SupXML

A memory-safe, fast, spec-compliant XML library for Rust, with a drop-in C ABI replacement for libxml2.

Why SupXML

Memory-safe

Pure Rust, memory-safe by construction. Roughly 70% of CVEs come from memory-safety bugs, a whole class that Rust rules out at compile time, so they can’t happen in SupXML.

Fast and Efficient

~2× faster than libxml2 on full-validation DOM parse (median across 21 fixtures), and ~2.4× faster on the W3C XSD 1.0 test suite. Bumpalo-backed arena DOM.
Full bench numbers →

Spec-compliant

Zero failures on the W3C XML Conformance Test Suite: all 2274 deterministic cases pass; the other 21 of 2295 are implementation-defined by the spec. 98.9% schema / 98.8% instance on the W3C XSD 1.0 suite (libxml2: 92.9% / 98.3%).
Cross-parser comparison →

Stream big files

XmlByteStreamReader pulls from any io::Read through a rolling buffer, processing files larger than memory in bounded RAM. The in-memory XmlBytesReader is a zero-copy SAX reader, median ~1.04× faster than quick-xml at the matched-contract comparison.

Drop-in for libxml2

Byte-compatible C ABI matches libsupxml2.so. Once a consumer dynamically links libxml2, swapping the load command points it at SupXML.
Per-binding setup →

Full-featured

XPath 1.0 + 2.0, XSD 1.0 / 1.1, XSLT 1.0 + 2.0 (3.0 partial), Schematron, Canonical XML / Exc-C14N, HTML5, EXSLT, recovery mode… all in one library.

W3C XML Conformance Test Suite

The W3C XML Conformance Test Suite (revision xmlts20130923) is the canonical test catalog for XML 1.0 parsers. It defines 2295 tests across submissions from James Clark, Sun, IBM, OASIS, and others. 2274 of them have a deterministic expected outcome (well-formed / not-well-formed / invalid) and SupXML matches every one with zero failures. The remaining 21 are tagged error in the catalog itself, meaning XML 1.0 explicitly leaves their handling implementation-defined, both accepting and rejecting them satisfies the spec, so whatever SupXML does is conformant by definition. Full breakdown →

How parsers compare on malformed input

For a like-for-like comparison across parsers we use the catalog’s not-wf corpus, files engineered to violate one specific XML 1.0 well-formedness rule, so a conforming parser must reject them. (We focus on not-wf because xml-rs and quick-xml don’t load external DTDs, making fair scoring on the valid/invalid corpora difficult.) Score = percentage correctly rejected:

Corpus	Files	SupXML	libxml2	xml-rs	quick-xml
xmltest (James Clark)	200	99.0%	97.0%	58.5%	10.0%
Sun Microsystems	57	100%	98.2%	33.3%	8.8%
IBM (incl. XML 1.1)	890	94.5%	59.6%	42.7%	5.2%
All vendors	1147	95.6%	68.0%	45.0%	6.2%

This table walks every .xml file on disk under a not-wf/ directory, which is a superset of what the official catalog scores (it also includes files the catalog marks error or scopes to a specific XML 1.0 edition). That’s why SupXML reads clean on the catalog (2274/2274 deterministic) but 94.5% on IBM here because the bench is asking a broader question, and counts implementation-defined fixtures against the parser even though the catalog allows either outcome.

quick-xml’s score is low because it doesn’t check well-formedness at all. It’s a fast tokenizer but unsafe to use in practice. SupXML beats libxml2 by ~27 points overall.

Quick example

use sup_xml::{parse_str, ParseOptions, XPathContext};

let opts = ParseOptions { namespace_aware: true, ..Default::default() };
let doc = parse_str(
    "<catalog><book id='b1'/><book id='b2'/></catalog>",
    &opts,
)?;

let ctx = XPathContext::new(&doc);
assert_eq!(ctx.eval_count("/catalog/book")?, 2);

[dependencies]
sup-xml = { version = "*", features = ["xsd", "xslt", "html"] }

Feature matrix

Feature	Cargo feature	Entry point
XML 1.0 parse / serialize	(default)	`parse_str`, `parse_bytes`, `serialize_to_string`
XPath 1.0 (default) + XPath 2.0 (opt-in)	(default)	`XPathContext`, `XPathOptions { xpath_2_0: true }`
HTML5 parse	`html`	`parse_html_str`
XSD 1.0 / 1.1 validation	`xsd`	`sup_xml::xsd::Schema`
XSLT 1.0 + 2.0 (3.0 partial)	`xslt`	`sup_xml::xslt::Stylesheet`
Schematron validation	`xslt`	`sup_xml::xslt::schematron::Schematron`
Canonical XML / Exc-C14N	(default)	`canonicalize_to_bytes`
Typed-struct deserialize	`serde`	`sup_xml::de::*`
HTTPS-fetched DTDs / entities	`network-resolver`	`NetworkResolver`
Async I/O entry points	`tokio`	`sup_xml::async_io::parse_async`