How I Use Declarative Macros in Rust

Macros are one of my favorite features of Rust. I use both declarative and procedural macros heavily in my projects. Coming from C/C++, I never really liked metaprogramming, but Rust makes it great.

In this post, I'll show common use cases of mine for writing declarative macros, and I don't mean just writing dbg! all over the place. I'll write about procedural macros in another post.

Quick intro

This is just a quick intro to macros. I recommend reading Macros chapter from Rust book and The Little Book of Rust Macros.

Declarative macros are rules that match against Rust syntax and produce Rust code. They are declared with macro_rules! macro_name {...} construct, and used as macro_name!(...).

Macro rules consist of patterns or matchers, and templates. Matchers define what Rust fragments to match. They are written as $name:expr => ..., where $name is a "metavariable" and expr is a fragment specifier.

For example, the macro below would bind any Rust expression to the metavariable $name.

macro_rules! say_hello {  ($name:expr) => {      println!("Hello, {}!", $name);  };}say_hello!("World");say_hello!({ 13 + 29 });

There are various fragment specifiers, but I mostly use ident, expr, ty/path and tt.

Repetitions define repeated patterns. You can define them with $(...), followed by a separator and an operator, which is * for any number of repetitions, + for at least one, and ? for an optional fragment. Metavariables inside repetitions are matched against the corresponding fragment in the input. They can be expanded inside the template with $(...)*. For example, the macro bellow matches any number of key-value pairs in its last matcher.

macro_rules! hash_map {  () => {    ::std::collections::HashMap::new()  };  ($capacity:expr) => {    ::std::collections::HashMap::with_capacity($capacity)  };  ($($key:expr => $value:expr),* $(,)?) => {{    let mut _map = hash_map!(count_repeating!($($key),*));    $(      _map.insert($key, $value);    )*    _map  }};}

This macro can be used to easily create a HashMap. It's from my utility library called Bomboni. In case you need to count the number of repetitions, you can look here.

One common pattern is handling trailing comma, because having a trailing comma at the end of a long list makes code prettier after formatting. This is done by $(,) at the end of the pattern, which will match one or zero commas.

Implementing traits

For me, the most common use case for declarative macros is implementing traits. If you ever looked at Rust's std library sources, you can see how they use macros for repetitive implementations (an example is int_impl! macro).

I typically write macros when implementing From and TryFrom traits for structs that represent data-transfer objects or API request/response types. For example, checking if fields exist, converting from Options, constructing errors based on field names, handling visibility, etc.

Here's an Id type that implements From for all integer types.

struct Id(u128);macro_rules! impl_from {  ( $( $source:ty ),* $(,)? ) => {    $(impl From<$source> for Id {      fn from(x: $source) -> Self {        Id::new(x as u128)      }    })*  };}impl_from!(i8, i16, i32, i64, i128, u8, u16, u32, u64);

This would require 9 Froms to be written by hand.

Here's an example of implementing serde traits for Google Protobuf's Well-Known Types. Another example could be serializing simple types as strings, where you utilize Display/FromStr and combine it with serde's Serialize and Deserialize.

macro_rules! impl_value_serde {  ($type:ty, $as:ty) => {    impl Serialize for $type {      fn serialize<S>(        &self,        serializer: S,      ) -> Result<<S as Serializer>::Ok, <S as Serializer>::Error>      where        S: Serializer,      {        <$as>::serialize(&self.value, serializer)      }    }    impl<'de> Deserialize<'de> for $type {      fn deserialize<D>(deserializer: D) -> Result<Self, <D as Deserializer<'de>>::Error>      where        D: Deserializer<'de>,      {        let value = <$as>::deserialize(deserializer)?;        Ok(value.into())      }    }  };}impl_value_serde!(DoubleValue, f64);impl_value_serde!(FloatValue, f32);impl_value_serde!(Int32Value, i32);impl_value_serde!(UInt32Value, u32);impl_value_serde!(BoolValue, bool);impl_value_serde!(StringValue, String);

The following trait is used to parse some API-specific "data-transfer" structs.

macro_rules! impl_data_conversion {  ($variant:ident, $type:ty) => {    impl ProtoParse<ParsedValue> for $type {      type Error = GraphError;      fn parse(x: ParsedValue) -> Result<Self, Self::Error> {        if let ParsedValue::Data(x) = x {          Ok(x.parse_into()?)        } else {          Err(GraphError::value_display(            &x.to_string(),            GraphErrorReason::ValueInvalidNumericData,          )          .with_data_type_kind(DataTypeKind::$variant))        }      }    }  };}impl_data_conversion!(String, String);impl_data_conversion!(Boolean, bool);impl_data_conversion!(Float32, f32);// ...

