Askama

Askama implements a template rendering engine based on Jinja. It generates Rust code from your templates at compile time based on a user-defined struct to hold the template's context. See below for an example.

All feedback welcome! Feel free to file bugs, requests for documentation and any other feedback to the issue tracker.

Have a look at our Askama Playground, if you want to try out askama's code generation online.

Feature highlights

Construct templates using a familiar, easy-to-use syntax
Benefit from the safety provided by Rust's type system
Template code is compiled into your crate for optimal performance
Debugging features to assist you in template development
Templates must be valid UTF-8 and produce UTF-8 when rendered
Works on stable Rust

Supported in templates

Template inheritance
Loops, if/else statements and include support
Macro support
Variables (no mutability allowed)
Many built-in filters, and the ability to use your own
Whitespace suppressing with '-' markers
Opt-out HTML escaping
Syntax customization

Getting Started

First, add the following to your crate's Cargo.toml:

# in [dependencies] section
askama = "0.14.0"

Now create a directory called templates in your crate root. In it, create a file called hello.html, containing the following:

Hello, {{ name }}!

In any Rust file inside your crate, add the following:

use askama::Template; // bring trait in scope

#[derive(Template)] // this will generate the code...
#[template(path = "hello.html")] // using the template in this path, relative
                                 // to the `templates` dir in the crate root
struct HelloTemplate<'a> { // the name of the struct can be anything
    name: &'a str, // the field name should match the variable name
                   // in your template
}

fn main() {
    let hello = HelloTemplate { name: "world" }; // instantiate your struct
    println!("{}", hello.render().unwrap()); // then render it.
}

You should now be able to compile and run this code.

Creating Templates

An Askama template is a struct definition which provides the template context combined with a UTF-8 encoded text file (or inline source, see below). Askama can be used to generate any kind of text-based format. The template file's extension may be used to provide content type hints.

A template consists of text contents, which are passed through as-is, expressions, which get replaced with content while being rendered, and tags, which control the template's logic. The template syntax is very similar to Jinja, as well as Jinja-derivatives like Twig or Tera.

#[derive(Template)] // this will generate the code...
#[template(path = "hello.html")] // using the template in this path, relative
                                 // to the `templates` dir in the crate root
struct HelloTemplate<'a> { // the name of the struct can be anything
    name: &'a str, // the field name should match the variable name
                   // in your template
}

The `template()` attribute

Askama works by generating one or more trait implementations for any struct type decorated with the #[derive(Template)] attribute. The code generation process takes some options that can be specified through the template() attribute. The following sub-attributes are currently recognized:

path (e.g. path = "foo.html"): sets the path to the template file. The path is interpreted as relative to the configured template directories (by default, this is a templates directory next to your Cargo.toml). The file name extension is used to infer an escape mode (see below). In web framework integrations, the path's extension may also be used to infer the content type of the resulting response. Cannot be used together with source.
```
#[derive(Template)]
#[template(path = "hello.html")]
struct HelloTemplate<'a> { ... }
```
source (e.g. source = "{{ foo }}"): directly sets the template source. This can be useful for test cases or short templates. The generated path is undefined, which generally makes it impossible to refer to this template from other templates. If source is specified, ext must also be specified (see below). Cannot be used together with path.
```
#[derive(Template)]
#[template(source = "Hello {{ name }}")]
struct HelloTemplate<'a> {
    name: &'a str,
}
```
in_doc (e.g. in_doc = true): please see the section "documentation as template code".
ext (e.g. ext = "txt"): lets you specify the content type as a file extension. This is used to infer an escape mode (see below), and some web framework integrations use it to determine the content type. Cannot be used together with path.
```
#[derive(Template)]
#[template(source = "Hello {{ name }}", ext = "txt")]
struct HelloTemplate<'a> {
    name: &'a str,
}
```
print (e.g. print = "code"): enable debugging by printing nothing (none), the parsed syntax tree (ast), the generated code (code) or all for both. The requested data will be printed to stdout at compile time.
```
#[derive(Template)]
#[template(path = "hello.html", print = "all")]
struct HelloTemplate<'a> { ... }
```
block (e.g. block = "block_name"): renders the block by itself. Expressions outside of the block are not required by the struct, and inheritance is also supported. This can be useful when you need to decompose your template for partial rendering, without needing to extract the partial into a separate template or macro.
```
#[derive(Template)]
#[template(path = "hello.html", block = "hello")]
struct HelloTemplate<'a> { ... }
```

blocks (e.g. blocks = ["title", "content"]): automatically generates (a number of) sub-templates that act as if they had a block = "..." attribute. You can access the sub-templates with the method my_template.as_block_name(), where block_name is the name of the block:

#[derive(Template)]
#[template(
    ext = "txt",
    source = "
        {% block title %} ... {% endblock %}
        {% block content %} ... {% endblock %}
    ",
    blocks = ["title", "content"]
)]
struct News<'a> {
    title: &'a str,
    message: &'a str,
}

let news = News {
    title: "Announcing Rust 1.84.1",
    message: "The Rust team has published a new point release of Rust, 1.84.1.",
};
assert_eq!(
    news.as_title().render().unwrap(),
    "<h1>Announcing Rust 1.84.1</h1>"
);

escape (e.g. escape = "none"): override the template's extension used for the purpose of determining the escaper for this template. See the section on configuring custom escapers for more information.
```
#[derive(Template)]
#[template(path = "hello.html", escape = "none")]
struct HelloTemplate<'a> { ... }
```
syntax (e.g. syntax = "foo"): set the syntax name for a parser defined in the configuration file. The default syntax , "default", is the one provided by Askama.
```
#[derive(Template)]
#[template(path = "hello.html", syntax = "foo")]
struct HelloTemplate<'a> { ... }
```
config (e.g. config = "config_file_path"): set the path for the config file to be used. The path is interpreted as relative to your crate root.
```
#[derive(Template)]
#[template(path = "hello.html", config = "config.toml")]
struct HelloTemplate<'a> { ... }
```

askama (e.g. askama = askama): If you are using askama in a subproject, a library or a macro, it might be necessary to specify the path where to find the module askama:

#[doc(hidden)]
use askama as __askama;

#[macro_export]
macro_rules! new_greeter {
    ($name:ident) => {
        #[derive(Debug, $crate::askama::Template)]
        #[template(
            ext = "txt",
            source = "Hello, world!",
            askama = $crate::__askama
        )]
        struct $name;
    }
}

new_greeter!(HelloWorld);
assert_eq!(HelloWorld.to_string(), Ok("Hello, world."));

Templating `enum`s

You can add derive Templates for structs and enums. If you add #[template()] only to the item itself, both item kinds work exactly the same. But with enums you also have the option to add a specialized implementation to one, some, or all variants:

#[derive(Debug, Template)]
#[template(path = "area.txt")]
enum Area {
    Square(f32),
    Rectangle { a: f32, b: f32 },
    Circle { radius: f32 },
}

{%- match self -%}
    {%- when Self::Square(side) -%}
        {{side}}^2
    {%- when Self::Rectangle { a, b} -%}
        {{a}} * {{b}}
    {%- when Self::Circle { radius } -%}
        pi * {{radius}}^2
{%- endmatch -%}

will give you the same results as:

#[derive(Template, Debug)]
#[template(ext = "txt")]
enum AreaPerVariant {
    #[template(source = "{{self.0}}^2")]
    Square(f32),
    #[template(source = "{{a}} * {{b}}")]
    Rectangle { a: f32, b: f32 },
    #[template(source = "pi * {{radius}}^2")]
    Circle { radius: f32 },
}

As you can see with the ext attribute, enum variants inherit most settings of the enum: config, escape, ext, syntax, and whitespace. Not inherited are: block, and print.

If there is no #[template] annotation for an enum variant, then the enum needs a default implementation, which will be used if self is this variant. A good compromise between annotating only the template, or all its variants, might be using the block argument on the members:

#[derive(Template, Debug)]
#[template(path = "area.txt")]
enum AreaWithBlocks {
    #[template(block = "square")]
    Square(f32),
    #[template(block = "rectangle")]
    Rectangle { a: f32, b: f32 },
    #[template(block = "circle")]
    Circle { radius: f32 },
}

{%- block square -%}
    {{self.0}}^2
{%- endblock -%}

{%- block rectangle -%}
    {{a}} * {{b}}
{%- endblock -%}

{%- block circle -%}
    pi * {{radius}}^2
{%- endblock -%}

Documentation as template code

As an alternative to supplying the code template code in an external file (e.g. path argument), or as a string (e.g. source argument), you can also enable the "code-in-doc" feature. With this feature, you can specify the template code directly in the documentation of the template item.

