Character encodings

The parser auto-detects the input’s encoding and transcodes to UTF-8 before parsing — matching libxml2 and the XML 1.0 spec’s requirement (§ 4.3.3) that processors accept both UTF-8 and UTF-16:

use sup_xml::{parse_bytes, ParseOptions};
let doc = parse_bytes(latin1_or_utf16_or_gb2312_bytes, &ParseOptions::default())?;

Detection follows XML 1.0 Appendix F: BOM, then the four-byte autodetect signatures for UTF-32 / UTF-16 / EBCDIC, then the <?xml encoding="..."?> declaration. UTF-8 input stays zero-copy; non-UTF-8 input pays one allocation for the decoded buffer.

Strict UTF-8 mode

use sup_xml::{parse_bytes, ParseOptions};
let opts = ParseOptions { auto_transcode: false, ..Default::default() };
let doc = parse_bytes(must_be_utf8_bytes, &opts)?;

Built-in encodings — no dependency

Hand-tuned, byte-for-byte audited:

• UTF-8, US-ASCII — zero-copy passthrough
• ISO-8859-1 (Latin-1), Windows-1252 — SWAR-accelerated
• UTF-16 LE / BE — BOM-aware, surrogate-pair validated
• UTF-32 LE / BE — BOM-aware (also accepts UCS-4LE / UCS-4BE aliases)
• EBCDIC variants:
  • IBM037 (CCSID 37) — US/Canada Latin
  • IBM500 (CCSID 500) — International EBCDIC
  • IBM1047 (CCSID 1047) — Open Systems / z/OS Unix Services Latin-1
  • IBM1140 (CCSID 1140) — IBM037 with the Euro sign update

Via encoding_rs — full-encodings feature (default on)

The full WHATWG Encoding set: Shift_JIS, EUC-JP, ISO-2022-JP, GB2312, GBK, GB18030, Big5, EUC-KR, ISO-8859-2…16, Windows-1250…1258, KOI8-R, KOI8-U, IBM866, macintosh, x-mac-cyrillic, TIS-620 (via windows-874), and others.

Disable the feature to drop the dependency and reject these inputs with a clean error.

Deliberately unsupported

• UTF-7 — not supported for security reasons (UTF-7 enables MIME header smuggling and XSS in HTML contexts; we won’t ship the decoder).