Skip to main content

Pluggable fulltext indexing solution for Zope and Plone

Project description

What is TextIndexNG V3?

TXNG 3 is the reimplementation of the well-known TextIndexNG product for Zope 2 using Zope 3 technologies. The current implementation runs out-of-the-box on Zope 2 (in combination with Five). The core implementation can be re-used easily in Zope 3.


  • multiple-field indexing (e.g. you can create one index to index content by title, author and body and perform queries against each field)

  • multi-lingual support (Products.LinguaPlone and

  • pluggable components (storages, lexicons, query parsers, splitters, stopwords, normalizers)

  • complex queries (AND, OR, NOT, phrase search, right truncation, wildcards, similarity search)

  • indexes foreign formats (DOC, PDF, XML, SGML, PPT etc.)

  • optional query autoexpansion to improve search results


  • Zope 2.10+

  • for running TextIndexNG3 with Plone, you need Plone 3.1 or higher

Download and project area

TextIndexNG 3 is currently hosted on Github:


Bugs, support issues are handled for free soley either through the TextIndexNG 3 bugtracker on Github:

Dedicated commercial support is available on a per-hour or per-issue basis from

The latest source code is available from Github


TextIndexNG 3 is published under the Zope Public License V 2.1 (see ZPL.txt) Other license agreements can be made. Contact us for details (

TextIndexNG 3 contains copies of ZCTextIndex/ and ZCTextIndex/ which are published under the Zope Public License ZPL.

TextIndexNG 3 ships with a copy of the Snowball code ( for implementing stemming. This code is (C) 2001, Dr. Martin Porter and published under the BSD license.

TextIndexNG3 ships with a modified version of PLY, (C) 2001, David M. Beazley. PLY is published under the GNU Lesser Public License (LGPL).

TextIndexNG3 ships with the python-levenshtein extension written by David Necas und published under the GNU Public License (GPL).


Many thanks to Yvo Schubbe for contributing a lot of code for the 3.2.0 release. Thanks to Christian Zagrodnik for working on the 3.4.0 release.


Third-party contributions that become part of the TextIndexNG3 core (means they are being checked into the TextIndexNG3 source repository) must be made under the same license as the source (ZPL 2.1). Contributed code is subject to be relicensed without further notice of the orginal contributor.


Installation with zc.buildout

  • starting TextIndexNG3 3.3.0 the only supported and recommended way to install TextIndexNG3 using zc.buildout. Add:

    eggs = Products.TextIndexNG3

    to your buildout.cfg file.

    Compiling and installing the extension modules requires a C compiler (usually GCC on Unix systems works perfectly). On Windows systems you need Visual C++ to compile the extension modules.

Installation on Plone

  • follow the steps above

  • go to “Plone setup” -> “Add/remove programs”

  • choose TextIndexNG3 to be added as new product

  • a new configlet for TextIndexNG3 will appear on the setup screen (left side)

  • click on the configlet and choose the only option to replace the existing index setup with TextIndexNG3 indexes

  • that’s it

External converters

To convert foreign formats like .doc, .pdf etc. you need to install some external converters. See

for details.

How to make your custom content-types searchable

Most current Zope index implementations are built on the fact that an index with id XX tries to lookup the indexable content either from an objects XX attribute or by calling the method XX() of the object. Although TextIndexNG V3 still supports this behaviour, the recommended way to make custom types indexable through TXNG3 is through providing dedicated methods that return indexable content. The API of these methods is defined in src/textindexng/interfaces/ Custom types must either implement the IIndexableContent API directly or provide the interface through an adapter registered through ZCML. The IndexContentCollector class should be used to return indexable content either as unicode string or as binary stream (to be transformed through external converters). Some example how to use the indexing API can be found in src/textindexng/tests/ (see classes Mock, MockPDF and StupidMockAdapter)

How to query the index?

TextIndexNG accepts multiple query options that influence the search results (options passed to the search() method):

  • query - the search query (see below). Warning: the search query must always be a Python unicode string

  • parser - id of a registered parser (default: txng.parsers.english)

  • language - one of the languages registered for a particular index (default: the first registered language)

  • field - perform query against this field (as registered with the index)

  • similarity_ratio - a float value between 0.0 and 1.0 to determine the ratio for measuring the similarity of terms based on the Levenshtein distance (default: 0.75)

  • autoexpand - ‘off’|’always’|’on_miss’ (default: ‘off’) determines how query terms are treated. ‘on_miss’ expands the query terms to all terms that are similiar to the original search term (if could not be found). ‘always’ expands the terms always. ‘off’ turns off auto-expansion. Auto-expansion helps you to improve the search result e.g. for mis-spelled words. Using auto-expansion might slow down the query performance.

  • ranking - 0|1|True|False enables/disables (default 0|False) relevance ranking based on the cosine measure. Using ‘ranking’ requires that the index uses storage(s) that implement IStorageWithTermFrequency (txng.storages.term_frequencies). By default an index does not use this storage (see ‘storage’ parameter of the index constructor).

