RE#

A high-performance, automata-based regex engine with first-class support for intersection (&), complement (~). Non-backtracking with linear-time matching. Built for complex patterns (large alternations, lookarounds, boolean combinations) that make traditional engines degrade or fall back to slower paths.

paper | blog post | syntax docs | dotnet version and web playground

Quick start

cargo add resharp

// 8+ alphanumeric & contains digit & contains uppercase
let re = resharp::Regex::new(r"[A-Za-z0-9]{8,}&_*[0-9]_*&_*[A-Z]_*").unwrap();

let found = re.is_match(b"Hunter2024").unwrap();
let matches = re.find_all(b"try Hunter2024 or password1").unwrap();

When to use RE# over regex

RE# operates on &[u8] / UTF-8 and aims to match regex crate throughput on standard patterns. Use RE# when you need:

• intersection (&), complement (~), or lookarounds
• large alternations with high throughput (at the cost of memory)
• fail-loud behavior: capacity / lookahead overflow returns Err instead of silently degrading

RE# is designed around is_match and find_all. It doesn't provide find or captures, but for simple cases you can often substitute find_anchored, or emulate a capture group with lookarounds. For example, a(b)c becomes (?<=a)b(?=c). For anything more involved, use the regex crate instead.

Syntax extensions

RE# supports standard regex syntax plus three extensions: _ (any byte), & (intersection), and ~(...) (complement). _* means "any string".

_*                any string
a_*               any string that starts with 'a'
_*a               any string that ends with 'a'
_*a_*             any string that contains 'a'
~(_*a_*)          any string that does NOT contain 'a'
(_*a_*)&~(_*b_*)  contains 'a' AND does not contain 'b'
(?<=b)_*&_*(?=a)  preceded by 'b' AND followed by 'a'

You combine all of these with & to get more complex patterns. RE# also supports lookarounds ((?=...), (?<=...), (?!...), (?<!...)), compiled directly into the automaton with no backtracking.

Differences from PCRE / regex

• Leftmost-longest, not leftmost-greedy. y|yes on "yes" matches yes. Branch order is irrelevant.
• Multiline on by default. ^/$ match start/end of line; disable with (?-m). \A/\z always anchor to input.
• \w defaults to 2-byte UTF-8. See UnicodeMode.

Lazy quantifiers (*?, +?, ...) are parse errors; rewrite with complement when possible: <div>.*?</div> -> <div>~(_*</div>_*)</div>. See syntax.md for the rest.

Configuration

let opts = resharp::RegexOptions {
    max_dfa_capacity: 65535,    // max automata states (default: u16::MAX)
    lookahead_context_max: 800, // max lookahead context distance (default: 800)
    hardened: false,            // linear find_all worst-case (slower but safer)
    unicode: resharp::UnicodeMode::Default, // Ascii | Default | Full | Javascript
    ..Default::default()
};
let re = resharp::Regex::with_options(r"pattern", opts).unwrap();

Benchmarks

RE# against regex, fancy-regex, and PCRE2 on a few popular patterns from crates.io. Regenerate with:

node scripts/bench-popular-table.mts

resharp runs with UnicodeMode::Full and multiline(false) to match the other engines. Ratios are vs the fastest per row.

Scan (find_all over a 1 MiB haystack), throughput

Validate (is_match on a single value), latency