Optimize array_diff() for long values

Author: mehmetcansahin

UPGRADING

@@ -544,6 +544,7 @@ PHP 8.6 UPGRADE NOTES
   . Reduced temporary allocations when iterating Phar directories.
 
 - Standard:
+  . Improved performance of array_diff() for integer-only arrays.
   . Improved performance of array_fill_keys().
   . Improved performance of array_map() with multiple arrays passed.
   . Improved performance of array_sum() and array_product() for

ext/standard/array.c

@@ -5753,6 +5753,68 @@ PHP_FUNCTION(array_diff_ukey)
 }
 /* }}} */
 
+static zend_always_inline bool php_array_values_are_all_long(HashTable *ht)
+{
+	zval *value;
+
+	ZEND_HASH_FOREACH_VAL(ht, value) {
+		ZVAL_DEREF(value);
+		if (UNEXPECTED(Z_TYPE_P(value) != IS_LONG)) {
+			return false;
+		}
+	} ZEND_HASH_FOREACH_END();
+
+	return true;
+}
+
+static zend_never_inline bool php_array_diff_long(zval *args, uint32_t argc, uint32_t exclude_count, zval *return_value)
+{
+	HashTable exclude;
+	zval *value, *entry;
+	zend_string *key;
+	zend_long idx;
+	uint32_t i;
+
+	/* The copy loop below cannot cheaply undo a partially built result, so
+	 * the first array must be checked upfront. */
+	if (!php_array_values_are_all_long(Z_ARRVAL(args[0]))) {
+		return false;
+	}
+
+	/* Create the exclude map while checking the value types, bailing out to
+	 * the generic implementation on the first non-long value. A late bailout
+	 * discards the partially built map, but that is cheaper than always
+	 * paying an extra type-check pass on the all-long path. */
+	zend_hash_init(&exclude, exclude_count, NULL, NULL, 0);
+	for (i = 1; i < argc; i++) {
+		ZEND_HASH_FOREACH_VAL(Z_ARRVAL(args[i]), value) {
+			ZVAL_DEREF(value);
+			if (UNEXPECTED(Z_TYPE_P(value) != IS_LONG)) {
+				zend_hash_destroy(&exclude);
+				return false;
+			}
+			zend_hash_index_add_empty_element(&exclude, Z_LVAL_P(value));
+		} ZEND_HASH_FOREACH_END();
+	}
+
+	array_init_size(return_value, zend_hash_num_elements(Z_ARRVAL(args[0])));
+	ZEND_HASH_FOREACH_KEY_VAL(Z_ARRVAL(args[0]), idx, key, value) {
+		entry = value;
+		ZVAL_DEREF(entry);
+		if (!zend_hash_index_exists(&exclude, Z_LVAL_P(entry))) {
+			if (key) {
+				value = zend_hash_add_new(Z_ARRVAL_P(return_value), key, value);
+			} else {
+				value = zend_hash_index_add_new(Z_ARRVAL_P(return_value), idx, value);
+			}
+			zval_add_ref(value);
+		}
+	} ZEND_HASH_FOREACH_END();
+
+	zend_hash_destroy(&exclude);
+	return true;
+}
+
 /* {{{ Returns the entries of arr1 that have values which are not present in any of the others arguments. */
 PHP_FUNCTION(array_diff)
 {
@@ -5853,6 +5915,10 @@ PHP_FUNCTION(array_diff)
 		RETURN_THROWS();
 	}
 
+	if (php_array_diff_long(args, argc, (uint32_t)num, return_value)) {
+		return;
+	}
+
 	ZVAL_NULL(&dummy);
 	/* create exclude map */
 	zend_hash_init(&exclude, num, NULL, NULL, 0);

ext/standard/tests/array/array_diff_long_values.phpt

@@ -0,0 +1,41 @@
+--TEST--
+array_diff() with long values
+--FILE--
+<?php
+function dump_array(array $value): void {
+    echo json_encode($value), "\n";
+}
+
+dump_array(array_diff([1, 2, 3], [2]));
+dump_array(array_diff([-2, -1, 0, 1], [-1, 1]));
+dump_array(array_diff([PHP_INT_MIN, -1, PHP_INT_MAX], [PHP_INT_MIN, PHP_INT_MAX]));
+dump_array(array_diff([1, 2, 2, 3], [2]));
+dump_array(array_diff(['a' => 1, 'b' => 2, 'c' => 3], [2]));
+dump_array(array_diff([1, 2, 3], ['2']));
+dump_array(array_diff([1, '2', 3], [2]));
+dump_array(array_diff([1, 2, 3, 4], [2], [4, 5]));
+dump_array(array_diff([1, 2, 3, 4], [2], ['3']));
+dump_array(array_diff([1, 2], [1, 2]));
+
+$y = 2;
+dump_array(array_diff([1, 2, 3], [&$y]));
+
+$x = 1;
+$array = [&$x, 2];
+$result = array_diff($array, [2]);
+$x = 9;
+var_dump($result[0]);
+?>
+--EXPECT--
+{"0":1,"2":3}
+{"0":-2,"2":0}
+{"1":-1}
+{"0":1,"3":3}
+{"a":1,"c":3}
+{"0":1,"2":3}
+{"0":1,"2":3}
+{"0":1,"2":3}
+{"0":1,"3":4}
+[]
+{"0":1,"2":3}
+int(9)