Consider adding an owned version of WeightedChoice

[mneumann]

This can more easily embedded within any other type, than the WeightedChoice type.

https://github.com/mneumann/graph-annealing/blob/master/src/owned_weighted_choice.rs

Or, maybe better would be to add code to reuse some code of weighted_choice?

[keisetsu]

I would really like to see this, too.

[mbrubeck]

I think it might even be better to provide only an owned version. The borrowed version does not seem to provide many useful advantages over an owned version. For example, you can't re-use the [Weighted<T>] slice after the borrow ends, because it has been mutated and no longer contains the same weights. This is a potential foot-gun too, since it leaves the [Weighted<T>] around in an unspecified state, easy to mis-use. Better to consume it and prevent such bugs.

On the other hand, the borrowed version allows building a WeightedChoice from a temporary stack array without allocating. Changing it to own a Vec<Weighted<T>> would prevent this...

[ramn]

I changed to this, which passes the test suite:

pub struct WeightedChoice<T: AsMut<[Weighted<U>]> + AsRef<[Weighted<U>]>, U: Clone> {
    items: T,
    weight_range: Range<u32>,
    _phantom: marker::PhantomData<U>,
}

But it looks a bit crazy.

[ramn]

Opened a PR #151 to get the ball rolling.

[mbrubeck]

Heh, I submitted PR #152 which takes a different approach, adding a new owned type while leaving the existing type unchanged. The main advantage of #152 is that it is not a breaking change. The disadvantage is that it adds a redundant type and impls.

[ramn]

Haha quick work! Yes yours looks more proper, and non-breaking!

[ramn]

Hm build failed for mine, but cargo test locally works fine

[ramn]

Not an argument for my specific implementation, but one could argue that the current version of WeightedChoice is not very idiomatic and perhaps should be retired. The design seems awkward.