Instead of path = "…" or source = "…", specify in_doc = true in the #[template] attribute, and in the item's documentation, add a code block with the askama attribute:

/// Here you can put our usual comments.
///
/// ```askama
/// <div>{{ lines|linebreaksbr }}</div>
/// ```
///
/// Any usual docs, including tests can be put in here, too:
///
/// ```rust
/// assert_eq!(
///     Example { lines: "a\nb\nc" }.to_string(),
///     "<div>a<br/>b<br/>c</div>"
/// );
/// ```
///
/// All comments are still optional, though.
#[derive(Template)]
#[template(ext = "html", in_doc = true)]
struct Example<'a> {
    lines: &'a str,
}

If you want to supply the template code in the comments, then you have to specify the ext argument, too, e.g. #[template(ext = "html")].

Instead of askama, you can also write jinja or jinja2, e.g. to get it to work better in conjunction with syntax highlighters.

Runtime values

It is possible to define variables at runtime and to use them in the templates using the values filter or the askama::get_value function and to call the _with_values variants of the render methods. It expects an extra argument implementing the Values trait. This trait is implemented on a few types provided by the std, like HashMap:

use std::collections::HashMap;

let mut values: HashMap<&str, Box<dyn Any>> = HashMap::new();
// We add a new value named "name" with the value "Bibop".
values.insert("name", Box::new("Bibop"));
values.insert("age", Box::new(12u32));

The Values trait is expecting types storing data with the Any trait, allowing to store any type.

Then to render with these values:

template_struct.render_with_values(&values).unwrap();

There are two ways to get the values from the template, either by using the value filter or by calling directly the askama::get_value function:

{% if let Ok(name) = "name"|value::<&str> %}
  name is {{ name }}
{% endif %}
{% if let Ok(age) = askama::get_value::<u32>("age") %}
  age is {{ age }}
{% endif %}

If you try to retrieve a value with the wrong type or that you didn't set, you will get an Err(askama::Error::ValueType) or a Err(askama::Error::ValueMissing).

Another example with a key-value tuple:

let value = "value".to_string();
let tuple: (&str, &dyn Any) = ("a", &value);

template_struct.render_with_values(&tuple).unwrap();

With this setup, only the "a" key will return a value:

{% if let Ok(name) = "a"|value::<String> %}
  a is {{ a }}
{% endif %}

Debugging and Troubleshooting

You can view the parse tree for a template as well as the generated code by changing the template attribute item list for the template struct:

#[derive(Template)]
#[template(path = "hello.html", print = "all")]
struct HelloTemplate<'a> { ... }

The print key can take one of four values:

none (the default value)
ast (print the parse tree)
code (print the generated code)
all (print both parse tree and code)

The resulting output will be printed to stderr during the compilation process.

The parse tree looks like this for the example template:

[Lit("", "Hello,", " "), Expr(WS(false, false), Var("name")), Lit("", "!", "\n")]

The generated code looks like this:

impl<'a> askama::Template for HelloWorld<'a> {
    fn render_into<AskamaW>(&self, __askama_writer: &mut AskamaW) -> askama::Result<()>
    where
        AskamaW: core::fmt::Write + ?Sized,
    {
        __askama_writer.write_str("Hello, ")?;
        match (
            &((&&askama::filters::AutoEscaper::new(
                &(self.name),
                askama::filters::Html,
            ))
                .askama_auto_escape()?),
        ) {
            (expr2,) => {
                (&&askama::filters::Writable(expr2)).askama_write(__askama_writer)?;
            }
        }
        __askama_writer.write_str("!")?;
        Ok(())
    }
    const SIZE_HINT: usize = 11usize;
}

Configuration

At compile time, Askama will read optional configuration values from askama.toml in the crate root (the directory where Cargo.toml can be found). Currently, this covers the directories to search for templates, custom syntax configuration and escaper configuration.

This example file demonstrates the default configuration:

[general]
# Directories to search for templates, relative to the crate root.
dirs = ["templates"]
# Unless you add a `-` in a block, whitespace characters won't be trimmed.
whitespace = "preserve"

Whitespace control

In the default configuration, you can use the - operator to indicate that whitespace should be suppressed before or after a block. For example:

<div>


{%- if something %}
Hello
{% endif %}

In the template above, only the whitespace between <div> and {%- will be suppressed. If you set whitespace to "suppress", whitespace characters before and after each block will be suppressed by default. To preserve the whitespace characters, you can use the + operator:

{% if something +%}
Hello
{%+ endif %}

In this example, Hello will be surrounded with newline characters.

There is a third possibility: in case you want to suppress all whitespace characters except one, you can use ~:

{% if something ~%}
Hello
{%~ endif %}

To be noted, if one of the trimmed characters is a newline, then the only character remaining will be a newline.

If you want this to be the default behavior, you can set whitespace to "minimize".

To be noted: you can also configure whitespace directly into the template derive proc macro:

#[derive(Template)]
#[template(whitespace = "suppress")]
pub struct SomeTemplate;

If you configure whitespace directly into the template derive proc-macro, it will take precedence over the one in your configuration file. So in this case, if you already set whitespace = "minimize" into your configuration file, it will be replaced by suppress for this template.

Custom syntaxes

Here is an example that defines two custom syntaxes:

[general]
default_syntax = "foo"

[[syntax]]
name = "foo"
block_start = "%{"
comment_start = "#{"
expr_end = "^^"

[[syntax]]
name = "bar"
block_start = "%%"
block_end = "%%"
comment_start = "%#"
expr_start = "%{"

A syntax block consists of at least the attribute name which uniquely names this syntax in the project.

The following keys can currently be used to customize template syntax:

block_start, defaults to {%
block_end, defaults to %}
comment_start, defaults to {#
comment_end, defaults to #}
expr_start, defaults to {{
expr_end, defaults to }}

Values must be at least two characters long. If a key is omitted, the value from the default syntax is used.

Escapers

Here is an example of a custom escaper:

[[escaper]]
path = "::tex_escape::Tex"
extensions = ["tex"]

An escaper block consists of the attributes path and extensions. path contains a Rust identifier that must be in scope for templates using this escaper. This type must implement the Escaper trait.

extensions defines a list of file extensions that will trigger the use of that escaper. Extensions are matched in order, starting with the first escaper configured and ending with the default escapers for HTML (extensions html, htm, xml, j2, jinja, jinja2) and plain text (no escaping; md, yml, none, txt, and the empty string). Note that this means you can also define other escapers that match different extensions to the same escaper.

You can then use templates with this extension or use the escape filter with the name of your extension in your template:

{{ some_string|escape("tex") }}

As an example, we want .js files to be treated like "txt" files. To do so:

[[escaper]]
path = "askama::filters::Text"
extensions = ["js"]

Text implements the Escaper trait so since we don't need want any escaping on our .js files, we use it.

Template Syntax

Variables

Top-level template variables are defined by the template's context type. You can use a dot (.) to access variable's attributes or methods. Reading from variables is subject to the usual borrowing policies. For example, {{ name }} will get the name field from the template context, while {{ user.name }} will get the name field of the user field from the template context.

Using constants in templates

You can use constants defined in your Rust code. For example if you have:

pub const MAX_NB_USERS: usize = 2;

defined in your crate root, you can then use it in your templates by using crate::MAX_NB_USERS:

<p>The user limit is {{ crate::MAX_NB_USERS }}.</p>
{% set value = 4 %}
{% if value > crate::MAX_NB_USERS %}
    <p>{{ value }} is bigger than MAX_NB_USERS.</p>
{% else %}
    <p>{{ value }} is less than MAX_NB_USERS.</p>
{% endif %}

Assignments

Inside code blocks, you can also declare variables or assign values to variables. Assignments can't be imported by other templates.

Assignments use the let tag:

{% let name = user.name %}
{% let len = name.len() %}

{% let val -%}
{% if len == 0 -%}
  {% let val = "foo" -%}
{% else -%}
  {% let val = name -%}
{% endif -%}
{{ val }}

Like Rust, Askama also supports shadowing variables.

{% let foo = "bar" %}
{{ foo }}

{% let foo = "baz" %}
{{ foo }}

For compatibility with Jinja, set can be used in place of let.

Borrow rules

In some cases, the value of a variable initialization will be put behind a reference to prevent changing ownership. The rules are as follows:

If the value is an expression of more than one element (like x + 2), it WILL NOT BE put behind a reference.
If the value is a variable defined in the templates, it WILL NOT BE put behind a reference.
If the value has a filter applied to it (x|capitalize), it WILL NOT BE put behind a reference.
If the value is a field (x.y), it WILL BE put behind a reference.
If the expression ends with a question mark (like x?), it WILL NOT BE put behind a reference.

