skip to navigation
skip to content

acora 2.1

Fast multi-keyword search engine for text strings

What is Acora?

Acora is ‘fgrep’ for Python, a fast multi-keyword text search engine.

Based on a set of keywords and the Aho-Corasick algorithm, it generates a search automaton and runs it over string input, either unicode or bytes.

Acora comes with both a pure Python implementation and a fast binary module written in Cython. However, note that the current construction algorithm is not suitable for really large sets of keywords (i.e. more than a couple of thousand).

You can find the latest source code on github.

To report a bug or request new features, use the github bug tracker. Please try to provide a short test case that reproduces the problem without requiring too much experimentation or large amounts of data. The easier it is to reproduce the problem, the easier it is to solve it.

Features

  • works with unicode strings and byte strings
  • about 2-3x as fast as Python’s regular expression engine for most input
  • finds overlapping matches, i.e. all matches of all keywords
  • support for case insensitive search (~10x as fast as ‘re’)
  • frees the GIL while searching
  • additional (slow but short) pure Python implementation
  • support for Python 2.5+ and 3.x
  • support for searching in files
  • permissive BSD license

How do I use it?

Import the package:

>>> from acora import AcoraBuilder

Collect some keywords:

>>> builder = AcoraBuilder('ab', 'bc', 'de')
>>> builder.add('a', 'b')

Or:

>>> builder.update(['a', 'b'])  # new in version 2.0

Generate the Acora search engine for the current keyword set:

>>> ac = builder.build()

Search a string for all occurrences:

>>> ac.findall('abc')
[('a', 0), ('ab', 0), ('b', 1), ('bc', 1)]
>>> ac.findall('abde')
[('a', 0), ('ab', 0), ('b', 1), ('de', 2)]

Iterate over the search results as they come in:

>>> for kw, pos in ac.finditer('abde'):
...     print("%2s[%d]" % (kw, pos))
 a[0]
ab[0]
 b[1]
de[2]

Acora also has direct support for parsing files (in binary mode):

>>> keywords = ['Import', 'FAQ', 'Acora', 'NotHere'.upper()]

>>> builder = AcoraBuilder([s.encode('ascii') for s in keywords])
>>> ac = builder.build()

>>> found = set(kw for kw, pos in ac.filefind('README.rst'))
>>> len(found)
3

>>> sorted(str(s.decode('ascii')) for s in found)
['Acora', 'FAQ', 'Import']

FAQs and recipes

  1. How do I run a greedy search for the longest matching keywords?

    >>> builder = AcoraBuilder('a', 'ab', 'abc')
    >>> ac = builder.build()
    
    >>> for kw, pos in ac.finditer('abbabc'):
    ...     print(kw)
    a
    ab
    a
    ab
    abc
    
    >>> from itertools import groupby
    >>> from operator import itemgetter
    
    >>> def longest_match(matches):
    ...     for pos, match_set in groupby(matches, itemgetter(1)):
    ...         yield max(match_set)
    
    >>> for kw, pos in longest_match(ac.finditer('abbabc')):
    ...     print(kw)
    ab
    abc
    

    Note that this recipe assumes search terms that do not have inner overlaps apart from their prefix.

  2. How do I parse line-by-line with arbitrary line endings?

    >>> def group_by_lines(s, *keywords):
    ...     builder = AcoraBuilder('\r', '\n', *keywords)
    ...     ac = builder.build()
    ...
    ...     current_line_matches = []
    ...     last_ending = None
    ...
    ...     for kw, pos in ac.finditer(s):
    ...         if kw in '\r\n':
    ...             if last_ending == '\r' and kw == '\n':
    ...                 continue # combined CRLF
    ...             yield tuple(current_line_matches)
    ...             del current_line_matches[:]
    ...             last_ending = kw
    ...         else:
    ...             last_ending = None
    ...             current_line_matches.append(kw)
    ...     yield tuple(current_line_matches)
    
    >>> kwds = ['ab', 'bc', 'de']
    >>> for matches in group_by_lines('a\r\r\nbc\r\ndede\n\nab', *kwds):
    ...     print(matches)
    ()
    ()
    ('bc',)
    ('de', 'de')
    ()
    ('ab',)
    
  3. How do I find whole lines that contain keywords, as fgrep does?

    >>> def match_lines(s, *keywords):
    ...     builder = AcoraBuilder('\r', '\n', *keywords)
    ...     ac = builder.build()
    ...
    ...     line_start = 0
    ...     matches = False
    ...     for kw, pos in ac.finditer(s):
    ...         if kw in '\r\n':
    ...             if matches:
    ...                  yield s[line_start:pos]
    ...                  matches = False
    ...             line_start = pos + 1
    ...         else:
    ...             matches = True
    ...     if matches:
    ...         yield s[line_start:]
    
