Alfred–Margaret

Alfred–Margaret is a fast implementation of the Aho–Corasick string searching algorithm in Haskell. It powers many string-related operations in Channable.

The library is designed to work with the text package. It matches directly on the internal UTF-16 representation of Text for efficiency. See the announcement blog post for a deeper dive into Aho–Corasick, and the optimizations that make this library fast.

Alfred–Margaret is named after Alfred Aho and Margaret Corasick.

Performance

Running time to count all matches, in a real-world data set, comparing a Java implementation and a Rust implementation against Alfred–Margaret, and against memcopy to establish a lower bound:

For the full details of this benchmark, see our announcement blog post, which includes more details about the data set, the benchmark setup, and a few things to keep in mind when interpreting this graph.

LLVM

If you are using LLVM instead of the GHC backend, make sure to compare different versions. Using LLVM 9 instead of LLVM 12 with GHC 8.10.7 made a significant dent in benchmark times. For more information, see this issue.

Example

Check if a string contains one of the needles:

import qualified Data.Text.AhoCorasick.Automaton as Aho
import qualified Data.Text.AhoCorasick.Searcher as Searcher

searcher = Searcher.build Aho.CaseSensitive ["tshirt", "shirts", "shorts"]

Searcher.containsAny searcher "short tshirts"
> True

Searcher.containsAny searcher "long shirt"
> False

Searcher.containsAny searcher "Short TSHIRTS"
> False

searcher' = Searcher.build Aho.IgnoreCase ["tshirt", "shirts", "shorts"]

Searcher.containsAny searcher' "Short TSHIRTS"
> True

Sequentially replace many needles:

import Data.Text.AhoCorasick.Automaton (CaseSensitivity (..))
import qualified Data.Text.AhoCorasick.Replacer as Replacer

replacer = Replacer.build CaseSensitive [("tshirt", "banana"), ("shirt", "pear")]

Replacer.run replacer "tshirts for sale"
> "bananas for sale"

Replacer.run replacer "tshirts and shirts for sale"
> "bananas and pears for sale"

Replacer.run replacer "sweatshirts and shirtshirts"
> "sweabananas and shirbananas"

Replacer.run replacer "sweatshirts and shirttshirts"
> "sweabananas and pearbananas"

Get all matches, possibly overlapping:

import qualified Data.Text.AhoCorasick.Automaton as Aho

pairNeedleWithSelf text = (Aho.unpackUtf16 text, text)
automaton = Aho.build $ fmap pairNeedleWithSelf ["tshirt", "shirts", "shorts"]
allMatches = Aho.runText [] (\matches match -> Aho.Step (match : matches))

allMatches automaton "short tshirts"
> [ Match {matchPos = CodeUnitIndex 13, matchValue = "shirts"}
> , Match {matchPos = CodeUnitIndex 12, matchValue = "tshirt"}
> ]

allMatches automaton "sweatshirts and shirtshirts"
> [ Match {matchPos = CodeUnitIndex 27, matchValue = "shirts"}
> , Match {matchPos = CodeUnitIndex 26, matchValue = "tshirt"}
> , Match {matchPos = CodeUnitIndex 22, matchValue = "shirts"}
> , Match {matchPos = CodeUnitIndex 11, matchValue = "shirts"}
> , Match {matchPos = CodeUnitIndex 10, matchValue = "tshirt"}
> ]