0001053: Battle continues after AI clears Bank

(0002932)
Tow (developer)
2012-08-26 19:17

Hmm... when trying to fix this issues, I found and fixed three serious bugs. Not sure which was that or if it really fixed, but I'm not able to reproduce crash anymore.
So seems to be fixed in r2848.

By the way, the attached save seems to be corrupted — creatures are missing bonuses with damage values. They deal 0 damage and thus battle goes on indefinitely. However my saves work fine, so I believe it's something on your side.

(0002946)
Warmonger (administrator)
2012-08-28 19:27
edited on: 2012-08-28 19:29

It still happens often for me, crash in bank with level-up on AI turn.

Uploaded another save, but not sure if AI will trigger the crash again.
Didn't happen on tiny test map for thousands of turns, though.

(0002953)
Tow (developer)
2012-08-30 15:52

I spent quite a while trying to reproduce this but I'm unable. I don't think it is the same bug as originally reported, since... it seems pretty well fixed. :P

When the crash happens again, please try capturing the stacktrace from debugger, saving whole log files, and so on. I need more information to investigate and I can't get them being unable to reproduce the issue.

(0002955)
Warmonger (administrator)
2012-08-31 07:19

Uploaded some crash dumps.

Also, similiar issue happens when neutral creatures flee, but battle is started nonetheless. Didn't reproduce it today yet.

(0002956)
Tow (developer)
2012-08-31 10:56

To read crashdumps I need your debug symbols (.pdb files) for the exact build you used.
[By exact build I mean that .pdb must come from the very same compilation as the exe that generated dump — even rebuilding same sources generates incompatible symbols.]

I'll look into logs in the meantime.
Look to be a race condition between post-battle callbacks and client-handling connection thread.

(0002957)
Ivan (developer)
2012-08-31 11:38

I got some crashes with "Armageddon rains, everyone dies" scenario.
Server have tendency to crash in
CGameHandler::checkForBattleEnd()
because gs->curB is null. (don't have full stacktrace right now).
Looks like race condition to me.

I thought that this was same as 0000294 and haven't reported it.

(0002958)
Tow (developer)
2012-08-31 13:44

I doubt it is related, since the crashes here are related to creature bank handling. (unless it is bigger race condition bug affecting all battles)
But I'll investigate the Armageddon scenario as well, VCMI should never crash.

(0002960)
Zamolxis (viewer)
2012-08-31 16:30

In regards to the original report: does Blue attack Cyclops Stockpile at the end of his movement points? There's a bug in VCMI 0.88 & 0.89 when attacking the Crypt, that after victory & reward a 2nd battle (with no enemy) is called. One of the conditions to reproduce it, is that hero has to enter the Crypt with the last of his movement points. > Does that sound like a bug that would have been corrected by Tow's fixes in r2848, or should I create a separate report?

(0002962)
Tow (developer)
2012-08-31 16:41

@Ivan
When using Armageddon to kill I did observe the crash... but on the client side, caused by the stack queue fail to update. Fixed in r2869. Now I can't observe any crash after casting Armageddon.

@Zamolxis
Great to see you again! :)
Fix from r2848 corrected bugs that I introduced when rewriting battle callbacks recently. The bug you describe... well, it looks very similar to the issue Warmonger is reporting here, it may be the same bug. I'll try reproducing that last movement points condition.

(0002965)
Ivan (developer)
2012-08-31 17:01

Tried same scenario, same hero, same battle - no crash. Probably was fixed recently.

(0002966)
Zamolxis (viewer)
2012-08-31 17:09
edited on: 2012-08-31 17:10

@Tow
Nice to be back (though by far not at the end of my hurdles with the house renovations, so it's hard to predict how active I'll be able to be here in the near future)
Take Ivor in the map I just attached to reproduce:
- send him south-east on the road just before FoW, then the underground gate
- attack halflings, but let them flee
- attack nearest Crypt on the upper row (should be reachable with last move)
- win the battle, take the skill, but then, instead of reward, you're asked if you want to attack again... and then follow some more buggy behaviors.

Let me know if you can still reproduce. If not, I'll create a separate (more detailed) report with screenshots & crashdump.

(0002968)
Tow (developer)
2012-08-31 17:35

Thanks, well done. I managed to reproduce issue with SVN build. After levelling-up another battle was offered. However there was no enemy when it started... I managed to go through it without crash but it's obviously a bug (and 0.89 or AI-only battle may crash).

I guess it's rather related to getting a level in creature bank than to last movement points. At the second attempt, there was less creatures in crypt, Ivor didn't level-up and the bug didn't show up.

