Drop-in for the Rust regex crate#

Full drop-in, strict by default — the regex crate’s API on REAL’s linear engine; a pattern REAL cannot run raises Error::Unsupported instead of silently backtracking.

Adopt / swap#

use real_regex::Regex;   // drop-in for the regex crate

let re = Regex::new(r"(\w+)@(\w+)")?;
let caps = re.captures("info@example.com").unwrap();
&caps[2];                // "example"

API offered#

Regex (is_match / find / find_iter / captures / captures_iter / replace / replace_all / split, Display / FromStr), Captures, RegexBuilder, RegexSet (which-matched), and the bytes module — the regex crate surface, same signatures. Not offered: Captures::expand (use replace / replace_all). The opt-in fallback cargo feature adds RegexBuilder::fallback(true): a rejected pattern delegates to the regex crate for that pattern, trading its linear-time guarantee — engine() always says which backend ran (Real / Fallback).

Differences & limitations#

Condensed — the full list with tables is the crate’s own README / docs.rs page:

  • Bounded lookarounds, a positive divergence — (?=…) (?<=a|bb) match in linear time; the regex crate and RE2 reject lookarounds entirely.

  • \p{…} natively — General_Category, Script, Script_Extensions and the standard binary properties run on REAL (engine() = Real); other UAX44 namespaces raise Error::Unsupported. fallback delegates only what the regex crate actually has — \p{Bidi_Class=L} is refused on both sides.

  • CPython word/space/case semantics — \w \s and IGNORECASE folding follow Python re, not UTS#18; class-set syntax ([a[b]], &&) declines.

  • \u{…} is accepted as \x{…} (the regex crate spelling). \U{…} is still eight-digit \U, not braces.

  • \ before a non-ASCII character is that character (\é matches é, [\à-\é] is a code-point range) — Python re’s rule. The crate answers unrecognized escape sequence, the same error it gives \q.

  • Possessive quantifiers, a positive divergence — but silent — REAL reads x?+ / x*+ / x++ as possessive (the Python 3.11+ grammar); the regex crate reads them as nested repetition. Both compile — check the README entry before migrating.

  • shortest_match is leftmost-first (greedy end), not earliest completion.

  • Empty-match iteration follows the regex crate’s rule at the wrapper, so find_iter / split agree with it.

Comparison#

REAL’s closest peer on the linear-time axis. The honest duel, both ISAs, lives in the performance ledger; the reading is Performance.