Character Set

UTF-8 and single-byte character set support

Zend_Search_Lucene works with the UTF-8 charset internally. Index files store unicode data in Java's "modified UTF-8 encoding". Zend_Search_Lucene core completely supports this encoding with one exception. [16]

Actual input data encoding may be specified through Zend_Search_Lucene API. Data will be automatically converted into UTF-8 encoding.

Default text analyzer

However, the default text analyzer (which is also used within query parser) uses ctype_alpha() for tokenizing text and queries.

ctype_alpha() is not UTF-8 compatible, so the analyzer converts text to 'ASCII//TRANSLIT' encoding before indexing. The same processing is transparently performed during query parsing. [17]

Default analyzer doesn't treats numbers as parts of terms. Use corresponding 'Num' analyzer if you don't want words to be broken by numbers.

UTF-8 compatible text analyzers

Zend_Search_Lucene also contains a set of UTF-8 compatible analyzers: Zend_Search_Lucene_Analysis_Analyzer_Common_Utf8, Zend_Search_Lucene_Analysis_Analyzer_Common_Utf8Num, Zend_Search_Lucene_Analysis_Analyzer_Common_Utf8_CaseInsensitive, Zend_Search_Lucene_Analysis_Analyzer_Common_Utf8Num_CaseInsensitive.

Any of this analyzers can be enabled with the code like this:

Zend_Search_Lucene_Analysis_Analyzer::setDefault(
    new Zend_Search_Lucene_Analysis_Analyzer_Common_Utf8());

UTF-8 compatible analyzers were improved in Zend Framework 1.5. Early versions of analyzers assumed all non-ascii characters are letters. New analyzers implementation has more accurate behavior.

This may need you to re-build index to have data and search queries tokenized in the same way, otherwise search engine may return wrong result sets.

All of these analyzers need PCRE (Perl-compatible regular expressions) library to be compiled with UTF-8 support turned on. PCRE UTF-8 support is turned on for the PCRE library sources bundled with PHP source code distribution, but if shared library is used instead of bundled with PHP sources, then UTF-8 support state may depend on you operating system.

Use the following code to check, if PCRE UTF-8 support is enabled:

if (@preg_match('/\pL/u', 'a') == 1) {
    echo "PCRE unicode support is turned on.\n";
} else {
    echo "PCRE unicode support is turned off.\n";
}

Case insensitive versions of UTF-8 compatible analyzers also need mbstring extension to be enabled.

If you don't want mbstring extension to be turned on, but need case insensitive search, you may use the following approach: normalize source data before indexing and query string before searching by converting them to lowercase:

// Indexing
setlocale(LC_CTYPE, 'de_DE.iso-8859-1');

...

Zend_Search_Lucene_Analysis_Analyzer::setDefault(
    new Zend_Search_Lucene_Analysis_Analyzer_Common_Utf8());

...

$doc = new Zend_Search_Lucene_Document();

$doc->addField(Zend_Search_Lucene_Field::UnStored('contents',
                                                  strtolower($contents)));

// Title field for search through (indexed, unstored)
$doc->addField(Zend_Search_Lucene_Field::UnStored('title',
                                                  strtolower($title)));

// Title field for retrieving (unindexed, stored)
$doc->addField(Zend_Search_Lucene_Field::UnIndexed('_title', $title));
// Searching
setlocale(LC_CTYPE, 'de_DE.iso-8859-1');

...

Zend_Search_Lucene_Analysis_Analyzer::setDefault(
    new Zend_Search_Lucene_Analysis_Analyzer_Common_Utf8());

...

$hits = $index->find(strtolower($query));


[16] Zend_Search_Lucene supports only Basic Multilingual Plane (BMP) characters (from 0x0000 to 0xFFFF) and doesn't support "supplementary characters" (characters whose code points are greater than 0xFFFF)

Java 2 represents these characters as a pair of char (16-bit) values, the first from the high-surrogates range (0xD800-0xDBFF), the second from the low-surrogates range (0xDC00-0xDFFF). Then they are encoded as usual UTF-8 characters in six bytes. Standard UTF-8 representation uses four bytes for supplementary characters.

[17] Conversion to 'ASCII//TRANSLIT' may depend on current locale and OS.