Filters

Values such as those obtained from variables can be post-processed using filters. Filters are applied to values using the pipe symbol (|) and may have optional extra arguments in parentheses. Filters can be chained, in which case the output from one filter is passed to the next.

For example, {{ "{:?}"|format(name|escape) }} will escape HTML characters from the value obtained by accessing the name field, and print the resulting string as a Rust literal.

The built-in filters are documented as part of the filters documentation.

To define your own filters, simply have a module named filters in scope of the context deriving a Template impl. Note that in case of name collision, the built in filters take precedence.

Filter blocks

You can apply a filter on a whole block at once using filter blocks:

{% filter lower %}
    {{ t }} / HELLO / {{ u }}
{% endfilter %}

The lower filter will be applied on the whole content.

Just like filters, you can combine them:

{% filter lower|capitalize %}
    {{ t }} / HELLO / {{ u }}
{% endfilter %}

In this case, lower will be called and then capitalize will be called on what lower returned.

Whitespace control

Askama considers all tabs, spaces, newlines and carriage returns to be whitespace. By default, it preserves all whitespace in template code, except that a single trailing newline character is suppressed. However, whitespace before and after expression and block delimiters can be suppressed by writing a minus sign directly following a start delimiter or leading into an end delimiter.

Here is an example:

{% if foo %}
  {{- bar -}}
{% else if -%}
  nothing
{%- endif %}

This discards all whitespace inside the if/else block. If a literal (any part of the template not surrounded by {% %} or {{ }}) includes only whitespace, whitespace suppression on either side will completely suppress that literal content.

If the whitespace default control is set to "suppress" and you want to preserve whitespace characters on one side of a block or of an expression, you need to use +. Example:

<a href="/" {#+ #}
   class="something">text</a>

In the above example, one whitespace character is kept between the href and the class attributes.

There is a third possibility. In case you want to suppress all whitespace characters except one ("minimize"), you can use ~:

{% if something ~%}
Hello
{%~ endif %}

To be noted, if one of the trimmed characters is a newline, then the only character remaining will be a newline.

Whitespace controls can also be defined by a configuration file or in the derive macro. These definitions follow the global-to-local preference:

Inline (-, +, ~)
Derive (#[template(whitespace = "suppress")])
Configuration (in askama.toml, whitespace = "preserve")

Two inline whitespace controls may point to the same whitespace span. In this case, they are resolved by the following preference.

Suppress (-)
Minimize (~)
Preserve (+)

Functions

There are several ways that functions can be called within templates, depending on where the function definition resides. These are:

Template struct fields
Static functions
Struct/Trait implementations

Template struct field

When the function is a field of the template struct, we can simply call it by invoking the name of the field, followed by parentheses containing any required arguments. For example, we can invoke the function foo for the following MyTemplate struct:

#[derive(Template)]
#[template(source = "{{ foo(123) }}", ext = "txt")]
struct MyTemplate {
  foo: fn(u32) -> String,
}

However, since we'll need to define this function every time we create an instance of MyTemplate, it's probably not the most ideal way to associate some behavior for our template.

Static functions

When a function exists within the same Rust module as the template definition, we can invoke it using the self path prefix, where self represents the scope of the module in which the template struct resides.

For example, here we call the function foo by writing self::foo(123) within the MyTemplate struct source:

fn foo(val: u32) -> String {
  format!("{}", val)
}

#[derive(Template)]
#[template(source = "{{ self::foo(123) }}", ext = "txt")]
struct MyTemplate;

This has the advantage of being able to share functionality across multiple templates, without needing to expose the function publicly outside of its module.

However, we are not limited to local functions defined within the same module. We can call any public function by specifying the full path to that function within the template source. For example, given a utilities module such as:

// src/templates/utils/mod.rs

pub fn foo(val: u32) -> String {
  format!("{}", val)
}

Within our MyTemplate source, we can call the foo function by writing:

// src/templates/my_template.rs

#[derive(Template)]
#[template(source = "{{ crate::templates::utils::foo(123) }}", ext = "txt")]
struct MyTemplate;

Struct / trait implementations

Finally, we can call methods of our template struct:

#[derive(Template)]
#[template(source = "{{ foo(123) }}", ext = "txt")]
struct MyTemplate {
  count: u32,
};

impl MyTemplate {
  fn foo(&self, val: u32) -> String {
    format!("{} is the count, {} is the value", self.count, val)
  }
}

You can also use self.foo(123), or even Self::foo(self, 123), as you see fit.

Similarly, using the Self path, we can also call any method belonging to a trait that has been implemented for our template struct:

trait Hello {
  fn greet(name: &str) -> String;
}

#[derive(Template)]
#[template(source = r#"{{ Self::greet("world") }}"#, ext = "txt")]
struct MyTemplate;

impl Hello for MyTemplate {
  fn greet(name: &str) -> String {
    format!("Hello {}", name)
  }
}

If you want to call a closure which is a field, you'll need to follow Rust's syntax by surrounding the call with parens:

#[derive(Template)]
#[template(source = "{{ (closure)(12) }}", ext = "txt")]
struct MyTemplate {
    closure: fn(i32) -> i32,
}

Calling functions

If you only provide a function name, askama will assume it's a method. If you want to call a method, you will need to use a path instead:

{# This is the equivalent of `self.method()`. #}
{{ method() }}
{# This is the equivalent of `self::function()`, which will call the
`function` function from the current module. #}
{{ self::function() }}
{# This is the equivalent of `super::b::f()`. #}
{{ super::b::f() }}

Template inheritance

Template inheritance allows you to build a base template with common elements that can be shared by all inheriting templates. A base template defines blocks that child templates can override.

Base template

<!DOCTYPE html>
<html lang="en">
  <head>
    <title>{% block title %}{{ title }} - My Site{% endblock %}</title>
    {% block head %}{% endblock %}
  </head>
  <body>
    <div id="content">
      {% block content %}<p>Placeholder content</p>{% endblock %}
    </div>
  </body>
</html>

The block tags define three blocks that can be filled in by child templates. The base template defines a default version of the block. A base template must define one or more blocks in order to enable inheritance. Blocks can only be specified at the top level of a template or inside other blocks, not inside if/else branches or in for-loop bodies.

It is also possible to use the name of the block in endblock (both in declaration and use):

{% block content %}<p>Placeholder content</p>{% endblock content %}

Child template

Here's an example child template:

{% extends "base.html" %}

{% block title %}Index{% endblock %}

{% block head %}
  <style>
  </style>
{% endblock %}

{% block content %}
  <h1>Index</h1>
  <p>Hello, world!</p>
  {% call super() %}
{% endblock %}

The extends tag tells the code generator that this template inherits from another template. It will search for the base template relative to itself before looking relative to the template base directory. It will render the top-level content from the base template, and substitute blocks from the base template with those from the child template. Inside a block in a child template, the super() macro can be called to render the parent block's contents.

Because top-level content from the child template is thus ignored, the extends tag doesn't support whitespace control:

{%- extends "base.html" +%}

The above code is rejected because we used - and +. For more information about whitespace control, take a look here.

Block fragments

Additionally, a block can be rendered by itself. This can be useful when you need to decompose your template for partial rendering, without needing to extract the partial into a separate template or macro. This can be done with the block parameter.

#[derive(Template)]
#[template(path = "...", block = "my_block")]
struct BlockFragment {
    name: String,
}

HTML escaping

Askama by default escapes variables if it thinks it is rendering HTML content. It infers the escaping context from the extension of template filenames, escaping by default if the extension is one of html, htm, or xml. When specifying a template as source in an attribute, the ext attribute parameter must be used to specify a type. Additionally, you can specify an escape mode explicitly for your template by setting the escape attribute parameter value (to none or html).

Askama escapes <, >, &, ", and ', according to the OWASP escaping recommendations. Use the safe filter to prevent escaping for a single expression, or the escape (or e) filter to escape a single expression in an unescaped context.

