Tile entities are leaked upon recursively despawning the Map entity

[forbjok]

When the Map entity is despawned with .despawn_recursive(), all tile entities belonging to that tilemap continue to exist even though they are no longer visible. As far as I can tell, this is due to the tile entities not being marked as children of the Map entity.

[StarArawn]

Thanks for the report! I think we likely will want a map.despawn(commands, map_entity); api for this. I can't have tiles be children of chunks for various reasons.

[forbjok]

They don't necessarily need to be children of chunks, do they?
Both chunks and tiles could all just be children of the Map entity itself.

[StarArawn]

Part of the issue with having tiles be a child of the map entity or the chunk entity is that tiles don't have standard transforms, and bevy currently complains about that. We could give them standard transforms, but then its just a lot of extra data that is not needed and people might assume that changing a tile transform would change where a tile is located which isn't supported and isn't easy to support.

[forbjok]

tiles don't have standard transforms, and bevy currently complains about that

If that's the case, maybe we should suggest fixing it in Bevy. I don't see any reason why entities should be forced to have a Transform in order to be part of a hierarchy. Parent-child relationships are useful for a lot of things not related to positioning, such as in this case, for organizational purposes to make it easier to despawn all sub-entities when despawning the container entity.

[StarArawn]

tiles don't have standard transforms, and bevy currently complains about that

If that's the case, maybe we should suggest fixing it in Bevy. I don't see any reason why entities should be forced to have a Transform in order to be part of a hierarchy.

Bevy relations RFC would fix this issue. :)

See:
bevyengine/rfcs#18

[forbjok]

tiles don't have standard transforms, and bevy currently complains about that

If that's the case, maybe we should suggest fixing it in Bevy. I don't see any reason why entities should be forced to have a Transform in order to be part of a hierarchy.

Bevy relations RFC would fix this issue. :)

See:
bevyengine/rfcs#18

Yeah, this would probably allow it, assuming you can have the ECS enforce the deletion of certain relation types when their container entity is despawned.
This seems like a much more complex change though, and way overkill if the goal is simply to ensure sub-entities are despawned.

[StarArawn]

@forbjok I ran some tests with this, and I discovered a performance drop almost in half due to having a parent per child. If I instead have a map_query.despawn_layer(layer_id) would that be acceptable?

[forbjok]

Interesting. I wonder why that would affect performance when tiles don't have transforms that would need to be propagated. Have you looked into why it's affecting performance?

[StarArawn]

The performance is mainly lost due to how bevy_transform propagates the transform information for parents/children. It always iterates over the children regardless of whether they have transforms or not.

[StarArawn]

You can now use map_query.depsawn_map to despawn an entire map including tiles. Optionally you can despawn layers with depsawn_layer or tiles with: despawn_layer_tiles.