Freeciv

git bisect points to b74c41379c659066aaaf5e27e704c71a91362c4e. Whether the bug is there or just surfaces it, I don't know.

I couldn't reproduce with current S2_6 HEAD.

Tried again, with fddec02b63370ad8aa99e696f58f4548206f0e71. Fails for me.

I failed to detail all the steps:
...
Build a buoy
Move your units out of the zone
End turn and see your visibility reduced
End another turn and see yet more FoW.

If you were already changing turns and still worked for you, these are my build flags (though I don't know how they could be related to this kind of problem):

./autogen.sh --disable-nls --no-configure-run --disable-ruledit --disable-freeciv-manual
configure --prefix="${MY_BUILDS_DIR}/buoy" --enable-gitrev --enable-client=gtk3 --disable-nls --disable-ruledit --enable-fcmp=no --enable-mapimg=no --disable-sdl-mixer --enable-freeciv-manual=no

I'd say this is a duplicate of #781551. Same assertions (only one is referenced in the other report, but both appear), there are buoys, and visibility is reduced after turn change.

I did play several turns forward with no FoW problems. I did take some shortcuts with editor compared to your reproducing instructions first, though, but now that I tried to follow them as accurately as possible I still can't reproduce a problem.

Do you have a savegame where from this should be more trivial to reproduce (just ending a turn?)

The one attached in my first comment should do the trick. Ending turn reduces visibility and another turn reduces it further taking even the city into obscurity.

Ok, got it reproduced from that savegame. It might be relevant that the buoy tile is also worked by the city. Or that it's adjacent to a city and within ones borders (in earlier attempts I usually built buoy one tile further (in hex topology).

First backtrace from the failing asserts suggests that the real vision count gets messed up when autosettlers do virtual changes (in this case: pillaging buoy) to see what their overall effect would be.

Marko Lindqvist wrote:

First backtrace from the failing asserts suggests that the real vision count gets messed up when autosettlers do virtual changes (in this case: pillaging buoy) to see what their overall effect would be.

Actually it considers Transforming the ocean tile to land, which would indirectly cause destruction of the buoy.

1. It doesn't sound like it's relevant, but just in case ... the ocean tile where the buoy was put had originally been swamp. Buoy was built on it after transformation.

2. Turn change would create erratic behaviours. For example, you could move a unit on a seen tile, and the tile you move onto with the unit becomes fogged, and seemingly random tiles nearby might toggle fogged or unfogged in chaotic unpredicted ways.

So it's definitely something about how it calculates visible tiles and some kind of bad logic there is toggling