Expand description

Derive a builder for a struct

This crate implements the builder pattern for you. Just apply #[derive(Builder)] to a struct Foo, and it will derive an additional struct FooBuilder with setter-methods for all fields and a build-method — the way you want it.

Quick Start

Add derive_builder as a dependency to you Cargo.toml.

What you write

#[macro_use]
extern crate derive_builder;

#[derive(Builder)]
struct Lorem {
    ipsum: u32,
    // ..
}

What you get

#[derive(Clone, Default)]
struct LoremBuilder {
    ipsum: Option<u32>,
}

#[allow(dead_code)]
impl LoremBuilder {
    pub fn ipsum(&mut self, value: u32) -> &mut Self {
        let mut new = self;
        new.ipsum = Some(value);
        new
    }

    fn build(&self) -> Result<Lorem, LoremBuilderError> {
        Ok(Lorem {
            ipsum: Clone::clone(self.ipsum
                .as_ref()
                .ok_or(LoremBuilderError::from(UninitializedFieldError::new("ipsum")))?),
        })
    }
}

By default all generated setter-methods take and return &mut self (aka non-conusuming builder pattern). Accordingly, the build method also takes a reference by default.

You can easily opt into different patterns and control many other aspects.

The build method returns Result<T, E>, where T is the struct you started with and E is a generated builder error type. It returns Err if you didn’t initialize all fields and no default values were provided.

Builder Patterns

Let’s look again at the example above. You can now build structs like this:

let x: Lorem = LoremBuilder::default().ipsum(42).build()?;

Ok, chaining method calls is nice, but what if ipsum(42) should only happen if geek = true?

So let’s make this call conditional

let mut builder = LoremBuilder::default();
if geek {
    builder.ipsum(42);
}
let x: Lorem = builder.build()?;

Now it comes in handy that our setter methods take and return mutable references. Otherwise we would need to write something more clumsy like builder = builder.ipsum(42) to reassign the return value each time we have to call a setter conditionally.

Setters with mutable references are therefore a convenient default for the builder pattern in Rust.

But this is a free world and the choice is still yours!

Owned, aka Consuming

Precede your struct (or field) with #[builder(pattern = "owned")] to opt into this pattern. Builders generated with this pattern do not automatically derive Clone, which allows builders to be generated for structs with fields that do not derive Clone.

  • Setters take and return self.
  • PRO: Setter calls and final build method can be chained.
  • CON: If you don’t chain your calls, you have to create a reference to each return value, e.g. builder = builder.ipsum(42).

This pattern is recommended and active by default if you don’t specify anything else. You can precede your struct (or field) with #[builder(pattern = "mutable")] to make this choice explicit.

  • Setters take and return &mut self.
  • PRO: Setter calls and final build method can be chained.
  • CON: The build method must clone or copy data to create something owned out of a mutable reference. Otherwise it could not be used in a chain. (*)

Immutable

Precede your struct (or field) with #[builder(pattern = "immutable")] to opt into this pattern.

  • Setters take and return &self.
  • PRO: Setter calls and final build method can be chained.
  • CON: If you don’t chain your calls, you have to create a reference to each return value, e.g. builder = builder.ipsum(42).
  • CON: The build method and each setter must clone or copy data to create something owned out of a reference. (*)

(*) Performance Considerations

Luckily Rust is clever enough to optimize these clone-calls away in release builds for your every-day use cases. Thats quite a safe bet - we checked this for you. ;-) Switching to consuming signatures (=self) is unlikely to give you any performance gain, but very likely to restrict your API for non-chained use cases.

More Features

Hidden Fields

You can hide fields by skipping their setters on (and presence in) the builder struct.

  • Opt-out — skip setters via #[builder(setter(skip))] on individual fields.
  • Opt-in — set #[builder(setter(skip))] on the whole struct and enable individual setters via #[builder(setter)].

The types of skipped fields must implement Default.

#[derive(Builder)]
struct HiddenField {
    setter_present: u32,
    #[builder(setter(skip))]
    setter_skipped: u32,
}

Alternatively, you can use the more verbose form:

  • #[builder(setter(skip = true))]
  • #[builder(setter(skip = false))]

Custom setters (skip autogenerated setters)

Similarly to setter(skip), you can say that you will provide your own setter methods. This simply suppresses the generation of the setter, leaveing the field in the builder, as Option<T>.

