Fixes #1173

Problem

The search functionality was case-sensitive, causing different results for queries like "Symfony" vs "symfony". This was due to ElasticSearch term queries being case-sensitive exact matches, and the indexed content not being normalized for case-insensitive matching.

Example of the issue:

• https://www.yiiframework.com/search?type=news&q=Symfony
• https://www.yiiframework.com/search?type=news&q=symfony

These URLs would return different search results, which is unexpected behavior for users.

Root Cause

The original implementation had two issues:

1. Query-side: Used case-sensitive term queries on text fields in models/search/SearchActiveRecord.php
2. Index-side: ElasticSearch mappings didn't provide case-insensitive keyword fields for exact matching

Even with query-side lowercasing, the indexed content remained in original case, causing term queries to fail when searching for lowercased terms against mixed-case indexed data.

Solution

Implemented a comprehensive fix using ElasticSearch's built-in normalization capabilities:

1. Added Lowercase Normalizer

Added custom lowercase normalizer to index settings across all search models:

'analysis' => [
    'normalizer' => [
        'lowercase' => [
            'type' => 'custom',
            'filter' => ['lowercase']
        ]
    ]
]

2. Added Keyword Subfields

Enhanced field mappings to include keyword subfields with lowercase normalizer:

'name' => [
    'type' => 'text',
    'fields' => [
        // existing subfields...
        'keyword' => [
            'type' => 'keyword',
            'normalizer' => 'lowercase'
        ],
    ],
],

3. Updated Term Queries

Modified exact match queries to use new keyword subfields:

// Before: case-sensitive with manual lowercasing
['term' => ['name' => mb_strtolower($queryString)]]

// After: case-insensitive via normalized keyword field  
['term' => ['name.keyword' => $queryString]]

Changes Made

SearchActiveRecord.php:

• Updated term queries to use .keyword subfields
• Removed manual mb_strtolower() calls since normalization happens at index level
• Updated comments to reflect new approach

All Search Models (SearchApiType, SearchExtension, SearchGuideSection, SearchNews, SearchWiki):

• Added lowercase normalizer to index settings
• Added keyword subfields to name and title field mappings

Benefits

• ✅ Complete case-insensitivity: Both indexing and querying handle case normalization properly
• ✅ Better performance: Keyword fields are more efficient for exact matching than text fields with manual lowercasing
• ✅ Backward compatibility: Original text fields remain unchanged for fuzzy/analyzed searches
• ✅ Maintainable: Uses ElasticSearch's built-in capabilities instead of application-level workarounds
• ✅ International support: Proper Unicode lowercasing via ElasticSearch filters

Testing

The fix handles various cases correctly:

• "Symfony" and "symfony" now produce identical ElasticSearch queries
• International characters work properly ("ÁÉÍÓÚ" → "áéíóú")
• Mixed case inputs are normalized ("ArrayHelper" → "arrayhelper")
• Autocompletion suggestions are case-insensitive
• Fuzzy matching on analyzed fields remains intact

💬 Share your feedback on Copilot coding agent for the chance to win a $200 gift card! Click here to start the survey.