#xml #serde #parser #writer #html

archived fast-xml

High performance xml reader and writer

4 releases

0.23.1 May 30, 2022
0.23.0 May 8, 2022
0.23.0-alpha3 May 2, 2022
0.22.0 May 2, 2022

#36 in #writer

Download history 564/week @ 2023-06-08 590/week @ 2023-06-15 435/week @ 2023-06-22 261/week @ 2023-06-29 383/week @ 2023-07-06 508/week @ 2023-07-13 735/week @ 2023-07-20 491/week @ 2023-07-27 479/week @ 2023-08-03 339/week @ 2023-08-10 452/week @ 2023-08-17 811/week @ 2023-08-24 543/week @ 2023-08-31 941/week @ 2023-09-07 719/week @ 2023-09-14 923/week @ 2023-09-21

3,319 downloads per month
Used in 4 crates (2 directly)

MIT license


quick-xml returned back and active again. Please use it

fast-xml -- successor of quick-xml

status Crate

High performance xml pull reader/writer.

The reader:

  • is almost zero-copy (use of Cow whenever possible)
  • is easy on memory allocation (the API provides a way to reuse buffers)
  • support various encoding (with encoding feature), namespaces resolution, special characters.


Syntax is inspired by xml-rs.

Migration from quick-xml

If you using quick-xml 0.22.0 or 0.23.0-alpha3, you can just replace quick-xml in your Cargo.toml with fast-xml. Replace each occurrence of quick_xml crate name to fast_xml in your code base.

That two releases of fast-xml was specifically made for migration and contains the same code as original quick-xml, except updated cargo metadata and extern crate names in tests, benches and examples.



use fast_xml::Reader;
use fast_xml::events::Event;

let xml = r#"<tag1 att1 = "test">
                <tag2><!--Test comment-->Test</tag2>
                    Test 2

let mut reader = Reader::from_str(xml);

let mut count = 0;
let mut txt = Vec::new();
let mut buf = Vec::new();

// The `Reader` does not implement `Iterator` because it outputs borrowed data (`Cow`s)
loop {
    // NOTE: this is the generic case when we don't know about the input BufRead.
    // when the input is a &str or a &[u8], we don't actually need to use another
    // buffer, we could directly call `reader.read_event_unbuffered()`
    match reader.read_event(&mut buf) {
        Ok(Event::Start(ref e)) => {
            match e.name() {
                b"tag1" => println!("attributes values: {:?}",
                                    e.attributes().map(|a| a.unwrap().value).collect::<Vec<_>>()),
                b"tag2" => count += 1,
                _ => (),
        Ok(Event::Text(e)) => txt.push(e.unescape_and_decode(&reader).unwrap()),
        Ok(Event::Eof) => break, // exits the loop when reaching end of file
        Err(e) => panic!("Error at position {}: {:?}", reader.buffer_position(), e),
        _ => (), // There are several other `Event`s we do not consider here

    // if we don't keep a borrow elsewhere, we can clear the buffer to keep memory usage low


use fast_xml::Writer;
use fast_xml::Reader;
use fast_xml::events::{Event, BytesEnd, BytesStart};
use std::io::Cursor;
use std::iter;

let xml = r#"<this_tag k1="v1" k2="v2"><child>text</child></this_tag>"#;
let mut reader = Reader::from_str(xml);
let mut writer = Writer::new(Cursor::new(Vec::new()));
let mut buf = Vec::new();
loop {
    match reader.read_event(&mut buf) {
        Ok(Event::Start(ref e)) if e.name() == b"this_tag" => {

            // crates a new element ... alternatively we could reuse `e` by calling
            // `e.into_owned()`
            let mut elem = BytesStart::owned(b"my_elem".to_vec(), "my_elem".len());

            // collect existing attributes
            elem.extend_attributes(e.attributes().map(|attr| attr.unwrap()));

            // copy existing attributes, adds a new my-key="some value" attribute
            elem.push_attribute(("my-key", "some value"));

            // writes the event to the writer
        Ok(Event::End(ref e)) if e.name() == b"this_tag" => {
        Ok(Event::Eof) => break,
	// you can use either `e` or `&e` if you don't want to move the event
        Ok(e) => assert!(writer.write_event(&e).is_ok()),
        Err(e) => panic!("Error at position {}: {:?}", reader.buffer_position(), e),

let result = writer.into_inner().into_inner();
let expected = r#"<my_elem k1="v1" k2="v2" my-key="some value"><child>text</child></my_elem>"#;
assert_eq!(result, expected.as_bytes());


When using the serialize feature, fast-xml can be used with serde's Serialize/Deserialize traits.

Here is an example deserializing crates.io source:

// Cargo.toml
// [dependencies]
// serde = { version = "1.0", features = [ "derive" ] }
// fast-xml = { version = "0.22", features = [ "serialize" ] }
use serde::Deserialize;
use fast_xml::de::{from_str, DeError};

#[derive(Debug, Deserialize, PartialEq)]
struct Link {
    rel: String,
    href: String,
    sizes: Option<String>,

#[derive(Debug, Deserialize, PartialEq)]
#[serde(rename_all = "lowercase")]
enum Lang {

#[derive(Debug, Deserialize, PartialEq)]
struct Head {
    title: String,
    #[serde(rename = "link", default)]
    links: Vec<Link>,

#[derive(Debug, Deserialize, PartialEq)]
struct Script {
    src: String,
    integrity: String,

#[derive(Debug, Deserialize, PartialEq)]
struct Body {
    #[serde(rename = "script", default)]
    scripts: Vec<Script>,