#[derive(Template)]
#[template(source = "{{strvar}}")]
struct TestTemplate {
    strvar: String,
}

fn main() {
    let s = TestTemplate {
        strvar: "// my <html> is \"unsafe\" & should be 'escaped'".to_string(),
    };
    assert_eq!(
        s.render().unwrap(),
        "&#x2f;&#x2f; my &lt;html&gt; is &quot;unsafe&quot; &amp; \
         should be &#x27;escaped&#x27;"
    );
}

Control structures

For

Loop over each item in an iterator. For example:

<h1>Users</h1>
<ul>
{% for user in users %}
  <li>{{ user.name|e }}</li>
{% endfor %}
</ul>

Inside for-loop blocks, some useful variables are accessible:

loop.index: current loop iteration (starting from 1)
loop.index0: current loop iteration (starting from 0)
loop.first: whether this is the first iteration of the loop
loop.last: whether this is the last iteration of the loop

<h1>Users</h1>
<ul>
{% for user in users %}
   {% if loop.first %}
   <li>First: {{user.name}}</li>
   {% else %}
   <li>User#{{loop.index}}: {{user.name}}</li>
   {% endif %}
{% endfor %}
</ul>

If

The if statement essentially mirrors Rust's if expression, and is used as you might expect:

{% if users.len() == 0 %}
  No users
{% else if users.len() == 1 %}
  1 user
{% elif users.len() == 2 %}
  2 users
{% else %}
  {{ users.len() }} users
{% endif %}

If Let

Additionally, if let statements are also supported and similarly mirror Rust's if let expressions:

{% if let Some(user) = user %}
  {{ user.name }}
{% else %}
  No user
{% endif %}

`is (not) defined`

You can use is (not) defined to ensure a variable exists (or not):

{% if x is defined %}
  x is defined!
{% endif %}
{% if y is not defined %}
  y is not defined
{% else %}
  y is defined
{% endif %}

You can combine conditions with this feature and even use it in expressions:

{% if x is defined && x == "12" && y == Some(true) %}
...
{% endif %}

<script>
// It will generate `const x = true;` (or false is `x` is not defined).
const x = {{ x is defined }};
</script>

Due to proc-macro limitations, askama can only see the fields of your current type and the variables declared in the templates. Because of this, you can not check if a field or a function is defined:

{% if x.y is defined %}
  This code will not compile
{% endif %}

Match

In order to deal with Rust enums in a type-safe way, templates support match blocks from version 0.6. Here is a simple example showing how to expand an Option:

{% match item %}
  {% when Some with ("foo") %}
    Found literal foo
  {% when Some with (val) %}
    Found {{ val }}
  {% when None %}
{% endmatch %}

That is, a {% match %} block may contain whitespaces (but no other literal content) and comment blocks, followed by a number of {% when %} blocks and an optional {% else %} block.

Like in Rust, the matching is done against a pattern. Such a pattern may be a literal, e.g.

{% match multiple_choice_answer %}
  {% when 3 %} Correct!
  {% else %} Sorry, the right answer is "3".
{% endmatch %}

Or some more complex type, such as a Result<T, E>:

{% match result %}
  {% when Ok(val) %} Good: {{ val }}.
  {% when Err(err) %} Bad: {{ err }}.
{% endmatch %}

Using the placeholder _ to match against any value without capturing the datum, works too. The wildcard operator .. is used to match against an arbitrary amount of items, and the same restrictions as in Rust, e.g. that it can be used only once in a slice or struct:

{% match list_of_ints %}
  {% when [first, ..] %} The list starts with a {{ first }}
  {% when _ %} The list is empty.
{% endmatch %}

The {% else %} node is syntactical sugar for {% when _ %}. If used, it must come last, after all other {% when %} blocks:

{% match answer %}
  {% when Ok(42) %} The answer is "42".
  {% else %} No answer wrong answer?
{% endmatch %}

A {% match %} must be exhaustive, i.e. all possible inputs must have a case. This is most easily done by using proving an {% else %} case, if not all possible values need an individual handling.

Because a {% match %} block could not generate valid code otherwise, you have to provide at least one {% when %} case and/or an {% else %} case.

You can also match against multiple alternative patterns at once:

{% match number %}
  {% when 1 | 4 | 86 %} Some numbers
  {% when n %} Number is {{ n }}
{% endmatch %}

For better interoperability with linters and auto-formatters like djLint, you can also use an optional {% endwhen %} node to close a {% when %} case:

{% match number %}
  {% when 0 | 2 | 4 | 6 | 8 %}
    even
  {% endwhen %}
  {% when 1 | 3 | 5 | 7 | 9 %}
    odd
  {% endwhen %}
  {% else }
    unknown
{% endmatch %}

Referencing and dereferencing variables

If you need to put something behind a reference or to dereference it, you can use & and * operators:

{% let x = &"bla" %}
{% if *x == "bla" %}
Just talking
{% else if x == &"another" %}
Another?!
{% endif %}

They have the same effect as in Rust and you can put multiple of them:

{% let x = &&"bla" %}
{% if *&**x == "bla" %}
You got it
{% endif %}

Include

The include statement lets you split large or repetitive blocks into separate template files. Included templates get full access to the context in which they're used, including local variables like those from loops:

{% for i in iter %}
  {% include "item.html" %}
{% endfor %}

* Item: {{ i }}

The path to include must be a string literal, so that it is known at compile time. Askama will try to find the specified template relative to the including template's path before falling back to the absolute template path. Use include within the branches of an if/else block to use includes more dynamically.

Expressions

Askama supports string literals ("foo") and integer literals (1). It supports almost all binary operators that Rust supports, including arithmetic, comparison and logic operators. The parser applies the same operator precedence as the Rust compiler. Expressions can be grouped using parentheses.

{{ 3 * 4 / 2 }}
{{ 26 / 2 % 7 }}
{{ 3 % 2 * 6 }}
{{ 1 * 2 + 4 }}
{{ 11 - 15 / 3 }}
{{ (4 + 5) % 3 }}

The HTML special characters &, < and > will be replaced with their character entities unless the escape mode is disabled for a template, or the filter |safe is used.

Methods can be called on variables that are in scope, including self.

Warning: if the result of an expression (a {{ }} block) is equivalent to self, this can result in a stack overflow from infinite recursion. This is because the Display implementation for that expression will in turn evaluate the expression and yield self again.

Expressions containing bit-operators

In Askama, the binary AND, OR, and XOR operators (called &, |, ^ in Rust, resp.), are renamed to bitand, bitor, xor to avoid confusion with filter expressions. They still have the same operator precedende as in Rust. E.g. to test if the least significant bit is set in an integer field:

{% if my_bitset bitand 1 != 0 %}
    It is set!
{% endif %}

Type conversion

You can use the as operator in {{ … }} expressions, and {% … %} blocks. It works the same as in Rust, but with some deliberate restrictions:

You can only use primitive types like i32 or f64 both as source variable type and as target type.
If the source is a reference to a primitive type, e.g. &&&bool, then askama automatically dereferences the value until it gets the underlying bool.

String concatenation

As a short-hand for {{ a }}{{ b }}{{ c }} you can use the concat operator ~: {{ a ~ b ~ c }}. The tilde ~ has to be surrounded by spaces to avoid confusion with the whitespace control operator.

Templates in templates

Using expressions, it is possible to delegate rendering part of a template to another template. This makes it possible to inject modular template sections into other templates and facilitates testing and reuse.

use askama::Template;
#[derive(Template)]
#[template(source = "Section 1: {{ s1 }}", ext = "txt")]
struct RenderInPlace<'a> {
   s1: SectionOne<'a>
}

#[derive(Template)]
#[template(source = "A={{ a }}\nB={{ b }}", ext = "txt")]
struct SectionOne<'a> {
   a: &'a str,
   b: &'a str,
}

let t = RenderInPlace { s1: SectionOne { a: "a", b: "b" } };
assert_eq!(t.render().unwrap(), "Section 1: A=a\nB=b")

Note that if your inner template like SectionOne renders HTML content, then you may want to disable escaping when injecting it into an outer template, e.g. {{ s1|safe }}. Otherwise it will render the HTML content literally, because askama escapes HTML variables by default.

See the example render in place using a vector of templates in a for block.

Comments

Askama supports block comments delimited by {# and #}.

{# A Comment #}

Like Rust, Askama also supports nested block comments.

{#
A Comment
{# A nested comment #}
#}

Recursive Structures

Recursive implementations should preferably use a custom iterator and use a plain loop. If that is not doable, call .render() directly by using an expression as shown below.

use askama::Template;

#[derive(Template)]
#[template(source = r#"
{{ name }} {
{% for item in children %}
   {{ item.render()? }}
{% endfor %}
}
"#, ext = "html", escape = "none")]
struct Item<'a> {
    name: &'a str,
    children: &'a [Item<'a>],
}

Macros

You can define macros within your template by using {% macro name(args) %}, ending with {% endmacro %}.

You can then call it with {% call name(args) %}:

{% macro heading(arg) %}

<h1>{{arg}}</h1>

{% endmacro %}

{% call heading(s) %}

You can place macros in a separate file and use them in your templates by using {% import %}:

{%- import "macro.html" as scope -%}

{% call scope::heading(s) %}

You can optionally specify the name of the macro in endmacro:

{% macro heading(arg) %}<p>{{arg}}</p>{% endmacro heading %}

You can also specify arguments by their name (as defined in the macro):

{% macro heading(arg, bold) %}

<h1>{{arg}} <b>{{bold}}</b></h1>

{% endmacro %}

{% call heading(bold="something", arg="title") %}

You can use whitespace characters around =:

{% call heading(bold = "something", arg = "title") %}

You can mix named and non-named arguments when calling a macro:

{% call heading("title", bold="something") %}

However please note that named arguments must always come last.

