A friendly regular expression dialect for text analytics. Typical regex features are removed/adjusted to make natural language queries easier. Unicode-aware and able to search a stream with several patterns at once.

cargo add reggy

Use the high-level Pattern struct for simple search.

let mut p = Pattern::new("dogs?").unwrap();
assert_eq!(
    p.findall("cat dog dogs cats"),
    vec![(4, 7), (8, 12)]
);

Use the Ast struct to transpile to normal regex syntax.¹

let ast = Ast::parse(r"do(gg.)?|(!CAT|CAR FAR)").unwrap();
assert_eq!(
    ast.to_regex(),
    r"(?mi:do(?:gg\.)?|(?-i:CAT|CAR FAR))"
);

Use the Search struct to search a stream with several patterns at once.

let mut search = Search::compile(&[
    r"$#?#?#.##",
    r"(John|Jane) Doe",
]).unwrap();

Call Search::next to begin searching. It will yield any matches deemed definitely-complete immediately.

let jane_match = Match::new(1, (0, 8));
assert_eq!(
    search.next("Jane Doe paid John"),
    vec![jane_match]
);

Call Search::next again to continue with the same search state. Note that "John Doe" matched across the next boundary, and spans are relative to the start of the stream.

let john_match = Match::new(1, (14, 22));
let money_match_1 = Match::new(0, (23, 29));
let money_match_2 = Match::new(0, (41, 48));
assert_eq!(
    search.next(" Doe $45.66 instead of $499.00"),
    vec![john_match, money_match_1, money_match_2]
);

Call Search::finish to collect any not-definitely-complete matches once the stream is closed.

assert_eq!(search.finish(), vec![]);

See more in the API docs.

Reggy is case-insensitive by default. Spaces match any amount of whitespace (i.e. \s+). All the reserved characters mentioned below (\, (, ), ?, |, #, and !) may be escaped with a backslash for a literal match. Patterns are surrounded by implicit unicode word boundaries (i.e. \b). Empty patterns or subpatterns are not permitted.

Make a character optional with ?

dogs? matches dog and dogs

Create two or more alternatives with |

dog|cat matches dog and cat

Create a sub-pattern with (...)

the qualit(y|ies) required matches the quality required and the qualities required

the only( one)? around matches the only around and the only one around

Create a case-sensitive sub-pattern with (!...)

United States of America|(!USA) matches USA, not usa

Match digits with #

#.## matches 3.14

Reggy follows greedy matching semantics. A pattern may match after one step of a stream, yet may match a longer form depending on the next step. For example, ab|abb will match s.next("ab"), but a subsequent call to s.next("b") would create a longer match, "abb", which should supercede the match "ab".

Search only yields matches once they are definitely complete and cannot be superceded by future next calls. Each pattern has a maximum byte length L (this is why unbound quantifiers are absent from reggy). Once reggy has streamed at most L bytes (counting contiguous whitespace as 1 byte), past the start of a match without superceding it, that match will be yielded. Matches may be yielded earlier if the DFA reaches a dead state.

As a consequence, the Matches returned by a given Search are the same regardless of how a given haystack is segmented. Search::next returns Matches as soon as it practically can while respecting this invariant.

The pattern language is parsed with lalrpop (grammar).

The search routines use a regex_automata::dense::DFA. Compared to other regex engines, the dense DFA is memory-intensive and slow to construct, but searches are fast. Unicode word boundaries are handled by the unicode_segmentation crate.