#[derive(Debug, Deserialize, PartialEq)]
struct Html {
    lang: Option<String>,
    head: Head,
    body: Body,

fn crates_io() -> Result<Html, DeError> {
    let xml = "<!DOCTYPE html>
        <html lang=\"en\">
            <meta charset=\"utf-8\">
            <meta http-equiv=\"X-UA-Compatible\" content=\"IE=edge\">
            <meta name=\"viewport\" content=\"width=device-width, initial-scale=1\">

            <title>crates.io: Rust Package Registry</title>

        <link rel=\"manifest\" href=\"/manifest.webmanifest\">
        <link rel=\"apple-touch-icon\" href=\"/cargo-835dd6a18132048a52ac569f2615b59d.png\" sizes=\"227x227\">

            <link rel=\"stylesheet\" href=\"/assets/vendor-8d023d47762d5431764f589a6012123e.css\" integrity=\"sha256-EoB7fsYkdS7BZba47+C/9D7yxwPZojsE4pO7RIuUXdE= sha512-/SzGQGR0yj5AG6YPehZB3b6MjpnuNCTOGREQTStETobVRrpYPZKneJwcL/14B8ufcvobJGFDvnTKdcDDxbh6/A==\" >
            <link rel=\"stylesheet\" href=\"/assets/cargo-cedb8082b232ce89dd449d869fb54b98.css\" integrity=\"sha256-S9K9jZr6nSyYicYad3JdiTKrvsstXZrvYqmLUX9i3tc= sha512-CDGjy3xeyiqBgUMa+GelihW394pqAARXwsU+HIiOotlnp1sLBVgO6v2ZszL0arwKU8CpvL9wHyLYBIdfX92YbQ==\" >

            <link rel=\"shortcut icon\" href=\"/favicon.ico\" type=\"image/x-icon\">
            <link rel=\"icon\" href=\"/cargo-835dd6a18132048a52ac569f2615b59d.png\" type=\"image/png\">
            <link rel=\"search\" href=\"/opensearch.xml\" type=\"application/opensearchdescription+xml\" title=\"Cargo\">
            <!-- EMBER_CLI_FASTBOOT_BODY -->
                <div id=\"main\">
                    <div class='noscript'>
                        This site requires JavaScript to be enabled.

            <script src=\"/assets/vendor-bfe89101b20262535de5a5ccdc276965.js\" integrity=\"sha256-U12Xuwhz1bhJXWyFW/hRr+Wa8B6FFDheTowik5VLkbw= sha512-J/cUUuUN55TrdG8P6Zk3/slI0nTgzYb8pOQlrXfaLgzr9aEumr9D1EzmFyLy1nrhaDGpRN1T8EQrU21Jl81pJQ==\" ></script>
            <script src=\"/assets/cargo-4023b68501b7b3e17b2bb31f50f5eeea.js\" integrity=\"sha256-9atimKc1KC6HMJF/B07lP3Cjtgr2tmET8Vau0Re5mVI= sha512-XJyBDQU4wtA1aPyPXaFzTE5Wh/mYJwkKHqZ/Fn4p/ezgdKzSCFu6FYn81raBCnCBNsihfhrkb88uF6H5VraHMA==\" ></script>

    let html: Html = from_str(xml)?;
    assert_eq!(&html.head.title, "crates.io: Rust Package Registry");


This has largely been inspired by serde-xml-rs. fast-xml follows its convention for deserialization, including the $value special name.

Original quick-xml was developed by @tafia and abandoned around end of 2021.

Parsing the "value" of a tag

If you have an input of the form <foo abc="xyz">bar</foo>, and you want to get at the bar, you can use the special name $value:

struct Foo {
    pub abc: String,
    #[serde(rename = "$value")]
    pub body: String,

Unflattening structs into verbose XML

If your XML files look like <root><first>value</first><second>value</second></root>, you can (de)serialize them with the special name prefix $unflatten=:

struct Root {
    #[serde(rename = "$unflatten=first")]
    first: String,
    #[serde(rename = "$unflatten=second")]
    other_field: String,

Serializing unit variants as primitives

The $primitive prefix lets you serialize enum variants without associated values (internally referred to as unit variants) as primitive strings rather than self-closing tags. Consider the following definitions:

enum Foo {
    #[serde(rename = "$primitive=Bar")]

struct Root {
    foo: Foo

Serializing Root { foo: Foo::Bar } will then yield <Root foo="Bar"/> instead of <Root><Bar/></Root>.


Note that despite not focusing on performance (there are several unnecessary copies), it remains about 10x faster than serde-xml-rs.


  • encoding: support non utf8 xmls
  • serialize: support serde Serialize/Deserialize


Benchmarking is hard and the results depend on your input file and your machine.

Here on my particular file, fast-xml is around 50 times faster than xml-rs crate. (measurements was done while this crate named quick-xml)

// quick-xml benches
test bench_quick_xml            ... bench:     198,866 ns/iter (+/- 9,663)
test bench_quick_xml_escaped    ... bench:     282,740 ns/iter (+/- 61,625)
test bench_quick_xml_namespaced ... bench:     389,977 ns/iter (+/- 32,045)

// same bench with xml-rs
test bench_xml_rs               ... bench:  14,468,930 ns/iter (+/- 321,171)

// serde-xml-rs vs serialize feature
test bench_serde_quick_xml      ... bench:   1,181,198 ns/iter (+/- 138,290)
test bench_serde_xml_rs         ... bench:  15,039,564 ns/iter (+/- 783,485)

For a feature and performance comparison, you can also have a look at RazrFalcon's parser comparison table.


Any PR is welcomed!




~39K SLoC