Optionally keep around an empty `Children` component

[gwafotapa]

What problem does this solve or what need does it fill?

I have a game involving cards which can be in different zones (hand, deck, discard, ...). Each zone is implemented as an entity with a Children component holding its cards. I'm using the Children component instead of my own component because I want the transform propagation.

When the last card of a zone is removed, the Children component is removed from the zone. Because of that, I always have to query for the zones with Option<Children>. More annoying, I have systems reacting to Changed<Children> for the zones (e.g. to update graphics whenever a card entered or left the zone). I need to handle separately the special case where the last children are removed because there will be no change detected since the Children component is not there anymore.

What solution would you like?

I would like the Children component not to be removed when its underlying vector is empty for these entities. The current behavior is what you usually want so I would like that new behavior to be opt-in. I haven't really thought how to implement it. I'm curious if others have run into that issue.

What alternative(s) have you considered?

As explained above, I'm doing without so this is not really a limitation but it would be nice to have.

[alice-i-cecile]

We should probably start by making this behavior configurable on a relationship type level, then think about how to expose it as a runtime toggle.

Are you particularly attached to using Children here?

[gwafotapa]

I'm using Children such that the Transform of the cards is relative to the Transform of the zone they are in.
If I was using my own component I would have to:

• either do the transform propagation relative to the zone myself;
• or use it alongside Children and do the bookkeeping associated with keeping both components in sync.

I thought using Children was the best solution out of these three but if you have better I'm listening !

[alice-i-cecile]

either do the transform propagation relative to the zone myself;

If this behavior was configurable at the relationship level, this is probably the solution I would pick :) Especially for a relative small n, where you don't really need be fully optimizing transform propagation.

I don't have any immediate suggestions; just trying to figure out if type-level configuration would be a meaningful incremental step forward.

[gwafotapa]

just trying to figure out if type-level configuration would be a meaningful incremental step forward.

Probably not.