How to implement Iterator and IntoIterator for a simple struct?

Your iterator type is Iterator<Item = Self::Item>, but Iterator is a trait. Traits are implemented by structs, they don’t exist on their own. You could also have a reference trait object (&Iterator), a boxed trait object (Box<Iterator>) or an anonymous trait implementation (impl Iterator), all of which have a known sizes.

Instead, we create a PixelIntoIterator that has a known size and implements Iterator itself:

struct Pixel {
    r: i8,
    g: i8,
    b: i8,
}

impl IntoIterator for Pixel {
    type Item = i8;
    type IntoIter = PixelIntoIterator;

    fn into_iter(self) -> Self::IntoIter {
        PixelIntoIterator {
            pixel: self,
            index: 0,
        }
    }
}

pub struct PixelIntoIterator {
    pixel: Pixel,
    index: usize,
}

impl Iterator for PixelIntoIterator {
    type Item = i8;
    fn next(&mut self) -> Option<i8> {
        let result = match self.index {
            0 => self.pixel.r,
            1 => self.pixel.g,
            2 => self.pixel.b,
            _ => return None,
        };
        self.index += 1;
        Some(result)
    }
}

fn main() {
    let p = Pixel {
        r: 54,
        g: 23,
        b: 74,
    };
    for component in p {
        println!("{}", component);
    }
}

This has the nice benefit of returning actual i8s, not references. Since these are so small, you might as well pass them directly.

This consumes the Pixel. If you had a reference to a Pixel, you’d need to also implement an iterator that doesn’t consume it:

impl<'a> IntoIterator for &'a Pixel {
    type Item = i8;
    type IntoIter = PixelIterator<'a>;

    fn into_iter(self) -> Self::IntoIter {
        PixelIterator {
            pixel: self,
            index: 0,
        }
    }
}

pub struct PixelIterator<'a> {
    pixel: &'a Pixel,
    index: usize,
}

impl<'a> Iterator for PixelIterator<'a> {
    type Item = i8;
    fn next(&mut self) -> Option<i8> {
        let result = match self.index {
            0 => self.pixel.r,
            1 => self.pixel.g,
            2 => self.pixel.b,
            _ => return None,
        };
        self.index += 1;
        Some(result)
    }
}

If you wanted to support creating both a consuming iterator and a non-consuming iterator, you can implement both versions. You can always take a reference to a Pixel you own, so you only need the non-consuming variant. However, it’s often nice to have a consuming version so that you can return the iterator without worrying about lifetimes.

it’d be much more convenient to write this by reusing iterators that already exists, e.g., with [T; 3]

As of Rust 1.51, you can leverage array::IntoIter:

impl IntoIterator for Pixel {
    type Item = i8;
    type IntoIter = std::array::IntoIter<i8, 3>;

    fn into_iter(self) -> Self::IntoIter {
        std::array::IntoIter::new([self.r, self.b, self.g])
    }
}

In previous versions, it might be a bit silly, but you could avoid creating your own iterator type by gluing some existing types together and using impl Iterator:

use std::iter;

impl Pixel {
    fn values(&self) -> impl Iterator<Item = i8> {
        let r = iter::once(self.r);
        let b = iter::once(self.b);
        let g = iter::once(self.g);
        r.chain(b).chain(g)
    }
}