Go to file
Andrew Gallant d2ade94657 prefilter: fix rare byte prefilter
This fixes a rather nasty bug that occurred when the rare byte prefilter
computed its shift offset incorrectly. In particular, when a rare byte
is found using a prefilter, we shift backwards in the haystack by the
maximum amount possible before confirming whether a match exists or not.
If this shift is not actually the maximum amount possible, then it's
quite possible that we will miss a match. (N.B. The prefilter
infrastructure takes care to avoid accidentally quadratic behavior.)

The specific regression in this case was caused by searching for these
two patterns:

    ab/j/
    x/

which would erroneously fail to match this haystack

    ab/j/

When prefilters are enabled (the default), this particular search would
use the "rare two byte" prefilter. Specifically, it would detect '/' and
'j' as rare bytes, with '1' as the max offset for '/' and '3' as the max
offset for 'j'. The former is clearly incorrect, since '/' occurs at
offset 4 in the first pattern. This was being incorrectly computed
because we weren't actually looking at all possible bytes in all
patterns and recording their offsets. Once we found a rare byte, we
stopped trying to find more occurrences of it.

We fix this byte now recording the maximum offsets of _all_ bytes for
_all_ patterns given. That way, we're guaranteed to have the correct
maximal shift amount for any rare byte found.

Fixes #53
2020-02-26 20:25:13 -05:00
.github/workflows ci: remove superfluous component install 2020-02-26 20:25:13 -05:00
aho-corasick-debug debug: fix warnings in debug program 2019-08-03 09:47:19 -04:00
bench deps: update to criterion 0.3 2020-02-04 09:01:51 -05:00
src prefilter: fix rare byte prefilter 2020-02-26 20:25:13 -05:00
.gitignore gitignore: add scratch directory 2020-02-04 09:01:51 -05:00
Cargo.toml 0.7.8 2020-02-04 09:02:13 -05:00
COPYING Initial commit. Not finished. 2015-06-11 21:15:43 -04:00
DESIGN.md doc: fix typos in DESIGN.md 2020-02-04 09:01:51 -05:00
LICENSE-MIT Initial commit. Not finished. 2015-06-11 21:15:43 -04:00
README.md ci: switch to GitHub Actions 2020-01-11 12:51:33 -05:00
rustfmt.toml style: switch to rustfmt 2019-08-03 09:47:19 -04:00
UNLICENSE Initial commit. Not finished. 2015-06-11 21:15:43 -04:00

aho-corasick

A library for finding occurrences of many patterns at once with SIMD acceleration in some cases. This library provides multiple pattern search principally through an implementation of the Aho-Corasick algorithm, which builds a finite state machine for executing searches in linear time. Features include case insensitive matching, overlapping matches and search & replace in streams.

Build status

Dual-licensed under MIT or the UNLICENSE.

Documentation

https://docs.rs/aho-corasick

Usage

Add this to your Cargo.toml:

[dependencies]
aho-corasick = "0.7"

and this to your crate root (if you're using Rust 2015):

extern crate aho_corasick;

Example: basic searching

This example shows how to search for occurrences of multiple patterns simultaneously. Each match includes the pattern that matched along with the byte offsets of the match.

use aho_corasick::AhoCorasick;

let patterns = &["apple", "maple", "Snapple"];
let haystack = "Nobody likes maple in their apple flavored Snapple.";

let ac = AhoCorasick::new(patterns);
let mut matches = vec![];
for mat in ac.find_iter(haystack) {
    matches.push((mat.pattern(), mat.start(), mat.end()));
}
assert_eq!(matches, vec![
    (1, 13, 18),
    (0, 28, 33),
    (2, 43, 50),
]);

Example: case insensitivity

This is like the previous example, but matches Snapple case insensitively using AhoCorasickBuilder:

use aho_corasick::AhoCorasickBuilder;

let patterns = &["apple", "maple", "snapple"];
let haystack = "Nobody likes maple in their apple flavored Snapple.";

let ac = AhoCorasickBuilder::new()
    .ascii_case_insensitive(true)
    .build(patterns);
let mut matches = vec![];
for mat in ac.find_iter(haystack) {
    matches.push((mat.pattern(), mat.start(), mat.end()));
}
assert_eq!(matches, vec![
    (1, 13, 18),
    (0, 28, 33),
    (2, 43, 50),
]);

Example: replacing matches in a stream

This example shows how to execute a search and replace on a stream without loading the entire stream into memory first.

use aho_corasick::AhoCorasick;

let patterns = &["fox", "brown", "quick"];
let replace_with = &["sloth", "grey", "slow"];

// In a real example, these might be `std::fs::File`s instead. All you need to
// do is supply a pair of `std::io::Read` and `std::io::Write` implementations.
let rdr = "The quick brown fox.";
let mut wtr = vec![];

let ac = AhoCorasick::new(patterns);
ac.stream_replace_all(rdr.as_bytes(), &mut wtr, replace_with)?;
assert_eq!(b"The slow grey sloth.".to_vec(), wtr);

Example: finding the leftmost first match

In the textbook description of Aho-Corasick, its formulation is typically structured such that it reports all possible matches, even when they overlap with another. In many cases, overlapping matches may not be desired, such as the case of finding all successive non-overlapping matches like you might with a standard regular expression.

Unfortunately the "obvious" way to modify the Aho-Corasick algorithm to do this doesn't always work in the expected way, since it will report matches as soon as they are seen. For example, consider matching the regex Samwise|Sam against the text Samwise. Most regex engines (that are Perl-like, or non-POSIX) will report Samwise as a match, but the standard Aho-Corasick algorithm modified for reporting non-overlapping matches will report Sam.

A novel contribution of this library is the ability to change the match semantics of Aho-Corasick (without additional search time overhead) such that Samwise is reported instead. For example, here's the standard approach:

use aho_corasick::AhoCorasick;

let patterns = &["Samwise", "Sam"];
let haystack = "Samwise";

let ac = AhoCorasick::new(patterns);
let mat = ac.find(haystack).expect("should have a match");
assert_eq!("Sam", &haystack[mat.start()..mat.end()]);

And now here's the leftmost-first version, which matches how a Perl-like regex will work:

use aho_corasick::{AhoCorasickBuilder, MatchKind};

let patterns = &["Samwise", "Sam"];
let haystack = "Samwise";

let ac = AhoCorasickBuilder::new()
    .match_kind(MatchKind::LeftmostFirst)
    .build(patterns);
let mat = ac.find(haystack).expect("should have a match");
assert_eq!("Samwise", &haystack[mat.start()..mat.end()]);

In addition to leftmost-first semantics, this library also supports leftmost-longest semantics, which match the POSIX behavior of a regular expression alternation. See MatchKind in the docs for more details.

Minimum Rust version policy

This crate's minimum supported rustc version is 1.28.0.

The current policy is that the minimum Rust version required to use this crate can be increased in minor version updates. For example, if crate 1.0 requires Rust 1.20.0, then crate 1.0.z for all values of z will also require Rust 1.20.0 or newer. However, crate 1.y for y > 0 may require a newer minimum version of Rust.

In general, this crate will be conservative with respect to the minimum supported version of Rust.

Future work

Here are some plans for the future:

  • Assuming the current API is sufficient, I'd like to commit to it and release a 1.0 version of this crate some time in the next 6-12 months.
  • Support stream searching with leftmost match semantics. Currently, only standard match semantics are supported. Getting this right seems possible, but is tricky since the match state needs to be propagated through multiple searches. (With standard semantics, as soon as a match is seen the search ends.)