  • ranking_maxhits - the maximum number of documents obtained from the ranking (default 50). ‘ranking’ must be set to True to use this option otherwise an exception will be raised.


TextIndexNG comes with five query parsers. Each of them implements a different query syntax:

  • txng.parsers.en - implements the query syntax as described below (this is the default query parser)

  • - same as txng.parsers.en but it uses ‘UND’, ‘ODER’ and ‘NICHT’ instead of ‘AND’, ‘OR’ and ‘NOT’ (german parser)

  • - same as txng.parsers.en but it uses ‘ET’, ‘OUT’ and ‘PAS’ instead of ‘AND’, ‘OR’ and ‘NOT’ (french parser)

  • txng.parser.dumb_and - this is a very simple parser that accepts only a whitespace separated list of terms which are all combined using AND. No fancy query options as with the parsers above are allowed.

  • txng.parser.dumb_or - this is a very simple parser that accepts only a whitespace separated list of terms which are combined using OR. No fancy query options as with the parsers above are allowed.


Wikipedia defines:

A stemmer is a program or algorithm which determines the
morphological root of a given inflected (or, sometimes, derived)
word form -- generally a written word form.

TextIndexNG V3 includes the Snowball stemmer library written by Martin Porter ( which provides stemming support for eleven languages. Stemming is an optional feature and must be specified when you create a new index. But note that stemming is incompatible by design with a number of TextIndexNG’s features. In general all features except searching for words without wildcards, left/right truncation won’t work and raise an exception.


A thesaurus maps a query term to a sequence of related terms that will be used for searching. Therefore a thesaurus only affects searching but not indexing. Thesaurus are configured as named utitities implementing IThesaurus. The name of a configured thesaurus is by convention ‘txng.thesaurus.XXX’ where XXX is a country code. Multiple thesauruses for one language can be configured under different names. To tell TextIndexNG3 to use a thesaurus while searching you must use the name of a thesaurus or a list of thesaurus names as ‘thesarus’ parameter to the search() method.

Example:, thesaurus=('', ''), ..)

In this case TextIndexNG will use all query terms from 'some_query' and all related
terms from a lookup in the configured thesauruses *.de and *.de-special.


Using a thesaurus is not compatible with phrases searches. Terms appearing within a phrase search will never be used for a thesaurus lookup.

Running the unittests:

bin/zopectl test -s Products.TextIndexNG3

Query syntax for txng.parsers.en parser

  • AND search: word1 AND word2

  • OR search: word1 OR word2

  • PHRASE search: “The Zope Book”

  • NOT search: word1 NOT word2 (searches for - all documents containing ‘word1’ but not

  • Similarity search: %word (All words similiar to ‘word’. The similarity is measured based on the Levenshtein distance of two terms.)

  • Right truncation:

    use the ‘*’ operator at the end of a prefix to search for all words that start with this prefix:

    `foo*`   matches 'foo', 'foobar', 'foofoo', etc.
  • Left truncation

    use the * operator at the beginning of word to search for all words that end with this suffix:

    *bar matches ‘foobar’, ‘bar’, ‘abar’, etc.

  • Wildcard search

    use ‘?’ or ‘*’ within a term

  • Range search: WORD1..WORD2 to search for all words words between WORD1 and WORD2 where WORD1 <= word <= WORD2 (lexicographical ordering)

  • Combining queries (by example)

    • word1 and (word2 or word3)

    • word1 and (word2 word3) - a missing operator implies AND search

    • word1 and “this is a phrase” search for the phrase AND word1

    • (word1 or word2) and (word3 or word4)

  • multi-field queries

    Searching over multiple fields is supported using the following notation:

    `FIELDNAME::OPERATOR(term1 term2 ...)`

    where ‘FIELDNAME’ is the id of a field as configured for the index. ‘OPERATOR’ is either ‘phrase’, ‘near’, ‘and’ or ‘not’ (or the uppercase variant). ‘termX’ is either a word or a word with a modifier (truncation, wildcard search, similarity).


    title::phrase(The Zope Book)

    author::and(michel pelletier amos lattmeier)

    title::phrase(The Zope Book) AND author::and(michel pelletier amos lattmeier)

    title::phrase(The Zope Book) OR author::and(michel pelletier amos lattmeier)

Query constraints

  • a word is a sequence of characters that does not contain a whitespace

  • all queries must be passed as Python unicode string (not UTF-8 string). If a query is not unicode it will be converted to unicode using the configured default encoding.

  • terms of a phrase (using phrase search) can not contain any special operator (for truncation, similiarity search etc)…only whole words

  • Left truncation, wildcard search and similarity search are expensive operations because the index has to iterate over all indexed words from the vocabulary to filter out matching words.


