Summary

Lands the regex foundation for JSONata: /pat/flags literals, a first-class callable regex value, a lazy PCRE2-backed engine adapter, and the $contains/$split regex builtins. Faithful to jsonata-js v2.2.1. ($match/$replace follow in M7b.)

• Lexer (3a877ae): the tokenizer scans /pat/flags (i/m flags) when / appears in operand position (decided from the previous token) and division otherwise; depth-tracking finds the closing /. Empty // → S0301, unterminated → S0302. Parser emits a {type=regex} node.
• Value (b9b1b0f): a new src/jsonata/regex.lua wraps lrexlib-pcre2 — required lazily (i→CASELESS, m→MULTILINE; byte→char offsets). A regex evaluates to a first-class _jsonata_function whose impl returns a {match,start,end,groups,next} object; next() throws D1004 on a zero-width match. ($type(/x/)→"function".)
• Builtins (fc034ee): $contains/$split accept a string or a regex (driven by applying the regex value); string args keep their existing behaviour. Signatures <s-(sf):b> / <s-(sf)n?:a<s>> — also unblocks the (sf) signatures deferred in M5b.
• Dependency: lrexlib-pcre2 (lazy) — the zero-dependency drop-in property holds for everything except actual regex usage (verified: non-regex programs never load PCRE2).

A small Task-2 side-fix (a parenthesized (block) path-head is now self-contained) additionally turned +3 non-regex official cases green.

Results

• Official suite 1251 → 1261 (+10); regex contains/split slice +7, block-path +3.
• 514 unit tests green; zero-regression guard green (additions-only); /-disambiguation, division, and existing string $contains/$split unchanged.
• Adversarial review vs genuine jsonata-js v2.2.1: the core surface (/ disambiguation matrix, $contains/$split regex, regex-value match/start/end/flags/global-iterator, S0301/S0302, joins) is oracle-faithful.

Deferred to M7b (documented; suite-invisible in M7a)

• Single-capture .groups unwrap + cons-array navigation (entangled — the real fix is holistic cons-array spread/indexing, best done when $match makes groups central).
• Lazy invalid-regex error timing (eval-time S0303 vs jsonata's parse-time) — a tradeoff of the lazy PCRE2 design.
• Pre-existing function-= comparison crash ($string = $number crashes identically).
• Minor cleanups: H.serialize skip the match-object next field; hoist is_regex/apply to H; $split O(n²) re-slicing.

Test plan

• busted spec/ — 514/0
• busted spec/jsonata_suite_spec.lua — zero-regression guard green
• bash scripts/run-suite.sh — 1261/1682
• Lazy-require: non-regex programs do not load rex_pcre2
• Adversarial oracle review (jsonata@2.2.1) incl. the / disambiguation matrix

🤖 Generated with Claude Code