Builder with closure syntax

[Veetaha]

This is based on a comment from @cksac in https://github.com/elastio/bon/issues/85#issuecomment-2343780384 It should look something like this. An additional config should enable the generation of a separate syntax for the builder where all required parameters are passed as positional parameters at the start of the function and optional parameters configured with the builder passed as an input to the trailing closure argument.

    use bon::bon;

    struct Tensor {}

    #[bon]
    impl Tensor {
        #[builder(closure_fn = sum_with)]
        fn sum(
            &self,
            dims: &[i64],
            #[builder(default)] keep_dim: bool,
            #[builder(default)] mean: bool,
        ) -> Tensor {
            Tensor {}
        }

    }

    let t = Tensor{};

    let output = t
        .sum_with(&[1, 2], |b| b.keep_dim(true).mean(false))
        // Automatically sets all optional parameters to default values
        .sum(&[1]);

I think this API looks too exotic, and it's probably fine if it would require a bit more code to write for the user. For example, if we just generate a From<TBuilder> for T impl, this API can be written manually using existing bon primitives (see the comment https://github.com/elastio/bon/issues/85#issuecomment-2343850854).

A note for the community from the maintainers

Please vote on this issue by adding a 👍 reaction to help the maintainers with prioritizing it. You may add a comment describing your real use case related to this issue for us to better understand the problem domain.

[cksac]

for compare with bon-generate all, if bon only generate From<TBuilder> for T

use bon::bon;

// TODO: GENERATED By bon
impl<__KeepDim, __Mean> From<SumOptionalParamsBuilder<(__KeepDim, __Mean)>> for SumOptionalParams
where
    __KeepDim: bon::private::IntoSet<Option<bool>, SumOptionalParamsBuilder__keep_dim>,
    __Mean: bon::private::IntoSet<Option<bool>, SumOptionalParamsBuilder__mean>,
{
    fn from(builder: SumOptionalParamsBuilder<(__KeepDim, __Mean)>) -> Self {
        builder.build()
    }
}

// User code
#[derive(bon::Builder)]
struct SumOptionalParams {
    #[builder(default)]
    keep_dim: bool,
    #[builder(default)]
    mean: bool,
}

struct Tensor {}

impl Tensor {
    pub fn new() -> Self {
        Tensor {}
    }

    // NOTE: orignal function need to be rewrite to accept options struct
    fn _sum(&self, dims: &[i64], options: SumOptionalParams) -> Tensor {
        Tensor {}
    }

    pub fn sum(&self, dims: &[i64]) -> Tensor {
        let options = SumOptionalParams::builder().build();
        self._sum(dims, options)
    }

    pub fn sum_with<T>(
        &self,
        dims: &[i64],
        options: impl FnOnce(SumOptionalParamsBuilder) -> T,
    ) -> Tensor
    where
        T: Into<SumOptionalParams>,
    {
        let options = options(SumOptionalParams::builder()).into();
        self._sum(dims, options)
    }
}

fn main() {
    let t = Tensor::new();
    t.sum(&[1]).sum_with(&[1, 2], |b| b.keep_dim(true));
    println!("test");
}

[cksac]

another version without closure, which allow early return and accept partial builder

use bon::bon;

// TODO: GENERATED By bon
impl<__KeepDim, __Mean> From<SumOptionalParamsBuilder<(__KeepDim, __Mean)>> for SumOptionalParams
where
    __KeepDim: bon::private::IntoSet<Option<bool>, SumOptionalParamsBuilder__keep_dim>,
    __Mean: bon::private::IntoSet<Option<bool>, SumOptionalParamsBuilder__mean>,
{
    fn from(builder: SumOptionalParamsBuilder<(__KeepDim, __Mean)>) -> Self {
        builder.build()
    }
}