#[derive(Builder)]
struct SetterOptOut {
    #[builder(setter(custom))]
    custom_setter: u32,
}
impl SetterOptOutBuilder {
    fn custom_setter(&mut self, value: u32) {
        self.custom_setter = Some(value);
    }
}

Again, the more verbose form is accepted:

  • #[builder(setter(custom = true))]
  • #[builder(setter(custom = false))]

Setter Visibility

Setters are public by default. You can precede your struct (or field) with #[builder(public)] to make this explicit.

Otherwise precede your struct (or field) with #[builder(private)] to opt into private setters.

Generated builder struct name

By default, the builder struct for struct Foo is FooBuilder. You can override this:

#[derive(Builder)]
#[builder(name = "FooConstructor")]
struct Foo { }

let foo: Foo = FooConstructor::default().build()?;

Setter Name/Prefix

Setter methods are named after their corresponding field by default.

  • You can customize the setter name via #[builder(setter(name = "foo")).
  • Alternatively you can set a prefix via #[builder(setter(prefix = "xyz")), which will change the method name to xyz_foo if the field is named foo. Note that an underscore is inserted, since Rust favors snake case here.

Prefixes can also be defined on the struct level, but renames only work on fields. Renames take precedence over prefix definitions.

Generic Setters

You can make each setter generic over the Into-trait. It’s as simple as adding #[builder(setter(into))] to either a field or the whole struct.

#[derive(Builder, Debug, PartialEq)]
struct Lorem {
    #[builder(setter(into))]
    pub ipsum: String,
}

fn main() {
    // `"foo"` will be converted into a `String` automatically.
    let x = LoremBuilder::default().ipsum("foo").build().unwrap();

    assert_eq!(x, Lorem {
        ipsum: "foo".to_string(),
    });
}

Setters for Option

You can avoid to user to wrap value into Some(...) for field of type Option<T>. It’s as simple as adding #[builder(setter(strip_option))] to either a field or the whole struct.

#[derive(Builder, Debug, PartialEq)]
struct Lorem {
    #[builder(setter(into, strip_option))]
    pub ipsum: Option<String>,
    #[builder(setter(into, strip_option), default)]
    pub foo: Option<String>,
}

fn main() {
    // `"foo"` will be converted into a `String` automatically.
    let x = LoremBuilder::default().ipsum("foo").build().unwrap();

    assert_eq!(x, Lorem {
        ipsum: Some("foo".to_string()),
        foo: None
    });
}

If you want to set the value to None when unset, then enable default on this field (or do not use strip_option).

Limitation: only the Option type name is supported, not type alias nor std::option::Option.

Fallible Setters

Alongside the normal setter methods, you can expose fallible setters which are generic over the TryInto trait. TryInto is a not-yet-stable trait (see rust-lang issue #33417) similar to Into with the key distinction that the conversion can fail, and therefore produces a Result.

You can only declare the try_setter attribute today if you’re targeting nightly, and you have to add #![feature(try_from)] to your crate to use it.

#[derive(Builder, Debug, PartialEq)]
#[builder(try_setter, setter(into))]
struct Lorem {
    pub name: String,
    pub ipsum: u8,
}

#[derive(Builder, Debug, PartialEq)]
struct Ipsum {
    #[builder(try_setter, setter(into, name = "foo"))]
    pub dolor: u8,
}

fn main() {
   LoremBuilder::default()
       .try_ipsum(1u16).unwrap()
       .name("hello")
       .build()
       .expect("1 fits into a u8");

   IpsumBuilder::default()
       .try_foo(1u16).unwrap()
       .build()
       .expect("1 fits into a u8");
}

Default Values

You can define default values for each field via annotation by #[builder(default = "...")], where ... stands for any Rust expression and must be string-escaped, e.g.

  • #[builder(default = "42")]
  • #[builder(default)] delegates to the Default trait of the base type.

The expression will be evaluated with each call to build.