(0002969)
Zamolxis (viewer)
2012-08-31 17:45
edited on: 2012-08-31 22:00

If I end turn anywhere on the path, and continue next day (not end of movement), the bug is not reproducible. Anyway, it's clearly a different one that what has been fixed so far then, so I'll create a separate report. We can discuss more there.

EDIT: 0001063 created (map deleted here and attached over there)

(0003492)
Tow (developer)
2013-04-20 11:39

Fixed in r3336.

Issue History
Date Modified	Username	Field	Change
2012-08-26 06:48	Warmonger	New Issue
2012-08-26 06:48	Warmonger	Status	new => assigned
2012-08-26 06:48	Warmonger	Assigned To	=> Tow
2012-08-26 06:48	Warmonger	File Added: BankCrash.rar
2012-08-26 07:04	Warmonger	Additional Information Updated	View Revisions
2012-08-26 19:17	Tow	Note Added: 0002932
2012-08-26 19:17	Tow	Status	assigned => resolved
2012-08-26 19:17	Tow	Fixed in Version	=> 0.dev
2012-08-26 19:17	Tow	Resolution	open => fixed
2012-08-26 19:19	Tow	Fixed in Version	0.dev =>
2012-08-28 19:27	Warmonger	Note Added: 0002946
2012-08-28 19:27	Warmonger	Status	resolved => feedback
2012-08-28 19:27	Warmonger	Resolution	fixed => reopened
2012-08-28 19:28	Warmonger	File Deleted: BankCrash.rar
2012-08-28 19:28	Warmonger	File Added: BankCrash2.rar
2012-08-28 19:29	Warmonger	Note Edited: 0002946	View Revisions
2012-08-30 15:52	Tow	Note Added: 0002953
2012-08-31 07:17	Warmonger	File Deleted: BankCrash2.rar
2012-08-31 07:18	Warmonger	File Added: Crash_conservatory.rar
2012-08-31 07:18	Warmonger	File Added: Crash_crypt.rar
2012-08-31 07:18	Warmonger	File Added: Freeze_Griffin_Conservatory.rar
2012-08-31 07:19	Warmonger	Note Added: 0002955
2012-08-31 07:19	Warmonger	Status	feedback => assigned
2012-08-31 10:56	Tow	Note Added: 0002956
2012-08-31 11:38	Ivan	Note Added: 0002957
2012-08-31 13:44	Tow	Note Added: 0002958
2012-08-31 16:30	Zamolxis	Note Added: 0002960
2012-08-31 16:41	Tow	Note Added: 0002962
2012-08-31 16:59	Zamolxis	File Added: VCMI_Tests_2012.h3m
2012-08-31 17:01	Ivan	Note Added: 0002965
2012-08-31 17:09	Zamolxis	Note Added: 0002966
2012-08-31 17:10	Zamolxis	Note Edited: 0002966	View Revisions
2012-08-31 17:35	Tow	Note Added: 0002968
2012-08-31 17:45	Zamolxis	Note Added: 0002969
2012-08-31 18:05	Zamolxis	Note Edited: 0002969	View Revisions
2012-08-31 21:59	Zamolxis	File Deleted: VCMI_Tests_2012.h3m
2012-08-31 22:00	Zamolxis	Note Edited: 0002969	View Revisions
2012-09-28 11:49	Warmonger	Relationship added	parent of 0000955
2013-01-28 07:52	Warmonger	Relationship added	parent of 0001191
2013-01-28 14:16	Warmonger	Relationship added	has duplicate 0001063
2013-03-30 15:39	Warmonger	Relationship added	has duplicate 0001251
2013-04-01 13:55	Ivan	Relationship added	related to 0001249
2013-04-20 11:39	Tow	Note Added: 0003492
2013-04-20 11:39	Tow	Status	assigned => resolved
2013-04-20 11:39	Tow	Fixed in Version	=> 0.93
2013-04-20 11:39	Tow	Resolution	reopened => fixed
2013-04-21 12:06	Tow	Relationship deleted	parent of 0001191
2014-05-30 17:43	beegee	Status	resolved => closed

Notes
(0002932) Tow (developer) 2012-08-26 19:17	Hmm... when trying to fix this issues, I found and fixed three serious bugs. Not sure which was that or if it really fixed, but I'm not able to reproduce crash anymore. So seems to be fixed in r2848. By the way, the attached save seems to be corrupted — creatures are missing bonuses with damage values. They deal 0 damage and thus battle goes on indefinitely. However my saves work fine, so I believe it's something on your side.

