB]

I just encountered a crash, I run the game (AI-only game) under gdb so I collected some information:

Program received signal SIGSEGV, Segmentation fault.
0xb7b7a024 in std::string::c_str() const () from /usr/lib/i386-linux-gnu/libstdc++.so.6
(gdb) bt
#0 0xb7b7a024 in std::string::c_str() const () from /usr/lib/i386-linux-gnu/libstdc++.so.6
#1 0x08b46283 in DefaultAI::check_productionsites (this=0xbef6d80, gametime=370560) at /var/widelands/tibor-ai5/src/ai/defaultai.cc:2075
#2 0x08b3f478 in DefaultAI::think (this=0xbef6d80) at /var/widelands/tibor-ai5/src/ai/defaultai.cc:207

(gdb) frame 1
#1 0x08b46283 in DefaultAI::check_productionsites (this=0xbef6d80, gametime=370560) at /var/widelands/tibor-ai5/src/ai/defaultai.cc:2075
2075 BuildingObserver& en_bo = get_building_observer(bld.name().c_str());

the portion of code:

2072 // Only enhance buildings that are allowed (scenario mode)
2073 if (player_->is_building_type_allowed(enhancement)) {
2074 const Building_Descr& bld = *tribe_->get_building_descr(enhancement);
2075 BuildingObserver& en_bo = get_building_observer(bld.name().c_str());

I cannot repeat the crash, but I encountered the same crash before. Frequency is like once in 100 hours of gameplay. I may insert some printfs to find out at least what buildings are considered for upgrade and wait for next crash...

[Hans Joachim Desserud (hjd)]

>I cannot repeat the crash, but I encountered the same crash before.

Does it crash in the replay, or does that stop just before?

[TiborB (tiborb95)]

Well, I did not try, and now I dont know which game it was. I will next time...

[TiborB (tiborb95)]

I found the reason, basically two lines are relevant

const Building_Index enhancement = site.site->descr().enhancement();
if (player_->is_building_type_allowed(enhancement)) {

The problem was that if there was no enhancement to a building possible, first line returned INVALID_INDEX, and now it is bit funny, player_->is_building_type_allowed(INVALID_INDEX) returned true and algorithm went on as if INVALID_INDEX was real building.

I will commit a fix in tibor-ai5 branch if nobody minds

[TiborB (tiborb95)]

Fixed with comiting tibor-ai5 to trunk (r7204)

[GunChleoc (gunchleoc)]

I have now linked the branch to this bug report ;)

[GunChleoc (gunchleoc)]

Fixed in build19-rc1.