IBX-11401: Fixed embedding default model inheritance and added field name fallback prefixes

phpstan-baseline.neon

@@ -18480,18 +18480,6 @@ parameters:
 			count: 1
 			path: src/lib/Search/Common/EventSubscriber/UserEventSubscriber.php
 
-		-
-			message: '#^Method Ibexa\\Core\\Search\\Common\\FieldNameGenerator\:\:__construct\(\) has parameter \$fieldNameMapping with no value type specified in iterable type array\.$#'
-			identifier: missingType.iterableValue
-			count: 1
-			path: src/lib/Search/Common/FieldNameGenerator.php
-
-		-
-			message: '#^Property Ibexa\\Core\\Search\\Common\\FieldNameGenerator\:\:\$fieldNameMapping type has no value type specified in iterable type array\.$#'
-			identifier: missingType.iterableValue
-			count: 1
-			path: src/lib/Search/Common/FieldNameGenerator.php
-
 		-
 			message: '#^Method Ibexa\\Core\\Search\\Common\\FieldNameResolver\:\:getAggregationFieldName\(\) should return string\|null but returns false\.$#'
 			identifier: return.type

src/bundle/Core/DependencyInjection/Configuration/Parser/Embeddings.php

@@ -47,7 +47,6 @@ public function addSemanticConfig(NodeBuilder $nodeBuilder): void
             ->end()
             ->scalarNode('default_embedding_model')
                 ->info('Default embedding model identifier')
-                ->defaultValue('text-embedding-ada-002')
             ->end();
     }
 
@@ -57,6 +56,8 @@ public function addSemanticConfig(NodeBuilder $nodeBuilder): void
     public function preMap(array $config, ContextualizerInterface $contextualizer): void
     {
         $contextualizer->mapConfigArray('embedding_models', $config);
+        // Keep the default value in default settings only. Defining a scalar default on the semantic node would
+        // materialize it into every siteaccess/group scope and block inheritance from `system.default` overrides.
         $contextualizer->mapSetting('default_embedding_model', $config);
     }
 

src/lib/Resources/settings/search_engines/common.yml

@@ -25,6 +25,9 @@ parameters:
         ibexa_dense_vector_ada002: 'ada002_dv'
         ibexa_dense_vector_3small: '3small_dv'
         ibexa_dense_vector_3large: '3large_dv'
+    # Allows generic mapping of embedding field types (`ibexa_dense_vector_<suffix>`) without per-model entries.
+    ibexa.search.common.field_name_generator.fallback_prefixes:
+        - 'ibexa_dense_vector_'
 
 services:
     # Note: services tagged with 'ibexa.field_type.indexable'
@@ -37,6 +40,7 @@ services:
         class: Ibexa\Core\Search\Common\FieldNameGenerator
         arguments:
             - '%ibexa.search.common.field_name_generator.map%'
+            - '%ibexa.search.common.field_name_generator.fallback_prefixes%'
 
     Ibexa\Core\Search\Common\FieldNameResolver:
         class: Ibexa\Core\Search\Common\FieldNameResolver

src/lib/Search/Common/FieldNameGenerator.php

@@ -15,32 +15,19 @@
 class FieldNameGenerator
 {
     /**
-     * Simple mapping for our internal field types, consisting of an array
-     * of SPI Search FieldType identifier as key and search backend field type
-     * string as value.
+     * `$fieldNameMapping` maps internal search field type identifiers to backend
+     * suffixes (e.g. `ibexa_string` => `s`).
      *
-     * We implement this mapping, because those dynamic fields are common to
-     * search backend configurations.
+     * `$fallbackPrefixes` defines type prefixes for generic fallback normalization
+     * when no explicit mapping exists (e.g. `ibexa_dense_vector_<suffix>`).
      *
-     * @see \Ibexa\Contracts\Core\Search\FieldType
-     *
-     * Code example:
-     *
-     * <code>
-     *  array(
-     *      "ez_integer" => "i",
-     *      "ez_string" => "s",
-     *      ...
-     *  )
-     * </code>
-     *
-     * @var array
+     * @param array<string, string> $fieldNameMapping
+     * @param string[] $fallbackPrefixes
      */
-    protected $fieldNameMapping;
-
-    public function __construct(array $fieldNameMapping)
-    {
-        $this->fieldNameMapping = $fieldNameMapping;
+    public function __construct(
+        protected array $fieldNameMapping,
+        private readonly array $fallbackPrefixes = []
+    ) {
     }
 
     /**
@@ -76,8 +63,26 @@ public function getTypedName(string $name, FieldType $type): string
             return $name;
         }
 