(0002946) Warmonger (administrator) 2012-08-28 19:27 edited on: 2012-08-28 19:29	It still happens often for me, crash in bank with level-up on AI turn. Uploaded another save, but not sure if AI will trigger the crash again. Didn't happen on tiny test map for thousands of turns, though.

(0002953) Tow (developer) 2012-08-30 15:52	I spent quite a while trying to reproduce this but I'm unable. I don't think it is the same bug as originally reported, since... it seems pretty well fixed. :P When the crash happens again, please try capturing the stacktrace from debugger, saving whole log files, and so on. I need more information to investigate and I can't get them being unable to reproduce the issue.

(0002955) Warmonger (administrator) 2012-08-31 07:19	Uploaded some crash dumps. Also, similiar issue happens when neutral creatures flee, but battle is started nonetheless. Didn't reproduce it today yet.

(0002956) Tow (developer) 2012-08-31 10:56	To read crashdumps I need your debug symbols (.pdb files) for the exact build you used. [By exact build I mean that .pdb must come from the very same compilation as the exe that generated dump — even rebuilding same sources generates incompatible symbols.] I'll look into logs in the meantime. Look to be a race condition between post-battle callbacks and client-handling connection thread.

(0002957) Ivan (developer) 2012-08-31 11:38	I got some crashes with "Armageddon rains, everyone dies" scenario. Server have tendency to crash in CGameHandler::checkForBattleEnd() because gs->curB is null. (don't have full stacktrace right now). Looks like race condition to me. I thought that this was same as 0000294 and haven't reported it.

(0002958) Tow (developer) 2012-08-31 13:44	I doubt it is related, since the crashes here are related to creature bank handling. (unless it is bigger race condition bug affecting all battles) But I'll investigate the Armageddon scenario as well, VCMI should never crash.

(0002960) Zamolxis (viewer) 2012-08-31 16:30	In regards to the original report: does Blue attack Cyclops Stockpile at the end of his movement points? There's a bug in VCMI 0.88 & 0.89 when attacking the Crypt, that after victory & reward a 2nd battle (with no enemy) is called. One of the conditions to reproduce it, is that hero has to enter the Crypt with the last of his movement points. > Does that sound like a bug that would have been corrected by Tow's fixes in r2848, or should I create a separate report?

(0002962) Tow (developer) 2012-08-31 16:41	@Ivan When using Armageddon to kill I did observe the crash... but on the client side, caused by the stack queue fail to update. Fixed in r2869. Now I can't observe any crash after casting Armageddon. @Zamolxis Great to see you again! :) Fix from r2848 corrected bugs that I introduced when rewriting battle callbacks recently. The bug you describe... well, it looks very similar to the issue Warmonger is reporting here, it may be the same bug. I'll try reproducing that last movement points condition.

(0002965) Ivan (developer) 2012-08-31 17:01	Tried same scenario, same hero, same battle - no crash. Probably was fixed recently.

(0002966) Zamolxis (viewer) 2012-08-31 17:09 edited on: 2012-08-31 17:10	@Tow Nice to be back (though by far not at the end of my hurdles with the house renovations, so it's hard to predict how active I'll be able to be here in the near future) Take Ivor in the map I just attached to reproduce: - send him south-east on the road just before FoW, then the underground gate - attack halflings, but let them flee - attack nearest Crypt on the upper row (should be reachable with last move) - win the battle, take the skill, but then, instead of reward, you're asked if you want to attack again... and then follow some more buggy behaviors. Let me know if you can still reproduce. If not, I'll create a separate (more detailed) report with screenshots & crashdump.

(0002968) Tow (developer) 2012-08-31 17:35	Thanks, well done. I managed to reproduce issue with SVN build. After levelling-up another battle was offered. However there was no enemy when it started... I managed to go through it without crash but it's obviously a bug (and 0.89 or AI-only battle may crash). I guess it's rather related to getting a level in creature bank than to last movement points. At the second attempt, there was less creatures in crypt, Ivor didn't level-up and the bug didn't show up.

(0002969) Zamolxis (viewer) 2012-08-31 17:45 edited on: 2012-08-31 22:00	If I end turn anywhere on the path, and continue next day (not end of movement), the bug is not reproducible. Anyway, it's clearly a different one that what has been fixed so far then, so I'll create a separate report. We can discuss more there. EDIT: 0001063 created (map deleted here and attached over there)

(0003492) Tow (developer) 2013-04-20 11:39	Fixed in r3336.