Another thing to note, if a named argument is referring to an argument that would be used for a non-named argument, it will error:

{% macro heading(arg1, arg2, arg3, arg4) %}
{% endmacro %}

{% call heading("something", "b", arg4="ah", arg2="title") %}

In here it's invalid because arg2 is the second argument and would be used by "b". So either you replace "b" with arg3="b" or you pass "title" before:

{% call heading("something", arg3="b", arg4="ah", arg2="title") %}
{# Equivalent of: #}
{% call heading("something", "title", "b", arg4="ah") %}

Default value

You can specify default values for your macro arguments:

{% macro heading(arg1, arg2 = "something") %}
{% endmacro %}

Then if you don't pass a value for this argument, its default value will be used:

{# We only specify `arg1`, so `arg2` will be "something" #}
{% call heading(1) %}
{# We specify both `arg1` and `arg2` so no default value is used #}
{% call heading(1, 2) %}

Calling Rust macros

It is possible to call rust macros directly in your templates:

{% let s = format!("{}", 12) %}

One important thing to note is that contrary to the rest of the expressions, Askama cannot know if a token given to a macro is a variable or something else, so it will always default to generate it "as is". So if you have:

macro_rules! test_macro{
    ($entity:expr) => {
        println!("{:?}", &$entity);
    }
}

#[derive(Template)]
#[template(source = "{{ test_macro!(entity) }}", ext = "txt")]
struct TestTemplate<'a> {
    entity: &'a str,
}

It will not compile, telling you it doesn't know entity. It didn't infer that entity was a field of the current type unlike usual. You can go around this limitation by binding your field's value into a variable:

{% let entity = entity %}
{{ test_macro!(entity) }}

Opt-in features

Some features in askama are opt-in to reduce the amount of dependencies, and to keep the compilation time low.

To opt-in to a feature, you can use features = […]. E.g. if you want to use the filter |json, you have to opt-in to the feature "serde_json":

[dependencies]
askama = { version = "0.14.0", features = ["serde_json"] }

Please read the Cargo manual for more information.

Default features

Any semver-compatible upgrade (e.g. askama = "0.14.1" to askama = "0.14.2") will keep the same list of default features. We will treat upgrades to a newer dependency version as a semver breaking change.

`"default"`

You can opt-out of using the feature flags by using default-features = false:

[dependencies]
askama = { version = "0.14.0", default-features = false }

Without default-features = false, i.e with default features enabled, the following features are automatically selected for you:

default = ["config", "derive", "std", "urlencode"]

This should encompass most features an average user of askama might need.

If you are writing a library that depends on askama, and if you want it to be usable in by other users and in other projects, then you should probably opt-out of features you do not need.

`"derive"`

This feature enables #[derive(Template)]. Without it the trait askama::Template will still be available, but if you want to derive a template, you have to manually depend on askama_derive. askama_derive should be used with the same features as askama.

Not using this feature might be useful e.g. if you are writing a library with manual filters for askama, without any templates. It might also very slightly speed-up the compilation, because more dependencies can be compiled in parallel, because askama won't transitively depend on e.g. syn or proc-macro2. On the author's PC the compilation of a trivial hello-world example was about 0.2s faster without the feature when compiled in release mode.

If you are writing a library that uses askama, consider not using this default-feature.

`"config"`

Enables compile time configurations.

`"urlencode"`

Enables the filters |urlencode and |urlencode_strict.

Addition features

Please note that we reserve the right to add more features to the current list, without labeling it as a semver breaking change. The newly added features might even depend on a newer rustc version than the previous list.

The most useful catch-all feature for a quick start might be "full", which enables all implemented features, i.e.:

full = ["default", "blocks", "code-in-doc", "serde_json"]

In production or once your project is “maturing” you might want to manually opt-in to any needed features with a finer granularity instead of depending on "full".

`"blocks"`

Enables using the template attribute blocks.

`"serde_json"`

This feature depends on the crate serde_json. We won't treat upgrades to a newer serde_json version as a semver breaking change, even if it raises the MSRV.

Enables the filter |json.

`"code-in-doc"`

This feature depends on the crate pulldown-cmark. We won't treat upgrades to a newer pulldown-cmark version as a semver breaking change, even if it raises the MSRV.

Enables using documentations as template code.

“Anti-features” in a `#![no_std]` environment

Opting-out of the default features "std" and "alloc" is only interesting for the use in a #![no_std] environment. Please find more information in The Embedded Rust Book.

`"alloc"`

Without the default feature "alloc" askama can be used in a #![no_std] environment. The method Template::render() will be absent, because askama won't have access to a default allocator.

Many filters need intermediate allocations, and won't be usable without this feature.

You can still render templates using e.g. no_std_io2::io::Cursor or embedded_io::Write

`"std"`

Without the feature "std" askama can be used in a #![no_std] environment. The method Template::write_into() will be absent, because askama won't have access to standard IO operations.

Enabling "std" enables "alloc", too.

Filters

{{ "HELLO" | lower }}

Askama has a collection of built-in filters, documented below, but can also include custom filters. Additionally, the json filter is included in the built-in filters, but is disabled by default. Enable it with Cargo features (see below for more information).

Built-In Filters

Built-in filters that take (optional) arguments can be called with named arguments, too. The order of the named arguments does not matter, but named arguments must come after positional (i.e. unnamed) arguments.

E.g. the filter pluralize takes two optional arguments: singular and plural which are singular = "" and plural = "s" by default. If you are fine with the default empty string for the singular, and you only want to set a specific plural, then you can call the filter like dog{{ count | pluralize(plural = "gies") }}.

capitalize

{{ text_to_capitalize | capitalize }}

Capitalize a value. The first character will be uppercase, all others lowercase:

{{ "hello" | capitalize }}

Output:

Hello

center

{{ text_to_center | center(length) }}

Centers the value in a field of a given width:

-{{ "a" | center(5) }}-

Output:

-  a  -

deref

{{ expression | deref }}

Dereferences the given argument.

{% let s = String::from("a") | as_ref %}
{% if s | deref == String::from("b") %}
{% endif %}

will become:

let s = &String::from("a");
if *s == String::from("b") {}

escape | e

{{ text_to_escape | e }}
{{ text_to_escape | escape }}
{{ text_to_escape | escape(escaper) }}

Escapes HTML characters in strings:

{{ "Escape <>&" | e }}

Output:

Escape &lt;&gt;&amp;

Optionally, it is possible to specify and override which escaper is used. Consider a template where the escaper is configured as escape = "none". However, somewhere escaping using the HTML escaper is desired. Then it is possible to override and use the HTML escaper like this:

{{ "Don't Escape <>&" | escape }}
{{ "Don't Escape <>&" | e }}

{{ "Escape <>&" | escape("html") }}
{{ "Escape <>&" | e("html") }}

Output:

Don't Escape <>&
Don't Escape <>&

Escape &lt;&gt;&amp;
Escape &lt;&gt;&amp;

filesizeformat

{{ number_of_bytes | filesizeformat }}

Returns adequate string representation (in KB, ..) of number of bytes:

{{ 1000 | filesizeformat }}

Output:

1 KB

fmt

{{ expression | fmt("format_string") }}

Formats arguments according to the specified format

The second argument to this filter must be a string literal (as in normal Rust). The two arguments are passed through to format!() by the Askama code generator, but the order is swapped to support filter composition.

{{ value | fmt("{:?}") }}

As an example, this allows filters to be composed like the following. Which is not possible using the format filter.

{{ value | capitalize | fmt("{:?}") }}

format

{{ "format_string" | format([variables ...]) }}

Formats arguments according to the specified format.

The first argument to this filter must be a string literal (as in normal Rust).

All arguments are passed through to format!() by the Askama code generator.

{{ "{:?}" | format(var) }}

indent

{{ text_to_indent | indent(width, [first], [blank]) }}

Indent newlines with width spaces.

{{ "hello\nfoo\nbar" | indent(4) }}

Output:

hello
    foo
    bar

The first argument can also be a string that will be used to indent lines.

The first line and blank lines are not indented by default. The filter has two optional [bool] arguments first and blank, that can be set to true to indent the first and blank lines, resp.:

{{ "hello\n\nbar" | indent("$ ", true, true) }}

Output:

$ hello
$ 
$ bar

join

{{ iterable | join(separator) }}

Joins iterable into a string separated by provided argument.

array = &["foo", "bar", "bazz"]

{{ array | join(", ") }}

Output:

foo, bar, bazz

linebreaks

{{ text_to_break | linebreaks }}

Replaces line breaks in plain text with appropriate HTML.

