serp-re

A re drop-in with true backtracking semantics, written in the Serpentine subset. Patterns compile to a small instruction program run by a recursive backtracking VM: alternation is first-alternative-wins, quantifiers are greedy/lazy with full backtracking into repeated bodies, and capture groups record the last iteration — CPython semantics, verified by 30,000+ differential checks against the real re module (including a 3k random pattern fuzz).

API

compile(pattern, flags=0) -> Pattern, match, fullmatch, search (returning Match | None), findall, finditer (eager list[Match]), sub, split, escape, the error exception (CPython 3.11 messages), and flags IGNORECASE/I, MULTILINE/M, DOTALL/S (combine with + — the subset has no | on ints). Match supports group(n)/groups()/start/end/ span/string and prints as <re.Match object; span=(0, 3), match='abc'>.

Supported syntax

Literals, ., escapes (\d \D \w \W \s \S \b \B \A \Z, punctuation, \n \t \r \f \v), classes [abc] / [a-z] / [^...], quantifiers * + ? {m} {m,} {,n} {m,n} plus lazy variants, groups (...), (?:...), (?P<name>...) (positional access), alternation |, anchors ^ $.

Caps (loud errors or documented divergences)

• No backreferences or lookarounds (raise error).
• \w/\s classification is ASCII.
• findall with 2+ capture groups raises (use finditer).
• groups() returns a PyVal list, not a tuple; split() fills unmatched group parts with '' instead of None.
• sub replacement is a string template (\1, \g<n>, escapes) — no callables. finditer is eager. {m,n} capped at 1000.
• The engine recurses while backtracking; very long quantified matches can exhaust the stack.

Install

serp add serp-re
pip install serp-re