Crate xml_oxide[][src]

Expand description

xml_oxide github Released API docs

Rust XML parser implementation that parses any well-formed XML defined in the W3C Spec in a streaming way.


  • It uses constant-like memory for large XML files
  • Fast enough for most use cases. It can parse a 1GB XML file(in memory) around 23 seconds
  • Supports Namespaces in XML 1.0
  • It only supports UTF-8 encoding
  • It is a non-validating parser, it does important well-formedness checks
  • Currently, it ignores well-formedness inside Processing Instructions, DTD/DOCTYPE and parses them as raw strings. It checks the general well-formedness including these entities. (It even parses comments inside DOCTYPE to achieve this)
  • It can parse not-well-formed documents (please report as a bug)
  • Entities that can be large are parsed as chunks to keep memory usage low: Character Data, CDATA Section, Comment, Whitespace
  • Reading chunk size is currently default 8KB, not configurable. Internal buffer is 16KB. If you have an element tag or DOCTYPE declaration that is bigger than the buffer, it can backtrack and allocate more memory for the parsing operation. 1 byte chunk size is used for testing.

Unsafe usage

  • unsafe is used for function std::str::from_utf8_unchecked. It is used on a slice of bytes that is already checked to be a valid UTF8 string with std::str::from_utf8 before. The performance saving is not tested though.
  • RefCell is not used anymore. Interestingly, just changing RefCell<Vec<u8>> to circular::Buffer passed Rust borrow checks. I’m leaving this note as a reference. RefCell is used because Rust is too restrictive for using mutables in conditional loops. Hopefully, non-lexical lifetimes will get better over time.

To Do

  • Because the namespace spec brings constraints around the usage of “:” in names. Provide namespace-aware=false option to parse otherwise valid XML 1.0 documents .
  • More tests
  • Parsing every entity including DTD to be able to utilize conformance test suite.

Example Usage

In this example StartElement and EndElement events are counted. Note that you can find more examples under tests directory.

  • StartElement also include empty tags. Checked by is_empty.
  • Reference entities like &amp; or &#60; comes in its own event(Not in Characters).
  • Character/numerical and predefined entity references are resolved. Custom entity definitions are passed as raw.
  • Check sax::Event to see all available event types
use std::fs::File;
use xml_oxide::{sax::parser::Parser, sax::Event};

fn main() {

    let mut counter: usize = 0;
    let mut end_counter: usize = 0;

    let now = std::time::Instant::now();

    let f = File::open("./tests/xml_files/books.xml").unwrap();

    let mut p = Parser::from_reader(f);

    loop {
        let res = p.read_event();

        match res {
            Ok(event) => match event {
                Event::StartDocument => {}
                Event::EndDocument => {
                Event::StartElement(el) => {
                    //You can differantiate between Starting Tag and Empty Element Tag
                    if !el.is_empty {
                        counter = counter + 1;
                        // print every 10000th element name
                        if counter % 10000 == 0 {
                            println!("%10000 start {}",;
                Event::EndElement(el) => {
                    end_counter += 1;
                    if == "feed" {
                Event::Characters(_) => {}
                Event::Reference(_) => {}
                _ => {}
            Err(err) => {
                println!("{}", err);

    println!("Start event count:{}", counter);
    println!("End event count:{}", end_counter);

    let elapsed = now.elapsed();
    println!("Time elapsed: {:.2?}", elapsed);

History & Credits

I tried to specify a push parser interface like the Java SAX library and implement it in 2017. The idea was to provide an interface that can have multiple implementations in the community. It was working(albeit slowly) but the main problem was that a push parser is not ergonomic in Rust. After thinking for a long time and learning more about Rust I decided to implement a pull parser. Currently, the SAX(pull) interface is just an enum and its behavior(like the possibility of splitting characters for each call).

If you want to use xml_sax interface to implement another parser we can discuss improving the interface. Currently, it is integrated into this crate.

Why a pull parser? section in pulldown-cmark is a great explanation.

The current interface is inspired by quick-xml, xml-rs, and Java libraries.

nom is a great library. It is just a crystallized & better version of what you would do naively at first try(I know). It also shows the power of composability in Rust.