Introduce 'wp_ai_client_cache_group' filter to allow customization of the object cache group

[baikare.sandeep007]

Introduce 'wp_ai_client_cache_group' filter to allow customization of the object cache group

Description

This patch introduces a new filter, wp_ai_client_cache_group, to the WP_AI_Client_Cache class. This allows developers and integrators to change the object-cache group under which the AI client stores its keys.

Previously, the cache group was hardcoded as the class constant self::CACHE_GROUP ('wp_ai_client'). This patch replaces all direct references to the constant with a call to a new private method get_cache_group(), which applies the new filter.

Purpose

The wp_ai_client_cache_group filter provides integrators with greater control over cache placement and invalidation. Key benefits include:

1. Avoid Key Collisions

Plugins or themes can choose a distinct group to avoid clashing with other code that might use similar cache keys on the same site.

2. Environment-Specific Caches

On multisite or hosting platforms that namespace caches per-site, integrators can map AI client keys into a per-site or per-environment group (e.g., include the blog ID or environment name) to prevent cross-site data leakage.

3. Backend Constraints / Compatibility

Some persistent object cache implementations treat groups differently (or do not support group flushing). Allowing the group to be replaced lets integrators adapt—for example, to use a group the host supports or to fall back to transient-based storage.

4. Testing and Isolation

Unit and integration tests can switch the cache group to an ephemeral one (or a mock group) so test runs do not interfere with real cached data.

5. Targeted Invalidation / Performance

Integrators who want to implement custom invalidation strategies—such as namespacing by model/version or sharding keys—can change the group to a name that their backend can efficiently flush or monitor.

Implementation Notes

• A new private method get_cache_group(): string is introduced. This method applies the wp_ai_client_cache_group filter to the default self::CACHE_GROUP constant.
• All cache operations (wp_cache_get, wp_cache_set, wp_cache_delete, etc.) now use $this->get_cache_group() instead of the hardcoded constant.
• The filter is expected to return a string cache group name. For predictable behavior, the returned value is explicitly cast to a string.
• Two new unit tests are added to Tests_AI_Client_Cache:
  1. test_cache_group_filter_is_respected: Verifies the filter changes the group used for cache operations.
  2. test_cache_group_filter_returns_non_string: Confirms non-string filter return values are properly cast to strings.

Example Usage

Basic: Change to a custom group

<?php
add_filter( 'wp_ai_client_cache_group', function( $group ) {
    return 'my_plugin_ai_cache';
} );

This stores all AI client cache items under the 'my_plugin_ai_cache' group.

Multisite: Per-site isolation

<?php
add_filter( 'wp_ai_client_cache_group', function( $group ) {
    return $group . '_site_' . get_current_blog_id();
} );

This appends the current site ID to the cache group, preventing cross-site cache sharing.

Testing: Use an ephemeral group

<?php
add_filter( 'wp_ai_client_cache_group', function( $group ) {
    return 'tests_' . uniqid();
} );

This generates a unique, temporary cache group for the duration of a test run.

Backward Compatibility

• Fully backward compatible – When the filter is not used, the cache group defaults to the original 'wp_ai_client' constant, preserving existing behavior.
• No functional changes to existing code that relies on the hardcoded cache group.

Impact

• Performance: Negligible overhead – the filter only executes on cache operations and the method is lightweight.
• Security: None – this is an additive change that provides more control without introducing new attack surfaces.

Testing

The attached patch includes comprehensive unit test coverage:

1. Verifies that applying the filter changes the cache group used for set() and get() operations.
2. Ensures that non-string return values (e.g., integers) are gracefully cast to strings.

Manual testing has confirmed that existing cache functionality remains unchanged when the filter is not applied.

Props

Props @baikare.sandeep007.

[baikare.sandeep007]

Updated patch file with ticket umber in tests

Trac ticket: https://core.trac.wordpress.org/ticket/65127

## Use of AI Tools

AI assistance: Yes
Tool(s): GitHub Copilot
Model(s): GPT-5-mini
Used for: test suggestions

[baikare.sandeep007]

Updated filter hook comment as per WPCS

[audrasjb]

Removing trunk version as this is not going to be shipped with WP 7.0 but in the next releases.

[desrosj]

Since this is an enhancement, there's no first version of WordPress this can be reproduced in. Removing trunk version.