flatten and flatten_ref methods on ActiveValue

[PacificBird]

Closes #3145

[Huliiiiii]

Why not just:

impl<V> ActiveValue<Option<V>>
where
    V: Into<Value> + Nullable, 
{
...
}

[Huliiiiii]

Naming them as_option and into_option might be better.

[PacificBird]

Why not just:

impl<V> ActiveValue<Option<V>>
where
    V: Into<Value> + Nullable, 
{
...
}

This requires specialization afaik. The impl<V> ActiveValue<V> gives a conflicting implementation as far as the compiler is currently concerned.

[PacificBird]

Naming them as_option and into_option might be better.

That's definitely shorter and nicer.

[Huliiiiii]

Why not just:

impl<V> ActiveValue<Option<V>>
where
    V: Into<Value> + Nullable, 
{
...
}

This requires specialization afaik. The impl<V> ActiveValue<V> gives a conflicting implementation as far as the compiler is currently concerned.

Implementing it on Option<V> instead of V works correctly. (on my machine)

impl<V> ActiveValue<Option<V>>
where
    V: Into<Value> + Nullable,
{
    pub fn as_option(&self) -> Option<&V> {
        self.try_as_ref().and_then(Option::as_ref)
    }
 
    pub fn into_option(self) -> Option<V> {
        match self {
            Self::Set(value) | Self::Unchanged(value) => value,
            Self::NotSet => None,
        }
    }
}

[tyt2y3]

Agree with @Huliiiiii's approach — implementing on ActiveValue<Option<V>> drops the two sealed helper traits entirely and, as a bonus, sidesteps the compile errors the current trait version has:

• the AsRefOption<T> impl is missing its lifetime arg (AsRefOption<'a, T>),
• the flatten_ref bound constrains the owned V instead of &V, so it's unsatisfiable for any real caller,
• flatten(&self) can't move out of &self to return an owned value.

CI is red on all three doctests.

The as_option / into_option naming is good. One thing the rename gets for free: into_option takes self by value (matching the into_ convention), which is exactly the fix flatten needed.