oxixml-encoding

Character-encoding decoders and encoders for the OxiXML workspace — the COOLJAPAN Pure-Rust foundation for XML and RDF processing. It plays the role that encoding_rs plays in other Rust XML stacks, written from scratch against std alone: incremental, chunk-safe decoding (a partial multi-byte sequence at a chunk boundary is carried over correctly), the inverse encoding direction, byte-order-mark detection, and the XML 1.0 Appendix F encoding-sniffing heuristic for <?xml ... encoding="..."?> prologues.

Install

[dependencies]
oxixml-encoding = "0.1.0"

Usage

use oxixml_encoding::{sniff_xml_encoding, Encoding};

// A UTF-8 document that self-declares its own encoding.
let xml = b"<?xml version=\"1.0\" encoding=\"UTF-8\"?><root/>";
let sniffed = sniff_xml_encoding(xml);
assert_eq!(sniffed.declared_encoding, Some(Encoding::Utf8));

// Decode a chunk of UTF-8 incrementally, one byte at a time, and get back the original text.
let mut decoder = Encoding::Utf8.new_decoder();
let mut out = String::new();
for (i, byte) in "héllo".as_bytes().iter().enumerate() {
    let last = i + 1 == "héllo".len();
    decoder
        .decode_to_string(std::slice::from_ref(byte), &mut out, last)
        .expect("valid UTF-8 fed one byte at a time");
}
assert_eq!(out, "héllo");

Supported encodings

UTF-8, UTF-16 (both byte orders), ISO-8859-1/windows-1252, US-ASCII, the rest of the WHATWG Encoding Standard's legacy single-byte encodings (the full ISO-8859-*, windows-125x, KOI8-*, IBM866, macintosh, x-mac-cyrillic families), and the three legacy Japanese multi-byte encodings (Shift_JIS, EUC-JP, ISO-2022-JP) prioritized for COOLJAPAN. UTF-32 and EBCDIC prologues are recognized by sniff_xml_encoding but are out of scope for decoding.

What it replaces

A from-scratch, dependency-free alternative to encoding_rs sized for what an XML processor actually needs, plus the XML-specific BOM detection and prologue-sniffing that encoding_rs itself does not provide.

Testing

Every supported encoding is round-tripped through a "chunk-boundary torture test" — encode, then feed the bytes back through an incremental decoder one, two, and three bytes at a time — and the three Japanese multi-byte encodings have dedicated coverage for their lead/trail byte error ranges and escape-sequence state machine.

License

Licensed under the Apache License, Version 2.0 (LICENSE or http://www.apache.org/licenses/LICENSE-2.0).