This conversion can return an error object, enriched by value and its type. For a project I'm working on, this is quite common. Writing this code by hand for many types would be annoying.

All of these macros are fairly simple, but they save a ton of time.

Repetitive code blocks

There are situations where I have long match expressions. The match arms are very simple, but there are many of them. Or, there are small procedures to be done, but I don't think writing a separate function or a closure would be appropriate.

One macro I really like is matches!. It's basically a single-arm match expression, but you can readably use it in if statements.

So, this:

match data_type.id {  ParsedDataTypeId::Standard(ref data_type_id) if data_type_id == "data-trait-name" => {    // ...  }  _ => {}}

Can be written as:

if matches!(data_type.id, ParsedDataTypeId::Standard(ref data_type_id)            if data_type_id == "data-trait-name") {  // ...}

I use it to implement utility "is" functions for enums.

pub fn is_standard(&self) -> bool {  matches!(self, ParsedDataTypeId::Standard(_))}

Here, I use a macro to write non-None fields to a buffer. The stringify! macro from Rust's standard library can print Rust's tokens into &'static str. In this example, I'm using it to get field names as strings.

macro_rules! impl_write_field {  ($ident:ident) => {    if let Some(value) = self.$ident.as_ref() {      if empty {        empty = false;      } else {        buf.write_str(", ");      }      buf.write_str(&format!("{}={}", stringify!($ident), value))?;    }  };}impl_write_field!(data_trait_id);impl_write_field!(data_type_id);impl_write_field!(node_instance_id);impl_write_field!(edge);// ...

Here, I'm extending an error status struct with builder pattern-like functions for each metadata field For a field named setting_id, it would give me with_setting_id and setting_id setter functions.

macro_rules! impl_platform_metadata_field {($ident:ident, $with_name:ident, $type:ty, into) => {    pub fn $ident<R: Into<PlatformErrorReason>>($ident: $type, reason: R) -> Self {      Self::new_with_metadata(        reason,        PlatformErrorMetadata {          $ident: Some($ident.into()),          ..Default::default()        },      )    }    pub fn $with_name(mut self, $ident: $type) -> Self {      self.modify_metadata(|metadata| {        metadata.$ident = Some($ident.into());      })    }};// .../*  err.with_reason(GraphErrorReason::DataTypeInvalidFieldValueType)    .with_field_id(&field.id)*/

Because of hygiene rules of Rust macros, we are somewhat limited in what we can do with declarative macros. You wouldn't be able to create a new identifier such as with_$name. With procedural macros, you can output any Rust code you want. But, if you don't want to write procedural macros, you can use dtolnay/paste crate, which allows you to concatenate together metavariables. There's a concat! macro, but it can only concatenate literals into a string.

Testing

Macros are very handy for testing. You can write your own assert statements or test cases.

Here's a macro for asserting correctness of Handlebars templating engine's rendering output.

let r = get_handlebars_registry();macro_rules! assert_case {  ($case:expr, $source:expr, $expected:expr $(,)?) => {    assert_eq!(      r.render_template(        &format!(r#"{{{{{} "{}" }}}}"#, $case, $source),        &Value::Null      )      .unwrap()      .as_str(),      $expected    );  };}macro_rules! assert_expr {  ($expr:expr, $result:expr) => {    assert_eq!(r.render_template($expr, &Value::Null).unwrap(), $result);  };}assert_case!(UPPER_CASE_HELPER_NAME, "variable name", "VARIABLE NAME");assert_case!(PASCAL_CASE_HELPER_NAME, "variable name", "VariableName");assert_expr!("{{add 1 2}}", "3.0");assert_expr!("{{sqrt 2}}", "1.4142135623730951");

Calling assert_eq! with render_template and unwrapping the result for each test case would be too verbose. It's possible that an inline array of test cases could be cleaner, but it depends. In a project of mine, there's a fairly long macro I have declared in the tests module, and I use it in several test functions. It's quite readable. For debugging, you might find that writing a macro is better than a function, especially if you use utilities like line!, file!, stringify!, etc.

In conclusion, declarative macros are a great tool for writing otherwise repetitive code. Sometimes expanding a macro can be helpful for understanding what's going on. If I'm writing a complicated macro, I write matchers case by case in a test function, and then expand them to see if it matches correctly.