0002 — Use regress for ECMAScript-compatible regular expressions

Status: Accepted

Context

JSONata regex literals (/ab+c/i) and the functions $match, $replace, $split and $contains follow JavaScript regular-expression semantics. The suite relies on features such as backreferences, lookahead and lookbehind, and on JavaScript’s capture-group behaviour.

The regex crate deliberately omits backreferences and lookaround, because it guarantees linear-time matching. fancy-regex adds them, but its syntax and semantics are Perl/PCRE-like rather than ECMAScript.

Decision

Use regress. It implements ECMAScript regular-expression syntax and semantics in pure Rust, and it compiles to wasm32-unknown-unknown.

Consequences

• We get regex behaviour that matches jsonata-js closely, including in the browser build.
• regress is a backtracking engine, so a pathological pattern can take exponential time. The evaluation timeout guards against this (see Options::timeout), and the guides will warn about untrusted patterns. (It did not: a match ran as one evaluation step. ADR 0009 makes the timeout reach inside a match.)
• Positions reported by $match must be converted from UTF-8 byte offsets to the UTF-16 code unit indices that JavaScript reports. That conversion is covered by dedicated tests.