Pathfinding automatism bug

There seems to be a bug in the pathfinding automatism where the navigating object stops when selecting a new destination. I couldn’t find any pattern in that behaviour. If you just click around randomly for some time setting a new destination it once in a while will stop and not move again until a new destination is selected.
I found the issue in one of my own projects without any obstacles in the scene but it is reproduceable in the pathfinding demo.
I tested this on windows (native), will try it on Linux later.

The project in the attachment is the official pathfinding demo with all obstacles removed.

Can someone reproduce this issue?

The project “tbstrat” uses rounded values for aligning the destination to the grid. The issue happens a lot more often there.

“” updated with images (8.94 KB) (338 KB)

Images are missing for your tbstrat projects (as they are relative to a directory not included in the zip), could you add them? :slight_smile:
Can’t find any problem it seems with the HTML5 version of the pathfinding demo. I guess there is a glitch somewhere in the native one.

The same problem has been reported by another user some time ago.
I tested its example and some destinations that should be reachable aren’t (but they are from another start point !). If I deactivate the maximum iterator count in the pathfinding automatism, I get a infinite loop with these destinations.

Example project updated
By the way I’m using version 3.6.84

I did some further testing, it looks like the same fields are affected as in this post:

Cool thanks I’m gonna do investigation on this problem :slight_smile:
Surely the obstacle is considered slightly too large and so the pathfinding consider the adjacent cell as being covered by the obstacle.

I’ve fixed the issue with cells that are adjacent to obstacles, see here: viewtopic.php?f=19&t=6975&p=52887#p52887

Your example is very useful, I’m going to use it to make further tests. HTML5 pathfinding seems to work ok, I still have to test the native one.

I have successfully tested the checkerboard project on Ubuntu (official repository) and there are no problems.
Couldn’t find a version number in the IDE but I guess it is the same version as the Windows one.

Finished compiling the github code on Linux, everything is working as it should.

Could this be a Windows only issue?

Meanwhile, we found other issues with the pathfinding automatism. So, I think you haven’t tested the automatism in the exactly same way on both OSes (it’s not related to an OS). This was fixed for the next version of GDevelop.

The two things I tested were moving to a field directly next to a wall and visiting the green fields mentioned here in the screenshot below, both worked fine. I couldn’t find any valid fields on linux that were not accessible.
I couldn’t test the current version from github on windows because I may not install the compiler at work.

What where the other issues you found?

The remaining issue that has been fixed some hours ago :