Edit screen freezes with many relationship connections (show() + n² options)

Hi Peter,

Thanks for confirming so quickly. Here are the two patches we run on our site. Both are temporary and will be removed once your fix ships. They are plain PHP, so they can go in a small plugin or a theme's functions.php.

1) JS: no forced style recalc during mb_ready

<?php
/**
 * Temporary site-side patch: Meta Box start-up (mb_ready) without a forced style recalc per field.
 * Only while the document-level "mb_ready" runs, show() on select_advanced dropdowns and on the
 * clone add/remove buttons just clears an inline display:none (these elements are visible by
 * default in Meta Box's CSS). Every other show() call keeps jQuery's behaviour, and show() is
 * restored right after "mb_ready".
 */
add_action( 'admin_enqueue_scripts', function () {
	if ( ! wp_script_is( 'rwmb', 'enqueued' ) ) {
		return;
	}

	$js = <<<'JS'
( function ( $, rwmb ) {
	'use strict';
	if ( ! $ || ! rwmb || ! rwmb.$document ) {
		return;
	}
	const selector = 'select.rwmb-select_advanced, select.rwmb-icon, .rwmb-button.remove-clone, .rwmb-button.add-clone';
	const $document = rwmb.$document;
	const originalTrigger = $document.trigger;

	$document.trigger = function ( event ) {
		const name = typeof event === 'string' ? event : ( event && event.type );
		if ( name !== 'mb_ready' ) {
			return originalTrigger.apply( this, arguments );
		}

		const originalShow = $.fn.show;
		$.fn.show = function () {
			const onlyFastTargets = arguments.length === 0 && this.length > 0
				&& Array.prototype.every.call( this, el => el.nodeType === 1 && el.matches( selector ) );
			if ( ! onlyFastTargets ) {
				return originalShow.apply( this, arguments );
			}
			for ( const el of this ) {
				if ( el.style.display === 'none' ) {
					el.style.display = '';
				}
			}
			return this;
		};

		try {
			return originalTrigger.apply( this, arguments );
		} finally {
			$.fn.show = originalShow;
		}
	};
} )( window.jQuery, window.rwmb );
JS;

	wp_add_inline_script( 'rwmb', $js, 'after' );
}, PHP_INT_MAX );

We kept it narrow on purpose. Other show() calls during mb_ready (tabs, file and media fields) still go through jQuery, because we didn't check whether their elements are hidden by CSS. In your code, the direct fix is still the one from my first post, in transform() and in the clone.js toggles.

2) PHP: each clone only renders its own option(s)

<?php
/**
 * Temporary site-side patch: each row of a cloneable ajax object field only renders its own option(s).
 * RWMB_Clone::html() runs the "rwmb_html" filter on each clone with ( $html, $sub_field, $sub_meta ).
 */
add_filter( 'rwmb_html', function ( $html, $field, $meta ) {
	if ( ! is_admin() || ! is_string( $html ) || empty( $field['clone'] ) || empty( $field['ajax'] ) ) {
		return $html;
	}
	if ( ! in_array( $field['type'] ?? '', [ 'post', 'taxonomy', 'taxonomy_advanced', 'user' ], true )
		|| 'select_advanced' !== ( $field['field_type'] ?? '' ) ) {
		return $html;
	}

	// The row's value: one ID, an array of IDs (multiple) or a comma list (taxonomy_advanced).
	$ids = [];
	if ( is_array( $meta ) ) {
		array_walk_recursive( $meta, function ( $value ) use ( &$ids ) {
			if ( is_scalar( $value ) ) {
				$ids[] = $value;
			}
		} );
	} elseif ( is_scalar( $meta ) && '' !== (string) $meta ) {
		$ids = (string) $meta;
	}
	$keep = array_flip( array_filter( wp_parse_id_list( $ids ) ) );

	// RWMB_Walker_Select writes <option value="ID" ...>escaped label</option>; the placeholder (value="") stays.
	$new = preg_replace_callback( '#<option value="(\d+)"[^>]*>.*?</option>#s', function ( $match ) use ( $keep ) {
		return isset( $keep[ (int) $match[1] ] ) ? $match[0] : '';
	}, $html );

	return null === $new ? $html : $new;
}, 99, 3 );

This one works on the HTML because there is no per-clone filter on the options themselves. In your code, filtering $field['options'] per clone is cleaner. One detail: the one-liner in my first post uses (array) $meta, which doesn't split taxonomy_advanced comma lists. The patch above handles those, as well as multiple fields.

Results on our post: mb_ready went from 54 s to 3.9 s with patch 1 alone, and to 752 ms with both. Saving keeps every connection; we checked the counts in wp_mb_relationships before and after.

Happy to test the build on a staging copy as soon as it's ready.

Thanks!

Adrien — Stellar Web Solutions