A single newline becomes an HTML line break <br> and a new line followed by a blank line becomes a paragraph break <p>.

{{ "hello\nworld\n\nfrom\naskama" | linebreaks }}

Output:

<p>hello<br />world</p><p>from<br />askama</p>

linebreaksbr

{{ text_to_break | linebreaksbr }}

Converts all newlines in a piece of plain text to HTML line breaks.

{{ "hello\nworld\n\nfrom\naskama" | linebreaks }}

Output:

hello<br />world<br /><br />from<br />askama

paragraphbreaks

{{ text_to_break | paragraphbreaks }}

A new line followed by a blank line becomes <p>, but, unlike linebreaks, single new lines are ignored and no <br/> tags are generated.

Consecutive double line breaks will be reduced down to a single paragraph break.

This is useful in contexts where changing single line breaks to line break tags would interfere with other HTML elements, such as lists and nested <div> tags.

{{ "hello\nworld\n\nfrom\n\n\n\naskama" | paragraphbreaks }}

Output:

<p>hello\nworld</p><p>from</p><p>askama</p>

lower | lowercase

{{ text_to_convert | lower }}
{{ text_to_convert | lowercase }}

Converts to lowercase.

{{ "HELLO" | lower }}

Output:

hello

`pluralize`

{{ integer | pluralize }}
{{ integer | pluralize([singular = ""], [plural = "s"]) }}

Select a singular or plural version of a word, depending on the input value.

If the value of self.count is +1 or -1, then "cat" is returned, otherwise "cats":

cat{{ count | pluralize }}

You can override the default empty singular suffix, e.g. to spell "doggo" for a single dog:

dog{{ count | pluralize("go") }}

If the word cannot be declined by simply adding a suffix, then you can also override singular and the plural, too:

{{ count | pluralize("mouse", "mice") }}

More complex languages that know multiple plurals might be impossible to implement with this filter, though.

ref

{{ expression | ref }}

Creates a reference to the given argument.

{{ "a" | ref }}
{{ self.x | ref }}

will become:

&"a"
&self.x

safe

{{ expression | safe }}

Marks a string (or other Display type) as safe. By default all strings are escaped according to the format.

{{ "<p>I'm Safe</p>" | safe }}

Output:

<p>I'm Safe</p>

title | titlecase

{{ text_to_convert | title }}
{{ text_to_convert | titlecase }}

Return a title cased version of the value. Words will start with uppercase letters, all remaining characters are lowercase.

{{ "hello WORLD" | title }}

Output:

Hello World

trim

{{ text_to_trim | trim }}

Strip leading and trailing whitespace.

{{ " hello " | trim }}

Output:

hello

truncate

{{ text_to_truncate | truncate(length) }}

Limit string length, appends '...' if truncated.

{{ "hello" | truncate(2) }}

Output:

he...

unique

Returns an iterator with all duplicates removed.

This filter is only available with the std feature enabled.

With this data:

vec!["a", "b", "a", "c"]

And this template:

{% for elem in data|unique %}{{ elem }},{% endfor %}

Output will be:

a,b,c,

upper | uppercase

{{ text_to_convert | upper }}
{{ text_to_convert | uppercase }}

Converts to uppercase.

{{ "hello" | upper }}

Output:

HELLO

urlencode | urlencode_strict

{{ text_to_escape | urlencode }}
{{ text_to_escape | urlencode_strict }}

Percent encodes the string. Replaces reserved characters with the % escape character followed by a byte value as two hexadecimal digits.

{{ "hello?world" | urlencode }}

Output:

hello%3Fworld

With |urlencode all characters except ASCII letters, digits, and _.-~/ are escaped. With |urlencode_strict a forward slash / is escaped, too.

wordcount

{{ text_with_words | wordcount }}

Count the words in that string.

{{ "askama is sort of cool" | wordcount }}

Output:

Optional / feature gated filters

The following filters can be enabled by requesting the respective feature in the Cargo.toml dependencies section, e.g.

[dependencies]
askama = { version = "0.12", features = ["serde_json"] }

`json` | `tojson`

{{ value_to_serialize | json }}
{{ value_to_serialize | json(indent) }}

Enabling the serde_json feature will enable the use of the json filter. This will output formatted JSON for any value that implements the required Serialize trait. The generated string does not contain ampersands &, chevrons < >, or apostrophes '.

To use it in a <script> you can combine it with the safe filter. In HTML attributes, you can either use it in quotation marks "{{data | json}}" as is, or in apostrophes with the (optional) safe filter '{{data | json | safe}}'. In HTML texts the output of e.g. <pre>{{data | json | safe}}</pre> is safe, too.

Good: <li data-extra="{{data | json}}">…</li>
Good: <li data-extra='{{data | json | safe}}'>…</li>
Good: <pre>{{data | json | safe}}</pre>
Good: <script>var data = {{data | json | safe}};</script>

Bad:  <li data-extra="{{data | json | safe}}">…</li>
Bad:  <script>var data = {{data | json}};</script>
Bad:  <script>var data = "{{data | json | safe}}";</script>

Ugly: <script>var data = "{{data | json}}";</script>
Ugly: <script>var data = '{{data | json | safe}}';</script>

By default, a compact representation of the data is generated, i.e. no whitespaces are generated between individual values. To generate a readable representation, you can either pass an integer how many spaces to use as indentation, or you can pass a string that gets used as prefix:

Prefix with four spaces:
<textarea>{{data | tojson(4)}}</textarea>

Prefix with two &nbsp; characters:
<p>{{data | tojson("\u{a0}\u{a0}")}}</p>

Custom Filters

To define your own filters, simply have a module named filters in scope of the context deriving a Template impl and define the filters as functions within this module. The functions must have at least two arguments and the return type must be askama::Result<T>. Although there are no restrictions on T for a single filter, the final result of a chain of filters must implement Display.

The arguments to the filters are passed as follows:

The first argument corresponds to the expression they are applied to.
The second argument are the runtime values: values: &dyn askama::Values.
Subsequent arguments, if any, must be given directly when calling the filter. The first argument may or may not be a reference, depending on the context in which the filter is called.

To abstract over ownership, consider defining your argument as a trait bound. For example, the trim built-in filter accepts any value implementing Display. Its signature is similar to fn trim(s: impl std::fmt::Display, values: &dyn askama::Values) -> askama::Result<String>.

Note that built-in filters have preference over custom filters, so, in case of name collision, the built-in filter is applied. Custom filters cannot have named or optional arguments.

Examples

Implementing a filter that replaces all instances of "oo" for "aa".

use askama::Template;

#[derive(Template)]
#[template(source = "{{ s | myfilter }}", ext = "txt")]
struct MyFilterTemplate<'a> {
    s: &'a str,
}

// Any filter defined in the module `filters` is accessible in your template.
mod filters {
    // This filter does not have extra arguments
    pub fn myfilter<T: std::fmt::Display>(
        s: T,
        _: &dyn askama::Values,
    ) -> askama::Result<String> {
        let s = s.to_string();
        Ok(s.replace("oo", "aa"))
    }
}

fn main() {
    let t = MyFilterTemplate { s: "foo" };
    assert_eq!(t.render().unwrap(), "faa");
}

Implementing a filter that replaces all instances of "oo" for n times "a".

use askama::Template;

#[derive(Template)]
#[template(source = "{{ s | myfilter(4) }}", ext = "txt")]
struct MyFilterTemplate<'a> {
    s: &'a str,
}

// Any filter defined in the module `filters` is accessible in your template.
mod filters {
    // This filter requires a `usize` input when called in templates
    pub fn myfilter<T: std::fmt::Display>(
        s: T,
        _: &dyn askama::Values,
        n: usize,
    ) -> askama::Result<String> {
        let s = s.to_string();
        let mut replace = String::with_capacity(n);
        replace.extend((0..n).map(|_| "a"));
        Ok(s.replace("oo", &replace))
    }
}

fn main() {
    let t = MyFilterTemplate { s: "foo" };
    assert_eq!(t.render().unwrap(), "faaaa");
}

Runtime values

It is possible to access runtime values in custom filters:

// This example contains a custom filter `|cased`.
// Depending on the runtime value `"case"`, the input is either turned
// into lower case, upper case, or left alone, if the runtime value is undefined.