// User code
#[derive(bon::Builder)]
struct SumOptionalParams {
    #[builder(default)]
    keep_dim: bool,
    #[builder(default)]
    mean: bool,
}

struct Tensor {}

impl Tensor {
    pub fn new() -> Self {
        Tensor {}
    }

    // NOTE: orignal function need to be rewrite to accept options struct
    fn _sum(&self, dims: &[i64], options: SumOptionalParams) -> Tensor {
        Tensor {}
    }

    pub fn sum(&self, dims: &[i64]) -> Tensor {
        let options = SumOptionalParams::builder().build();
        self._sum(dims, options)
    }

    pub fn sum_with(&self, dims: &[i64], options: impl Into<SumOptionalParams>) -> Tensor {
        let options = options.into();
        self._sum(dims, options)
    }
}

fn main() {
    let t = Tensor::new();
    let default_options = SumOptionalParams::builder().mean(true);

    let output = t
        .sum(&[1])
        .sum_with(&[1, 2], SumOptionalParams::builder().keep_dim(true))
        .sum_with(&[1, 2], default_options);
}

[Veetaha]

Yeah, I also was thinking of suggesting this, but this requires importing optional parameters struct type name.

[cksac]

another design is extract all params in function into param struct, like

use bon::bon;

// TODO: GENERATED By bon
impl<'a, __Dims, __KeepDim, __Mean> From<SumParamsBuilder<'a, (__Dims, __KeepDim, __Mean)>>
    for SumParams<'a>
where
    __Dims: bon::private::IntoSet<&'a [i64], SumParamsBuilder__dims>,
    __KeepDim: bon::private::IntoSet<Option<bool>, SumParamsBuilder__keep_dim>,
    __Mean: bon::private::IntoSet<Option<bool>, SumParamsBuilder__mean>,
{
    fn from(builder: SumParamsBuilder<'a, (__Dims, __KeepDim, __Mean)>) -> Self {
        builder.build()
    }
}

#[derive(bon::Builder)]
#[builder(start_fn = new)]
struct SumParams<'a> {
    dims: &'a [i64],
    #[builder(default)]
    keep_dim: bool,
    #[builder(default)]
    mean: bool,
}

struct Tensor {}

impl Tensor {
    pub fn new() -> Self {
        Tensor {}
    }

    pub fn sum<'b>(&self, params: impl Into<SumParams<'b>>) -> Tensor {
        let params = params.into();
        // todo logic
        Tensor {}
    }
}

// User can also implment into for SumParams
impl<'a, const N: usize> From<&'a [i64; N]> for SumParams<'a> {
    fn from(dims: &'a [i64; N]) -> Self {
        SumParams::new().dims(dims).build()
    }
}

fn main() {
    let t = Tensor::new();

    let output = t
        .sum(SumParams::new().dims(&[1, 2]))
        .sum(SumParams::new().dims(&[1, 2]).keep_dim(false))
        .sum(SumParams::new().dims(&[1, 2]).keep_dim(false).mean(false))
        .sum(&[1, 2]);
    //with https://github.com/elastio/bon/pull/125
    //.sum(SumParams::new(&[1, 2]).keep_dim(false).mean(false));
}

good: only 1 variant of original function bad: more verbose to call, can use custom implement into<FParams> if only single required params.

not sure which style is better.

[dzmitry-lahoda]

not sure if that was discussed, would be nice macro force all option parameters to be in the end in original function too(fail to compile if not). so macro will not reorder parameters as they written in text.

because if parameters will be reordered in macro, but not in original text, it may lead to subtle usage bugs.

[cksac]

Recap of above pattern. Personally, I prefer the closure style 1) > 2)

// 1)
let output = t
    .sum(&[1]);
    .sum_with(&[1, 2], |b| b.keep_dim(true).mean(false));

// 2) 
let output = t
    .sum(|b| b.dims(&[1, 2]))
    .sum(|b| b.dims(&[1, 2]).keep_dim(false).mean(false));

// 3)
let output = t
    .sum(&[1])
    .sum_with(&[1, 2], TensorSum__OptParams::builder().keep_dim(true).mean(false));

// 4)
let output = t
    .sum(TensorSum__Params::new().dims(&[1, 2]))
    .sum(TensorSum__Params::new().dims(&[1, 2]).keep_dim(false).mean(false));

• using closure

  • pros
    • hide the generated func param struct, don't need to be import
    • will not have name conflict with existing associated methods
  • cons
    • can't early return when set the params in clouser
    • can't use partial builder

• capture only optional function params to param struct

  • pros
    • easy to use when use default for optional params
  • cons
    • two variants for each method

• capture all function params to param struct

  • pros
    • only 1 variant for each method
    • consistent
  • cons
    • harder to know what params are required, need compile error to show