#[derive(Builder, Debug, PartialEq)]
struct Lorem {
    #[builder(default = "42")]
    pub ipsum: u32,
}

fn main() {
    // If we don't set the field `ipsum`,
    let x = LoremBuilder::default().build().unwrap();

    // .. the custom default will be used for `ipsum`:
    assert_eq!(x, Lorem {
        ipsum: 42,
    });
}

Tips on Defaults

  • The #[builder(default)] annotation can be used on the struct level, too. Overrides are still possible.
  • Delegate to a private helper method on FooBuilder for anything fancy. This way you will get much better error diagnostics from the rust compiler and it will be much more readable for other human beings. :-)
  • Defaults will not work while using #[builder(build_fn(skip))]. In this case, you’ll need to handle default values yourself when converting from the builder, such as by using .unwrap_or() and .unwrap_or_else().
struct Lorem {
    ipsum: String,
    // Custom defaults can delegate to helper methods
    // and pass errors to the enclosing `build()` method via `?`.
    #[builder(default = "self.default_dolor()?")]
    dolor: String,
}

impl LoremBuilder {
    // Private helper method with access to the builder struct.
    fn default_dolor(&self) -> Result<String, String> {
        match self.ipsum {
            Some(ref x) if x.chars().count() > 3 => Ok(format!("dolor {}", x)),
            _ => Err("ipsum must at least 3 chars to build dolor".to_string()),
        }
    }
}

You can even reference other fields, but you have to remember that the builder struct will wrap every type in an Option (as illustrated earlier).

Generic Structs

#[derive(Builder, Debug, PartialEq, Default, Clone)]
struct GenLorem<T: Clone> {
    ipsum: &'static str,
    dolor: T,
}

fn main() {
    let x = GenLoremBuilder::default().ipsum("sit").dolor(42).build().unwrap();
    assert_eq!(x, GenLorem { ipsum: "sit".into(), dolor: 42 });
}

Build Method Customization

You can rename or suppress the auto-generated build method, leaving you free to implement your own version. Suppression is done using #[builder(build_fn(skip))] at the struct level, and renaming is done with #[builder(build_fn(name = "YOUR_NAME"))].

Pre-Build Validation

If you’re using the provided build method, you can declare #[builder(build_fn(validate = "path::to::fn"))] to specify a validator function which gets access to the builder before construction. The path does not need to be fully-qualified, and will consider use statements made at module level. It must be accessible from the scope where the target struct is declared.

The provided function must have the signature (&FooBuilder) -> Result<_, String>; the Ok variant is not used by the build method.

#[derive(Builder, Debug, PartialEq)]
#[builder(build_fn(validate = "Self::validate"))]
struct Lorem {
    pub ipsum: u8,
}

impl LoremBuilder {
    /// Check that `Lorem` is putting in the right amount of effort.
    fn validate(&self) -> Result<(), String> {
        if let Some(ref ipsum) = self.ipsum {
            match *ipsum {
                i if i < 20 => Err("Try harder".to_string()),
                i if i > 100 => Err("You'll tire yourself out".to_string()),
                _ => Ok(())
            }
        } else {
            Ok(())
        }
    }
}

fn main() {
    // If we're trying too hard...
    let x = LoremBuilder::default().ipsum(120).build().unwrap_err();

    // .. the build will fail:
    assert_eq!(&x.to_string(), "You'll tire yourself out");
}

Note:

  • Default values are applied after validation, and will therefore not be validated!

Additional Trait Derivations

You can derive additional traits on the builder, including traits defined by other crates:

#[derive(Builder, Clone)]
#[builder(derive(Debug, PartialEq, Eq))]
pub struct Lorem {
    foo: u8,
    bar: String,
}

fn main() {
   assert_eq!(LoremBuilder::default(), LoremBuilder::default());
}

Attributes declared for those traits are not forwarded to the fields on the builder.

Documentation Comments and Attributes

#[derive(Builder)] copies doc comments and attributes (#[...]) from your fields to the according builder fields and setter-methods, if it is one of the following:

  • /// ...
  • #[doc = ...]
  • #[cfg(...)]
  • #[allow(...)]

The whitelisting minimizes interference with other custom attributes like those used by Serde, Diesel, or others.

