Spawning a child entity in or after POST_UPDATE won't be properly registred as child of the parent

[lassade]

Bevy version
0.3 (commit fae6287)

Operating system & version
Fedora 31

What you did
Spawn a child entity during or after POST_UPDATE stage
You should only use the spawn function and then manually add the parent ...

What you expected to happen
The entity GlobalTransform should reflect it's parent movement

What actually happened
The entity was sitting still in the origin (0, 0, 0) while it's parent was moving around

Additional information

[svents]

I think this happened in the following commit: 43aac1a

In the parent_update_system (which runs in POST_UPDATE) a query filter was added like this:
mut changed_parent_query: Query<(Entity, &Parent, Option<&mut PreviousParent>), Changed<Parent>,>,
Changed is reset after each frame, so the entities in question are no longer picked up in the next frame as before. I think removing Changed here does no great harm, because the system has to verify anyway that the value actually changed (Changed only tracks mutable access).

[lassade]

I think you are right but I also thinkg this pr #907 shouldn't be the right call for it. I think the parent_update_system should instead run at the end of each stage just like the command buffer;

[cart]

I think the parent_update_system should instead run at the end of each stage just like the command buffer;

It would fix this specific issue, but I would very much prefer to not run the same system in every stage. That seems unnecessary and a bit hack-ey. Ideally I think it makes more sense to make Parent/Child updates transactional to avoid the issue entirely (or remove the Changed filter like #907 does, which is probably the better short term solution).

[lassade]

Yeah, I guess the perf won't take that big of a hit

[alice-i-cecile]

If #68 is fixed this should be resolved.