ZOPYX/Andreas Jung
Hundskapfklinge 33
D-72074 Tuebingen, Germany
E-mail: info at zopyx dot com


3.4.14 (2016/02/04)

  • better Plone 5 compatibility

3.4.13 (2016/02/03)

  • better Plone 5 compatibility

3.4.12 (2016/02/01)

  • fixed Plone 5 installation issue

3.4.11 (2015/09/09)

  • fixed Plone 5 installation issue

3.4.10 (2015/29/08)

  • Plone 5.0 compatibility

  • support for CMF completely dropped

3.4.9 (2014/11/23)

  • Plone 5.0 compatibility

3.4.8 (2014/04/21)

  • fixed unit test failing between Plone 4.3.2 and Plone 4.3.3

3.4.7 (2013/05/26)

  • workaround for misbehaviour in cmf_adapters

3.4.6 (2013/05/26)

  • fixed deprecation warning about getSiteEncoding()

3.4.4 (2011/06/15)

  • fixed version number in metadata.xml

3.4.4 (2011/06/15)

  • fixed version number in metadata.xml

3.4.3 (2011/05/31)

  • fixed documenation issues

3.4.2 (2011/05/31)

  • fixed CMFContentAdapter fallbacks added in release 3.4.1

  • improved test setup

3.4.1 (2010/10/30)

  • fixes in for plone.indexer

  • some Python 2.6 compatiblity fixes

  • fixed some deprecation warnings

3.4.0 (2010/07/13)

  • official release (same as 3.4.0)

3.4.0b1 (2010/05/20)

  • Made the ranking method configurable on add.

  • Allow to specify search_all_fields in query.

3.3.4 (2010/01/02)

  • ZCTextIndex -> TextIndexNG3 indexes migration code did not take the result of getIndexSourceNames() into account

3.3.3 (2010/01/01)

  • fixed Zope 2.12 compatibility

3.3.2 (2010/01/01)

  • fixed Zope 2.12 compatibility

3.3.1 (2010/01/01)

  • fixed import error

3.3.0 (2009/12/30)

  • final release

3.3.0b2 (2009/12/08)

  • fixed default value for splitter_casefolding for programmatic creation

3.3.0b1 (2009/09/24)

  • fixed minor GenericSetup configuration issue

3.3.0a2 (2009/05/02)

  • replaced ICMFFile by CMFDefault’s IFile

  • removed ExtensibleObjectWrapperAdapter and related code

  • fixed CMF 2.2 compatibility

  • fixed Zope 2.12 compatibility

  • refactored code base into:

    • Products.TextIndexNG3

    • zopyx.txng3.core

    • zopyx.txng3:ext

  • existing indexes are not fully compatible with V 3.3.0. Existing indexes must not be removed. However existing indexes must be cleared and reindexed. An existing index configuration will not be lost. Only the internal storage and lexicon data must be rebuild.


  • browser/document_for_words takes request/language into account


  • the vocabulary form in the ZMI now deals correctly with a default language other than ‘en’


  • fixed unicode issue in test_index()


  • fixed issue in constructor causing a TypeError calling bool()


  • using index_unknown_languages=True as default when creating index instances programmatically


  • unreleased


  • cmf_adapters: fall back to ‘en’ if object does not implement the Language() method


  • fixed issue in uninstallation code (#2119945)


  • fixed ZMI related unicode issues (only affecting the ZMI testing mode)


  • added dutch thesaurus (wiggy)

  • better uninstall code (wiggy)


  • baseconverter: deal with pdata

  • html converter: deal with html files without implicit charset specs


  • generated parser files are now generated in temporary directories on a per-uid basis (Unix only)


  • fixed uninstall problem


  • the stopword remover did not work correctly under all circumstances (#1892700)

  • bugfix in (reported by Wichert Akkerman)


  • several minor bugfixes

  • updated extension module to latest zopyx.textindexng3 codebase with several new stemmers


  • improved installation code

  • fixed a bug in plone_adapters caused by a change in ExtensibleObjectWrapper

  • fixed a reindexing issue


  • requires Zope 2.10 or higher

  • update parser to PLY 2.1

  • Splitter: the SimpleSplitter is now the default splitter. The old C-implementation of the splitter is retired

  • IndexableContentCollector: the implementation now deals with both textual and binary content

  • Plone/CMF adapter no longer require manual configuration of the configure.zcml files

  • ATFile adapter for Plone: now deals with textual and binary content

  • adapters code overhauled


  • bugfix in (reported by Wichert Akkerman)


  • Fixed a bug in the installation code (undefined ‘existing_ids’ var)


  • fixes a compatibility issue with Plone 2.5.3 and the latest CMF version (_ob Attribute error while indexing)



  • phrase searches could potentially return false positives (reported by Dieter Maurer)

  • fixed storage._doc2wid datastructure in order to play more nicely with the ZODB (reported by Dieter Maurer). Existing indexes should be cleared and reindexed (reported by Dieter Maurer)

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution (69.5 kB view hashes)

Uploaded source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page