#[derive(Builder)]
struct Lorem {
    /// `ipsum` may be any `String` (be creative).
    ipsum: String,
    #[doc = r"`dolor` is the estimated amount of work."]
    dolor: i32,
    // `#[derive(Builder)]` understands conditional compilation via cfg-attributes,
    // i.e. => "no field = no setter".
    #[cfg(target_os = "macos")]
    #[allow(non_snake_case)]
    Im_a_Mac: bool,
}

Pass-through Attributes

You can set attributes on elements of the builder using the builder_*_attr attributes:

  • builder_struct_attr adds attributes after #[derive(...)] on the builder struct.
  • builder_impl_attr adds attributes on the impl block
  • builder_field_attr adds attributes to field declarations in the builder struct.
  • builder_setter_attr adds attributes to the setter in the impl block.
#[derive(Builder)]
#[builder(derive(serde::Serialize))]
#[builder_struct_attr(serde(rename_all = "camelCase"))]
struct Lorem {
    #[builder_field_attr(serde(rename="dolor"))]
    ipsum: String,
}

let mut show = LoremBuilder::default();
show.ipsum("sit".into());
assert_eq!(serde_json::to_string(&show).unwrap(), r#"{"dolor":"sit"}"#);

Error return type from autogenerated build function

By default, build returns an autogenerated error type:

#[doc="Error type for LoremBuilder"]
#[derive(Debug)]
#[non_exhaustive]
pub enum LoremBuilderError { // where `LoremBuilder` is the name of the builder struct
    /// Uninitialized field
    UninitializedField(&'static str),
    /// Custom validation error
    ValidationError(String),
}

impl From<String> for LoremBuilderError {
    fn from(s: String) -> Self { Self::ValidationError(s) }
}
impl From<UninitializedFieldError> for LoremBuilderError { // ...
impl Display for LoremBuilderError { // ...
impl std::error::Error for LoremBuilderError {}

Alternatively, you can specify your own error type:

#[derive(Builder, Debug, PartialEq)]
#[builder(build_fn(error = "OurLoremError"))]
struct Lorem {
    pub ipsum: u32,
}

struct OurLoremError(String);

impl From<UninitializedFieldError> for OurLoremError {
    fn from(ufe: UninitializedFieldError) -> OurLoremError { OurLoremError(ufe.to_string()) }
}

let err: OurLoremError = LoremBuilder::default().build().unwrap_err();
assert_eq!(&err.0, "Field not initialized: ipsum");

Completely custom fields in the builder

Instead of having an Option, you can have whatever type you like:

#[derive(Debug, PartialEq, Default, Builder, Clone)]
#[builder(derive(Debug, PartialEq))]
struct Lorem {
    #[builder(setter(into), field(type = "u32"))]
    ipsum: u32,

    #[builder(field(type = "String", build = "()"))]
    dolor: (),

    #[builder(field(type = "&'static str", build = "self.amet.parse()?"))]
    amet: u32,
}

impl From<std::num::ParseIntError> for LoremBuilderError { // ...

let mut builder = LoremBuilder::default();
builder.ipsum(42u16).dolor("sit".into()).amet("12");
assert_eq!(builder, LoremBuilder { ipsum: 42, dolor: "sit".into(), amet: "12" });
let lorem = builder.build().unwrap();
assert_eq!(lorem, Lorem { ipsum: 42, dolor: (), amet: 12 });

The builder field type (type =) must implement Default.

The argument to build must be a literal string containing Rust code for the contents of a block, which must evaluate to the type of the target field. It may refer to the builder struct as self, use ?, etc.

#![no_std] Support (on Nightly)

You can activate support for #![no_std] by adding #[builder(no_std)] to your struct and #![feature(alloc)] extern crate alloc to your crate.

The latter requires the nightly toolchain.

Troubleshooting

Gotchas

  • Tuple structs and unit structs are not supported as they have no field names.
  • Generic setters introduce a type parameter VALUE: Into<_>. Therefore you can’t use VALUE as a type parameter on a generic struct in combination with generic setters.
  • The try_setter attribute and owned builder pattern are not compatible in practice; an error during building will consume the builder, making it impossible to continue construction.
  • When re-exporting the underlying struct under a different name, the auto-generated documentation will not match.
  • If derive_builder depends on your crate, and vice versa, then a cyclic dependency would occur. To break it you could try to depend on the derive_builder_core crate instead.

Report Issues and Ideas

Open an issue on GitHub

If possible please try to provide the debugging info if you experience unexpected compilation errors (see above).

Structs

Runtime error when a build() method is called and one or more required fields do not have a value.