-        $typeName = $this->fieldNameMapping[$type->getType()] ?? $type->getType();
+        $typeIdentifier = $type->getType();
+        $typeName = $this->fieldNameMapping[$typeIdentifier] ?? $this->normalizeUsingFallbackPrefixes($typeIdentifier);
 
         return $name . '_' . $typeName;
     }
+
+    /**
+     * Generic fallback for field type families that encode backend suffix in the type identifier.
+     *
+     * Example:
+     * - `ibexa_dense_vector_gemini_embedding_001_1536_dv` => `gemini_embedding_001_1536_dv`
+     */
+    private function normalizeUsingFallbackPrefixes(string $typeIdentifier): string
+    {
+        foreach ($this->fallbackPrefixes as $prefix) {
+            if (str_starts_with($typeIdentifier, $prefix)) {
+                return substr($typeIdentifier, strlen($prefix));
+            }
+        }
+
+        return $typeIdentifier;
+    }
 }

tests/bundle/Core/DependencyInjection/Configuration/Parser/EmbeddingsTest.php

@@ -43,6 +43,34 @@ public function testDefaultEmbeddingsSettings(): void
         $this->assertConfigResolverParameterValue('default_embedding_model', 'text-embedding-ada-002', 'ibexa_demo_site');
     }
 
+    public function testDefaultEmbeddingModelSetInDefaultScopeIsNotMaterializedToOtherScopes(): void
+    {
+        $this->load([
+            'system' => [
+                'default' => [
+                    'default_embedding_model' => 'gemini_embedding_001_1536',
+                ],
+            ],
+        ]);
+
+        self::assertFalse(
+            $this->container->hasParameter('ibexa.site_access.config.ibexa_demo_group.default_embedding_model')
+        );
+        self::assertFalse(
+            $this->container->hasParameter('ibexa.site_access.config.ibexa_demo_site.default_embedding_model')
+        );
+
+        self::assertSame(
+            'gemini_embedding_001_1536',
+            $this->container->getParameter('ibexa.site_access.config.default.default_embedding_model')
+        );
+        $this->assertConfigResolverParameterValue(
+            'default_embedding_model',
+            'gemini_embedding_001_1536',
+            'ibexa_demo_site'
+        );
+    }
+
     /**
      * @param array<mixed> $config
      * @param array<mixed> $expected

tests/lib/Search/Common/FieldNameGeneratorTest.php

@@ -0,0 +1,57 @@
+<?php
+
+/**
+ * @copyright Copyright (C) Ibexa AS. All rights reserved.
+ * @license For full copyright and license information view LICENSE file distributed with this source code.
+ */
+declare(strict_types=1);
+
+namespace Ibexa\Tests\Core\Search\Common;
+
+use Ibexa\Contracts\Core\Search\FieldType;
+use Ibexa\Core\Search\Common\FieldNameGenerator;
+use PHPUnit\Framework\TestCase;
+
+final class FieldNameGeneratorTest extends TestCase
+{
+    public function testGetTypedNameUsesConfiguredMapping(): void
+    {
+        $generator = new FieldNameGenerator([
+            'ibexa_string' => 's',
+        ]);
+
+        $fieldType = $this->createMock(FieldType::class);
+        $fieldType
+            ->method('getType')
+            ->willReturn('ibexa_string');
+
+        self::assertSame('title_s', $generator->getTypedName('title', $fieldType));
+    }
+
+    public function testGetTypedNameNormalizesEmbeddingFieldTypeWithoutExplicitMapping(): void
+    {
+        $generator = new FieldNameGenerator([], ['ibexa_dense_vector_']);
+
+        $fieldType = $this->createMock(FieldType::class);
+        $fieldType
+            ->method('getType')
+            ->willReturn('ibexa_dense_vector_gemini_embedding_001_1536_dv');
+
+        self::assertSame(
+            'taxonomy_embeddings_gemini_embedding_001_1536_dv',
+            $generator->getTypedName('taxonomy_embeddings', $fieldType)
+        );
+    }
+
+    public function testGetTypedNameReturnsOriginalTypeWhenNoFallbackPrefixMatches(): void
+    {
+        $generator = new FieldNameGenerator([], ['ibexa_dense_vector_']);
+
+        $fieldType = $this->createMock(FieldType::class);
+        $fieldType
+            ->method('getType')
+            ->willReturn('custom_type');
+
+        self::assertSame('foo_custom_type', $generator->getTypedName('foo', $fieldType));
+    }
+}