    >>> kwds = ['x', 'de', '\nstart']
    >>> text = 'a line with\r\r\nsome text\r\ndede\n\nab\n start 1\nstart\n'
    >>> for line in match_lines(text, *kwds):
    ...     print(line)
    some text
    dede
    start
    

Changelog

  • 2.1 [2017-12-15]
    • fix handling of empty engines (Github issue #18)
  • 2.0 [2016-03-17]
    • rewrite of the construction algorithm to speed it up and save memory
  • 1.9 [2015-10-10]
    • recompiled with Cython 0.23.4 for better compatibility with recent Python versions.
  • 1.8 [2014-02-12]
    • pickle support for the pre-built search engines
    • performance optimisations in builder
    • Unicode parsing is optimised for Python 3.3 and later
    • no longer recompiles sources when Cython is installed, unless --with-cython option is passed to setup.py (requires Cython 0.20+)
    • build failed with recent Cython versions
    • built using Cython 0.20.1
  • 1.7 [2011-08-24]
    • searching binary strings for byte values > 127 was broken
    • built using Cython 0.15+
  • 1.6 [2011-07-24]
    • substantially faster automaton building
    • no longer includes .hg repo in source distribution
    • built using Cython 0.15 (rc0)
  • 1.5 [2011-01-24]
    • Cython compiled NFA-2-DFA construction runs substantially faster
    • always build extension modules even if Cython is not installed
    • --no-compile switch in setup.py to prevent extension module building
    • built using Cython 0.14.1 (rc2)
  • 1.4 [2009-02-10]
    • minor speed-up in inner search engine loop
    • some code cleanup
    • built using Cython 0.12.1 (final)
  • 1.3 [2009-01-30]
    • major fix for file search
    • built using Cython 0.12.1 (beta0)
  • 1.2 [2009-01-30]
    • deep-copy support for AcoraBuilder class
    • doc/test fixes
    • include .hg repo in source distribution
    • built using Cython 0.12.1 (beta0)
  • 1.1 [2009-01-29]
    • doc updates
    • some cleanup
    • built using Cython 0.12.1 (beta0)
  • 1.0 [2009-01-29]
    • initial release
 
File Type Py Version Uploaded on Size
acora-2.1-cp26-cp26m-manylinux1_i686.whl (md5, pgp) Python Wheel cp26 2017-12-15 146KB
acora-2.1-cp26-cp26m-manylinux1_x86_64.whl (md5, pgp) Python Wheel cp26 2017-12-15 173KB
acora-2.1-cp26-cp26mu-manylinux1_i686.whl (md5, pgp) Python Wheel cp26 2017-12-15 146KB
acora-2.1-cp26-cp26mu-manylinux1_x86_64.whl (md5, pgp) Python Wheel cp26 2017-12-15 171KB
acora-2.1-cp27-cp27m-manylinux1_i686.whl (md5, pgp) Python Wheel cp27 2017-12-15 148KB
acora-2.1-cp27-cp27m-manylinux1_x86_64.whl (md5, pgp) Python Wheel cp27 2017-12-15 173KB
acora-2.1-cp27-cp27mu-manylinux1_i686.whl (md5, pgp) Python Wheel cp27 2017-12-15 147KB
acora-2.1-cp27-cp27mu-manylinux1_x86_64.whl (md5, pgp) Python Wheel cp27 2017-12-15 172KB
acora-2.1-cp33-cp33m-manylinux1_i686.whl (md5, pgp) Python Wheel cp33 2017-12-15 148KB
acora-2.1-cp33-cp33m-manylinux1_x86_64.whl (md5, pgp) Python Wheel cp33 2017-12-15 174KB
acora-2.1-cp34-cp34m-manylinux1_i686.whl (md5, pgp) Python Wheel cp34 2017-12-15 149KB
acora-2.1-cp34-cp34m-manylinux1_x86_64.whl (md5, pgp) Python Wheel cp34 2017-12-15 174KB
acora-2.1-cp35-cp35m-manylinux1_i686.whl (md5, pgp) Python Wheel cp35 2017-12-15 150KB
acora-2.1-cp35-cp35m-manylinux1_x86_64.whl (md5, pgp) Python Wheel cp35 2017-12-15 174KB
acora-2.1-cp36-cp36m-manylinux1_i686.whl (md5, pgp) Python Wheel cp36 2017-12-15 152KB
acora-2.1-cp36-cp36m-manylinux1_x86_64.whl (md5, pgp) Python Wheel cp36 2017-12-15 176KB
acora-2.1.tar.gz (md5, pgp) Source 2017-12-15 191KB