// Step 1: Defining components for the "plain" analyzer
// All changes are made in the includes/Maintenance/AnalysisConfigBuilder.php file within the CirrusSearch extension. 
// First, we need to define our new rules within the main block for the Serbian/Croatian language.
// Find the customize() function (or defaults(), depending on the version), and within switch ($language) find case 'serbian':.

case 'bosnian':
case 'croatian':
case 'serbian':
case 'serbo-croatian':
    // 1. Define mapping for the letter Đ
    // We will add this mapping to the plain analyzers.
    $config['char_filter']['serbian_dj_mapping'] = [
        'type' => 'mapping',
        'mappings' => [
            'đ => dj',
            'Đ => Dj',
            'ђ => dj', // Support for users searching Cyrillic đ with bald Latin
            'Ђ => Dj'
        ]
    ];

    // 2. Define "aggressive" folding without TJones's exceptions
    // We need this because global icu_folding has unicodeSetFilter set for Serbian
    $config['filter']['serbian_plain_folding'] = [
        'type' => 'icu_folding'
        // Intentionally NOT adding unicodeSetFilter here!
    ];

    // The existing code for the 'text' analyzer remains intact so as not to break the stemmer
    $config = $myAnalyzerBuilder->withFilters( [ 'lowercase', 'icu_folding', 'serbian_stemmer' ] )
                                ->build( $config );
    break;

// Step 2: Intercepting the system method enableICUFolding
// This is the core of the solution. The enableICUFolding() method in Cirrus automatically goes through plain and plain_search analyzers 
// and glues the system icu_folding to them (which for the Serbian language contains the exceptions [^ĐđŽžĆ抚Čč]). 
// Because of this method, the search for "markovic" currently fails!
// Find the function enableICUFolding( array $config, $language ) and modify the logic that sets the plain filter.

public function enableICUFolding( array $config, $language ) {
    $unicodeSetFilter = $this->getICUSetFilter( $language );
    $filter = [
        'type' => 'icu_folding',
    ];
    if ( $unicodeSetFilter ) {
        $filter['unicodeSetFilter'] = $unicodeSetFilter;
    }
    $config['filter']['icu_folding'] = $filter;

    // ... [Existing code for replacing asciifolding in text analyzers] ...

    // THIS IS THE BLOCK WE ARE CHANGING: Explicitly enable icu_folding on plain analyzers
    if ( isset( $config['analyzer']['plain'] ) ) {
        
        // --- START OF CUSTOM LOGIC FOR T138858 ---
        if ( in_array( $language, [ 'sr', 'hr', 'bs', 'sh', 'serbian', 'croatian', 'bosnian', 'serbo-croatian' ] ) ) {
            // If plain/plain_search do not exist, initialize them
            if ( !isset( $config['analyzer']['plain']['filter'] ) ) {
                $config['analyzer']['plain']['filter'] = [];
            }
            if ( !isset( $config['analyzer']['plain_search']['filter'] ) ) {
                $config['analyzer']['plain_search']['filter'] = [];
            }
            
            // Set filters for plain
            $config['analyzer']['plain']['filter'] = [ 'lowercase', 'serbian_plain_folding' ];
            $config['analyzer']['plain_search']['filter'] = [ 'lowercase', 'serbian_plain_folding' ];

            // Set char_filter for Đ -> Dj
            if ( !isset( $config['analyzer']['plain']['char_filter'] ) ) {
                $config['analyzer']['plain']['char_filter'] = [];
            }
            if ( !isset( $config['analyzer']['plain_search']['char_filter'] ) ) {
                $config['analyzer']['plain_search']['char_filter'] = [];
            }

            // Insert mapping at the very beginning of the array (before tokenization)
            array_unshift( $config['analyzer']['plain']['char_filter'], 'serbian_dj_mapping' );
            array_unshift( $config['analyzer']['plain_search']['char_filter'], 'serbian_dj_mapping' );
        } else {
        // --- END OF CUSTOM LOGIC ---
            
            // Standard fallback logic for all other languages
            if ( !isset( $config['analyzer']['plain']['filter'] ) ) {
                $config['analyzer']['plain']['filter'] = [];
            }
            $config['analyzer']['plain']['filter'] =
                $this->switchFiltersToICUFoldingPreserve(
                    $config['analyzer']['plain']['filter'], true );
        }
    }

    return $config;
}

// Why is this plan perfect for production?
// Separation of domains: We keep the original text indexes intact, respecting Trey Jones's entire work and avoiding regression of the linguistic stemmer.
// Bypassing unicodeSetFilter: We explicitly assign the new serbian_plain_folding definition to the plain analyzer (which performs 100% "bald" removal of diacritics).
// Use of char_filter over token_filter: If you attempted the "đ to dj" transformation via token filter, Elasticsearch would apply it too late. 
// char_filter catches the raw string before tokenization, which is essential for correct indexing of words like "Novak Đoković" into "Novak Djokovic".

// Deployment steps
// After this change is saved and applied (merged) to the server, the operator must manually refresh the ES configuration and reindex the data. 
// In the MediaWiki terminal (on the cluster server):

// 1. Sends the new analyzer definition to the Elasticsearch registries (updates mappings and filters)
// php extensions/CirrusSearch/maintenance/UpdateSearchIndexConfig.php --wiki srwiki
// php extensions/CirrusSearch/maintenance/UpdateSearchIndexConfig.php --wiki hrwiki
// php extensions/CirrusSearch/maintenance/UpdateSearchIndexConfig.php --wiki bswiki
// php extensions/CirrusSearch/maintenance/UpdateSearchIndexConfig.php --wiki shwiki

// 2. Forces Elasticsearch to re-read (re-index) all existing text through the new plain analyzer
// php extensions/CirrusSearch/maintenance/ForceSearchIndex.php --wiki srwiki --skipLinks --indexOnSkip