Confused about what it means when a child translation is relative to a parent translation?

[errant warren]

For example, I have this simple piece of code which spawns a rectangle relative to its parent which is a circle:

    commands
        .spawn((MaterialMesh2dBundle {
            mesh: meshes.add(shape::Circle::default().into()).into(),
            material: materials.add(ColorMaterial::from(BALL_COLOR)),
            transform: Transform::from_translation(Vec3::new(10., 10., 0.)).with_scale(Vec3::new(50., 50., 1.)),
            ..default()
        },
        Ball)).with_children( |parent| {
                parent.spawn((
                    SpriteBundle{
                        transform: Transform {
                            translation: Vec3::new(2., 0., 1.),
                            scale: Vec3::new(0.2, 0.2, 1.),
                            ..default()
                        },
                        sprite: Sprite {
                            color: RECT_COLOR,
                            ..default()
                        },
                        ..default()
                    },
                    Rect
                ));
            });

I have a system which prints out both translations for both the Transform and the GlobalTransform of each of these shapes. For the ball entity the system prints: ball transform xy: [10, 10], ball global xy: [10, 10]. This is expected because the ball has no parent. However, for the rectangle the system prints: rect transform xy: [2, 0], rect global xy: [110, 10]. I would expect the child transform to be a simple addition onto the parent transform and print: rect transform xy: [2, 0], rect global xy: [12, 10] but that is not the case. So what is the relationship between the two? The extra 100 in the x value of the GlobalTransform of the child, the 2 in the Transform and the 50 as the width of the circle make me think its relative to the size of circle somehow. Is that the case?

[fiery zenith]

basically children's transforms are just matrix multiplications of it's parents and worlds

so for example, an objects transform in the world would be: world matrix * object matrix
If you have a child,
then it's calculated as such:
world matrix * object matrix * child matrix
or if we sum it up,
first_child = object * child

so it's not just an addition, it takes all the properties the parent has in the matrix, that is, translation, size, rotation

then it applies those to the child

you can have mutiple children, and they multiply it's parents matrix

hope that made some more sense

[errant warren]

@fiery zenith Ok this is starting to make sense. I don't have too much experience with matrix operations, is there any programmatic way to determine how to place the child transform so that it's global transform is at [12, 10]? Since setting the child transform to being [2, 0] does not work as I expected.

[fiery zenith]

there's a way to get global transfrom handle i believe

it's a different query

you can get it from there

[quick token]

GlobalTransform

it's called

I'm not sure if changing GlobalTransform gets propagated to Transform, @fiery zenith any idea?

I thought it was only the other way around

[fiery zenith]

you need to dereference it, it should work both ways

i never tried seeing if it actually changes local transform

[quick token]

it has to be synced up at some specific point too

[limpid cape]

I thought it only goes local -> global automatically. You shouldn't really be setting global transforms

[fiery zenith]

you can use it to see how to set it up

you can do math with parent to see how it works

[limpid cape]

You have .with_scale(Vec3::new(50., 50., 1.)), on your parent though, so that means all the transformes of the children will be scaled that much

[fiery zenith]

or reverse it

[quick token]

Parent's scale modifies the child local transform's translation?

[limpid cape]

the local transform is the same, but when it gets converted to global, the parent transform is used in the conversion

you can think of the parent transform as defining a new "space" for its children

if a parent has a scale of (5, 5) and a rotation of 90 degrees counterclockwise, then a child with a transform of (0, 2) (two units up) would become (-10, 0) 10 units left.

[quick token]

what if child was rotated, would the angle also get scaled? or would it shear

it's probably consequences of the mat multiplication, i don't remember the exact way it interacted

i usually worked with one transformation at a time

[limpid cape]

it wouldn't shear

you can think of doing all the children first, then doing the parent and all its children together

Say you were trying to draw this scene

with the ball at 3,3 with a scale of 1 and the plane at 0,0 with a scale of 10

if the ball was set to 3,3 and the plane has a default scale of 1 instead of 10, that would look like this, right?

point B is 3,3, point A is 5,5

[quick token]

yup

[limpid cape]

so now if you modify the parent scale, nothing moves relative to each other, we just have to update our labels now. C is 5,5 B is 30,30 (3from its translation*10 from the parent scale)

[quick token]

which one is the parent? the plane?

[limpid cape]

yes, sorry

[quick token]

so parent's scale only scales child's position

angle offsets angle

and position offsets scaled position

and child's scale is unaffected?

[limpid cape]

scale will also make the ball bigger

rotation will also move the ball around the parent;s origin

[quick token]

so parent scale is multiplied with child scale

hmm

alright

ty for explanation

[limpid cape]

so the two options to fix this are

1. to compensate when setting the ball transform and use numbers that are relevant to the parnet before it is transformed

2. move the parent's mesh to a second child so you can move it independently, and then only use the parent's transform when you are trying to move everything together

[quick token]

the 2nd solution seems much cleaner

that or just scaling mesh directly

but yeah, i ran into that with autodesk and it was very annoying

tis just how math works 🙁

[errant warren]

Ah this is much clearer now, thank you everyone for the explanations. I also think option 2 is the ideal solution.