use std::any::Any;

use askama::{Template, Values};

mod filters {
    use super::*;

    pub fn cased(value: impl ToString, values: &dyn Values) -> askama::Result<String> {
        let value = value.to_string();
        let case = askama::get_value(values, "case").ok();
        Ok(match case {
            Some(Case::Lower) => value.to_lowercase(),
            Some(Case::Upper) => value.to_uppercase(),
            None => value,
        })
    }
}

#[derive(Debug, Clone, Copy)]
pub enum Case {
    Lower,
    Upper,
}

#[test]
fn test_runtime_values_in_custom_filters() {
    #[derive(Template)]
    #[template(ext = "txt", source = "Hello, {{ user | cased }}!")]
    struct MyStruct<'a> {
        user: &'a str,
    }

    // The filter source ("wOrLd") should be written in lower case.
    let values: (&str, &dyn Any) = ("case", &Case::Lower);
    assert_eq!(
        MyStruct { user: "wOrLd" }
            .render_with_values(&values)
            .unwrap(),
        "Hello, world!"
    );

    // The filter source ("wOrLd") should be written in upper case.
    let values: (&str, &dyn Any) = ("case", &Case::Upper);
    assert_eq!(
        MyStruct { user: "wOrLd" }
            .render_with_values(&values)
            .unwrap(),
        "Hello, WORLD!"
    );

    // The filter source ("wOrLd") should be written as is.
    assert_eq!(
        MyStruct { user: "wOrLd" }.render().unwrap(),
        "Hello, wOrLd!"
    );
}

HTML-safe types

Askama will try to avoid escaping types that generate string representations that do not contain "HTML-unsafe characters". HTML-safe characters are characters that can be used in any context in HTML texts and attributes. The "unsafe" characters are: <, >, &, " and '.

In order to know which types do not need to be escaped, askama has the marker trait askama::filters::HtmlSafe, and any type that implements that trait won't get automatically escaped in a {{expr}} expression. By default e.g. all primitive integer types are marked as HTML-safe.

You can also mark your custom type MyStruct as HTML-safe using:

impl askama::filters::HtmlSafe for MyStruct {}

This automatically marks references &MyStruct as HTML-safe, too.

Safe output of custom filters

Say, you have a custom filter | strip that removes all HTML-unsafe characters:

fn strip(s: impl ToString) -> Result<String, askama::Error> {
    Ok(s.to_string()
        .chars()
        .filter(|c| !matches!(c, '<' | '>' | '&' | '"' | '\''))
        .collect()
    )
}

Then you can also mark the output as safe using askama::filters::Safe:

fn strip(s: impl ToString) -> Result<Safe<String>, askama::Error> {
    Ok(Safe(...))
}

There also is askama::filters::MaybeSafe that can be used to mark some output as safe, if you know that some inputs for our filter will always result in a safe output:

fn as_sign(i: i32) -> Result<MaybeSafe<&'static str>, askama::Error> {
    match i.into() {
        i if i < 0 => Ok(MaybeSafe::NeedsEscaping("<0")),
        i if i > 0 => Ok(MaybeSafe::NeedsEscaping(">0")),
        _          => Ok(MaybeSafe::Safe("=0")),
    }
}

Working with web-frameworks

Askama's Template::render() returns Result<String, askama::Error>. To make this result work in your preferred web-framework, you'll need to handle both cases: converting the String to a web-response with the correct Content-Type, and the Error case to a proper error message.

While in many cases it will be enough to simply convert the Error to

Box<dyn std::error::Error + Send + Sync> using err.into_box() or
std::io::Error using err.into_io_error()

it is recommended to use a custom error type. This way you can display the error message in your app's layout, and you are better prepared for the likely case that your app grows in the future. Maybe you'll need to access a database and handle errors? Maybe you'll add multiple languages and you want to localize error messages?

The crates thiserror and displaydoc can be useful to implement this error type.

Simplified alternative

Alternatively, you can use #[derive(askama_web::WebTemplate)] to automatically implement e.g. actix-web's Responder, axum's IntoResponse or warp's Reply. The library implements traits for all web-frameworks mentioned down below (and then some), but it does not stylize error messages.

If you don't need custom / stylized error messages, e.g. because you know that your templates won't have rendering errors, then using askama_web might work for you, too.

Actix-Web

To convert the String to an HTML response, you can use Html::new(_).

use actix_web::web::Html;
use actix_web::{Responder, handler};

#[handler]
fn handler() -> Result<impl Responder, AppError> {
    …
    Ok(Html::new(template.render()?))
}

To implement your own error type, you can use this boilerplate code:

use actix_web::{HttpResponse, Responder};
use actix_web::error::ResponseError;
use actix_web::http::StatusCode;
use actix_web::web::Html;
use askama::Template;

#[derive(Debug, displaydoc::Display, thiserror::Error)]
enum AppError {
    /// could not render template
    Render(#[from] askama::Error),
}

impl ResponseError for AppError {
    fn status_code(&self) -> StatusCode {
        match &self {
            AppError::Render(_) => StatusCode::INTERNAL_SERVER_ERROR,
        }
    }
}

impl Responder for AppError {
    type Body = String;

    fn respond_to(self, req: &HttpRequest) -> HttpResponse<Self::Body> {
        #[derive(Debug, Template)]
        #[template(path = "error.html")]
        struct Tmpl { … }

        let tmpl = Tmpl { … };
        if let Ok(body) = tmpl.render() {
            (Html::new(body), self.status_code()).respond_to(req)
        } else {
            (String::new(), self.status_code()).respond_to(req)
        }
    }
}

Axum

To convert the String to an HTML response, you can use Html(_).

use axum::response::{Html, IntoResponse};

async fn handler() -> Result<impl IntoResponse, AppError> {
    …
    Ok(Html(template.render()?))
}

To implement your own error type, you can use this boilerplate code:

use axum::response::IntoResponse;
use askama::Template;

#[derive(Debug, displaydoc::Display, thiserror::Error)]
enum AppError {
    /// could not render template
    Render(#[from] askama::Error),
}

impl IntoResponse for AppError {
    fn into_response(self) -> Response {
        #[derive(Debug, Template)]
        #[template(path = "error.html")]
        struct Tmpl { … }

        let status = match &self {
            AppError::Render(_) => StatusCode::INTERNAL_SERVER_ERROR,
        };
        let tmpl = Tmpl { … };
        if let Ok(body) = tmpl.render() {
            (status, Html(body)).into_response()
        } else {
            (status, "Something went wrong").into_response()
        }
    }
}

Poem

To convert the String to an HTML response, you can use Html(_).

use poem::web::Html;
use poem::{IntoResponse, handler};

#[handler]
async fn handler() -> Result<impl IntoResponse, AppError> {
    …
    Ok(Html(template.render()?))
}

To implement your own error type, you can use this boilerplate code:

use poem::error::ResponseError;
use poem::http::StatusCode;
use poem::web::Html;
use poem::{IntoResponse, Response};
use askama::Template;

#[derive(Debug, displaydoc::Display, thiserror::Error)]
enum AppError {
    /// could not render template
    Render(#[from] askama::Error),
}

impl ResponseError for AppError {
    fn status(&self) -> StatusCode {
        match &self {
            AppError::Render(_) => StatusCode::INTERNAL_SERVER_ERROR,
        }
    }
}

impl IntoResponse for AppError {
    fn into_response(self) -> Response {
        #[derive(Debug, Template)]
        #[template(path = "error.html")]
        struct Tmpl { … }

        let tmpl = Tmpl { … };
        if let Ok(body) = tmpl.render() {
            (self.status(), Html(body)).into_response()
        } else {
            (self.status(), "Something went wrong").into_response()
        }
    }
}

Rocket

To convert the String to an HTML response, you can use RawHtml(_).

use rocket::get;
use rocket::response::content::RawHtml;
use rocket::response::Responder;

#[get(…)]
fn handler<'r>() -> Result<impl Responder<'r, 'static>, AppError> {
    …
    Ok(RawHtml(template.render()?))
}

To implement your own error type, you can use this boilerplate code:

use askama::Template;
use rocket::http::Status;
use rocket::response::content::RawHtml;
use rocket::response::Responder;
use rocket::{Request, Response};

#[derive(Debug, displaydoc::Display, thiserror::Error)]
enum AppError {
    /// could not render template
    Render(#[from] askama::Error),
}

impl<'r> Responder<'r, 'static> for AppError {
    fn respond_to(
        self,
        request: &'r Request<'_>,
    ) -> Result<Response<'static>, Status> {
        #[derive(Debug, Template)]
        #[template(path = "error.html")]
        struct Tmpl { … }

        let status = match &self {
            AppError::Render(_) => Status::InternalServerError,
        };
        let template = Tmpl { … };
        if let Ok(body) = template.render() {
            (status, RawHtml(body)).respond_to(request)
        } else {
            (status, "Something went wrong").respond_to(request)
        }
    }
}

