Update tilemap space on tree notifications

[NewDefectus]

Fixes #83291. As mentioned in that issue, the relevant problem (collision bugginess when reparenting TileMaps to SubViewports) first appeared in #81070. I've tried researching the cause of this regression and why it didn't occur before that commit, but I'm not too sure honestly. So this fix doesn't directly tackle the issue of why this needs to be done at all, especially why this only occurs for SubViewports, but it definitely works and I think it's elegant enough not to cause any other issues.

[AThousandShips]

CC @groud

[groud]

Hmm, I've just checked. I am not sure this is the right fix. Technically, the re-creation of all physics shapes should already happen thanks to _physics_was_cleaned_up being true.

I suspect the issue would be due to RID space = tile_map_node->get_world_2d()->get_space();. As in a subviewport, the expected space might change.

I think the solution would be to instead update _physics_notify_tilemap_change to update the bodies space on DIRTY_FLAGS_TILE_MAP_IN_TREE change or something. That would avoid recreating all the physics shapes.

[NewDefectus]

Hmm, I've just checked. I am not sure this is the right fix. Technically, the re-creation of all physics shapes should already happen thanks to _physics_was_cleaned_up being true.

I suspect the issue would be due to RID space = tile_map_node->get_world_2d()->get_space();. As in a subviewport, the expected space might change.

I think the solution would be to instead update _physics_notify_tilemap_change to update the bodies space on DIRTY_FLAGS_TILE_MAP_IN_TREE change or something. That would avoid recreating all the physics shapes.

Yeah, I figured this wasn't the optimal solution - what you say about the space changing makes more sense. I'll replace the commit.

[groud]

I think it's the right solution, but I kind of dislike the refactor you added with the other notifications. I understand why you would like to change it that way but it makes the code a lot harder to read IMO.

Animatable collisions is kind of an easy thing to mess up (it basically does a complex thing, like reverting transforms changes in _process then re-applying them later on). So I'd rather keep it easier to read, even it it implies a few more lines.

[akien-mga]

Superseded by #84968, which is your fix with the additional changes requested (speed tracked to include it in RC1 today).