Gameplay suddenly gets very slow

Attached logfile

Thanks for your bugreport :-)

We will look at this... currently all developers are a bit busy. So this is just a comment to say that your bugreport was noticed ;)

Main reason for slowdown is gamesave. Did you run it in windowed mode to see in real time what is being printed into terminal?

I usually, pause the game, wait til fps gets back and then increase speed back...

I get a significant drop around 5h 22min in the attached savegame, so this is confirmed. Thanks for reporting :)

@TiborB: I attached the logfile. I do not see anything suspicious in there.

@all: I could "fix" the slowdown by fixing the aforementioned problem where the soldiers got stuck and wouldn't move. This issue (and how to resolve it) is described in #1395238. I ran in to the same problem a few hours later, this time with a lot more soldiers. The game slowed down to around 0.25FPS. I believe that the slowdown is connected to the "soldiers stuck" situation. It looks like the slowdown is proportional to the amount of soldiers being blocked.

We have a bug about soldiers getting stuck somewhere - these bugs are hard to reproduce. Are there stuck soldiers in the savegame that you attached? Could you give a bit of a description / a screenshot so we can find the sport on the map easily?

Soldiers getting stuck bugs:

https://bugs.launchpad.net/widelands/+bug/1395238 (recent)

https://bugs.launchpad.net/widelands/+bug/730070 (this one is pretty old)

https://bugs.launchpad.net/widelands/+bug/1251914 (might be fixed)

Can a problem be this?

MO(8557,road): splitting road: first part:
MO(8557,road): * (43, 101)
MO(8557,road): * (44, 100)
MO(8557,road): * (45, 100)
MO(8557,road): second part:
MO(8557,road): * (45, 100)
MO(8557,road): * (46, 100)
MO(8557,road): * (47, 100)
MO(8557,road): * (47, 101)
MO(8557,road): * (48, 101)
MO(8557,road): * (49, 101)

I tested the savegame few times, but everytime I would say it happened shortly after this. Can anybody confirm?

'splitting road' probably will not be a culprit at the end..

I think we should focus on solving #1395238 because I am quite sure those two problems are related. If soldiers do not get stuck, the game does not slow down.

However, even if there is a bug causing soldiers to get stuck, the game should not slow down. The goal of this bugfix should be to make the codebase more robust even in pathological situations like this one.

I just did some testing and recognized some important aspects:

* The game speed does not slow down but the framerate decreases
* Its only noticable if you accelerate the gamespeed (try 3x or 4x)
* One core of my CPU is at 100% when the situation ocurs (game speed: 3x)
* The slowdown starts when the opponents returned from the battlefield to their buildings (shot005). As long as they are on the battlefield, their is no slowdown (shot004)

Added shot005.png. This is the situation where the slowdown occurs.

I am pretty sure that the problem are the repeating calls to start_task_movepath. I created a patch which requires attacking soldiers to find an unblocked path to do so. If they do not find one, they retreat.

I posted the patch on Bug #1395238:

https://bugs.launchpad.net/widelands/+bug/1395238/comments/6

Are not they already on their way home? Generally I dont like additional flag only_unblocked, I think algorithm should be common for all bobs. Of course animals usually are not so crowded.

Oh, I looked better, indeed this is about attacking not heading home...

My solution is rubbish. Please ignore the patch. If I start an attack
with +10 soldiers, only half of them make it to the enemy lines. The
others return to their building before they even started an attack.

I am currently trying to implement GunChleocs idea [1] of random
destinations to break up the deadlock.

[1] https://bugs.launchpad.net/widelands/+bug/1395238/comments/8

Am 08.12.2015 um 13:43 schrieb TiborB:
> Oh, I looked better, indeed this is about attacking not heading home...
>

--
--------------------------
Janosch Peters
Südmährenstraße 78
85586 Poing

<email address hidden>
0151 40 25 25 69

Fixed in r7687.

Fixed in build19-rc1.