Salvo

To convert the String to an HTML response, you can use Text::Html(_).

use salvo::writing::Text;
use salvo::{Scribe, handler};

#[handler]
async fn handler() -> Result<impl Scribe, AppError> {
    …
    Ok(Text::Html(template.render()?))
}

To implement your own error type, you can use this boilerplate code:

use askama::Template;
use salvo::http::StatusCode;
use salvo::writing::Text;
use salvo::{Response, Scribe};

#[derive(Debug, displaydoc::Display, thiserror::Error)]
enum AppError {
    /// could not render template
    Render(#[from] askama::Error),
}

impl Scribe for AppError {
    fn render(self, res: &mut Response) {
        #[derive(Debug, Template)]
        #[template(path = "error.html")]
        struct Tmpl { … }

        res.status_code(match &self {
            AppError::Render(_) => StatusCode::INTERNAL_SERVER_ERROR,
        });
        let tmpl = Tmpl { … };
        if let Ok(body) = tmpl.render() {
            Text::Html(body).render(res);
        } else {
            Text::Plain("Something went wrong").render(res);
        }
    }
}

Warp

To convert the String to an HTML response, you can use html(_).

use warp::reply::{Reply, html};

fn handler() -> Result<impl Reply, AppError> {
    …
    Ok(html(template.render()?))
}

To implement your own error type, you can use this boilerplate code:

use http::StatusCode;
use warp::reply::{Reply, Response, html};

#[derive(Debug, displaydoc::Display, thiserror::Error)]
enum AppError {
    /// could not render template
    Render(#[from] askama::Error),
}

impl Reply for AppError {
    fn into_response(self) -> Response {
        #[derive(Debug, Template)]
        #[template(path = "error.html")]
        struct Tmpl { … }

        let status = match &self {
            AppError::Render(_) => StatusCode::INTERNAL_SERVER_ERROR,
        };
        let template = Tmpl { … };
        if let Ok(body) = template.render() {
            with_status(html(body), status).into_response()
        } else {
            status.into_response()
        }
    }
}

Performance

Rendering Performance

When rendering an askama template, you should prefer the methods

.render() (to render the content into a new string),
.render_into() (to render the content into an fmt::Write object, e.g. String) or
.write_into() (to render the content into an io::Write object, e.g. Vec<u8>)

over .to_string() or format!(). While .to_string() and format!() give you the same result, they generally perform much worse than askama's own methods, because fmt::Write uses dynamic methods calls instead of monomorphised code. On average, expect .to_string() to be 100% to 200% slower than .render().

Faster Rendering of Custom Types

Every type that implements fmt::Display can be used in askama expressions: {{ value }}. Rendering with fmt::Display can be slow, though, because it uses dynamic methods calls in its fmt::Formatter argument. To speed up rendering (by a lot, actually), askama adds the trait FastWritable. For any custom type you want to render, it has to implement fmt::Display, but if it also implements FastWritable, then – using autoref-based specialization – the latter implementation is automatically preferred.

To reduce the amount of code duplication, you can let your fmt::Display implementation call your FastWritable implementation:

use std::fmt::{self, Write};

use askama::{FastWritable, NO_VALUES};

// In a real application, please have a look at
// https://github.com/kdeldycke/awesome-falsehood/blob/690a070/readme.md#human-identity
struct Name<'a> {
    forename: &'a str,
    surname: &'a str,
}

impl fmt::Display for Name<'_> {
    // Because the method simply forwards the call, it should be `inline`.
    #[inline]
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        // `fmt::Write` has no access to runtime values,
        // so simply pass `NO_VALUES`.
        self.write_into(f, NO_VALUES)?;
        Ok(())
    }
}

impl FastWritable for Name<'_> {
    fn write_into<W: fmt::Write + ?Sized>(
        &self,
        dest: &mut W,
        _values: &dyn askama::Values,
    ) -> askama::Result<()> {
        dest.write_str(self.surname)?;
        dest.write_str(", ")?;
        dest.write_str(self.forename)?;
        Ok(())
    }
}

#[test]
fn both_implementations_should_render_the_same_text() {
    let person = Name {
        forename: "Max",
        surname: "Mustermann",
    };

    let mut buf_fmt = String::new();
    write!(buf_fmt, "{person}").unwrap();

    let mut buf_fast = String::new();
    person.write_into(&mut buf_fast, NO_VALUES).unwrap();

    assert_eq!(buf_fmt, buf_fast);
    assert_eq!(buf_fmt, "Mustermann, Max");
}

Slow Debug Recompilations

If you experience slow compile times when iterating with lots of templates, you can compile Askama's derive macros with a higher optimization level. This can speed up recompilation times dramatically.

Add the following to Cargo.toml or .cargo/config.toml:

[profile.dev.package.askama_derive]
opt-level = 3

This may affect clean compile times in debug mode, but incremental compiles will be faster.

Upgrading to new versions

This file only lists breaking changes you need to be aware of when you upgrade to a new askama version. Please see our release notes to get a list of all changes and improvements that might be useful to you.

From askama v0.13 to askama v0.14

The MSRV of this release is 1.83.
When assigning a new variable ({% let var = … %}) with a local variable was value, it is moved or copied, not referenced.
Try expressions ({{ expr? }}) are not placed behind a reference.
FastWritable implementations have access to runtime values.
Custom filters have access to runtime values.
|unique is a built-in filter; |titlecase is an alias for |title.

From askama v0.12 to askama v0.13

A blog post summarizing changes and also explaining the merge of rinja and askama is available here.

List of breaking changes:

The MSRV of this release is 1.81.
The integration crates were removed. Instead of depending on e.g. askama_axum / askama_axum, please use template.render() to render to a Result<String, askama::Error>.

Use e.g. .map_err(|err| err.into_io_error())? if your web-framework expects std::io::Errors, or err.into_box() if it expects Box<dyn std::error::Error + Send + Sync>.

Please read the documentation of your web-framework how to turn a String into a web-response.
The fields Template::EXTENSION and Template::MIME_TYPE were removed.
You may not give variables a name starting with __askama, or the name of a rust keyword.
#[derive(Template)] cannot be used with unions.
|linebreaks, |linebreaksbr and |paragraphbreaks escape their input automatically.
|json does not prettify its output by default anymore. Use e.g. |json(2) for readable output.
The binary operators |, & and ^ are now called bitor, bitand and xor, resp.
The feature "humansize" was removed. The filter |humansize is always available.
The feature "serde-json" is now called "serde_json".
The feature "markdown" was removed. Use comrak directly.
The feature "serde-yaml" was removed. Use e.g. yaml-rust2 directly.

From rinja v0.3 to askama v0.13

The MSRV of this release is 1.81.
The projects rinja and askama were re-unified into one project. You need to replace instances of rinja with askama, e.g.
```
-use rinja::Template;
+use askama::Template;
```
```
 [dependencies]
-rinja = "0.3.5"
+askama = "0.13.0"
```
The integration crates were removed. Instead of depending on e.g. rinja_axum / askama_axum, please use template.render() to render to a Result<String, askama::Error>.

Use e.g. .map_err(|err| err.into_io_error())? if your web-framework expects std::io::Errors, or err.into_box() if it expects Box<dyn std::error::Error + Send + Sync>.

Please read the documentation of your web-framework how to turn a String into a web-response.
The fields Template::EXTENSION and Template::MIME_TYPE were removed.
The feature "humansize" was removed. The filter |humansize is always available.
You may not give variables a name starting with __rinja, or the name of a rust keyword.
#[derive(Template)] cannot be used with unions.

From rinja v0.2 to rinja v0.3

You should be able to upgrade to v0.3 without changes.

From askama v0.12 to rinja v0.2

Have a look at our blog posts that highlight some of the best features of our releases, and give you more in-dept explanations: docs.rs switching jinja template framework from tera to rinja.

The MSRV of this release is 1.71.

You need to replace instances of askama with rinja, e.g.

-use askama::Template;
+use rinja::Template;

 [dependencies]
-askama = "0.12.1"
+rinja = "0.2.0"

|linebreaks, |linebreaksbr and |paragraphbreaks escape their input automatically.
|json does not prettify its output by default anymore. Use e.g. |json(2) for readable output.
The binary operators |, & and ^ are now called bitor, bitand and xor, resp.
The feature "serde-json" is now called "serde_json".
The feature "markdown" was removed. Use comrak directly.
The feature "serde-yaml" was removed. Use e.g. yaml-rust2 directly.

From askama v0.11 to askama v0.12

The magic _parent field to access &**self was removed.
Integration implementations do not need an ext argument anymore.
The iron integration was removed.