Thoughts on the stability of PHP sorting!

PHP sorting stability problem      

I recently encountered a very interesting problem at work. The online input is a sequence of sorted associative arrays. After a The array output after series processing is out of order, and cannot be reproduced when running locally. After checking the relevant code, the most concerning part is this paragraph:

$categories = Arr::sort($categories, function ($node) {
    return $node['default'];
}, true);

is used to bring elements to the front according to default priority. First, confirm the illuminate on the bottom line. /supportThe version is consistent with the local one, check Arr::sort()source code:

...
$descending ? arsort($results, $options)
            : asort($results, $options);

In the end, php's asort is called, and online is php5. Both local and testing were replaced by php7 due to recent upgrade considerations. Finally, it was successfully reproduced in the php5 environment and it was determined to be the sort problem.

The relative position of equal elements before and after sorting does not change, which means that this algorithm is stable.

If you have a certain understanding of quick sort, you can know that quick sort is unstable, so the output of this code will not be when the elements default have the same priority. The previous order, but why do the test results retain the original order in the php7 environment. Is it wrong that I understood that all default sorting in the world is quick sort?

Okay, let’s take a quick look at how the PHP source code solves this problem of. Go to the official php-src of github and directly search asort in this repository, find the c file, and find this function defined in arr.c:814

PHP_FUNCTION(asort)
{
	zval *array;
	zend_long sort_type = PHP_SORT_REGULAR;
	bucket_compare_func_t cmp;

	ZEND_PARSE_PARAMETERS_START(1, 2)
		Z_PARAM_ARRAY_EX(array, 0, 1)
		Z_PARAM_OPTIONAL
		Z_PARAM_LONG(sort_type)
	ZEND_PARSE_PARAMETERS_END();

	// 设置比较函数
	cmp = php_get_data_compare_func(sort_type, 0);

	zend_hash_sort(Z_ARRVAL_P(array), cmp, 0);

	RETURN_TRUE;
}

You can see that the final call is zend_hash_sort(), we continue to search:

and find that this function is the nesting doll of zend_hash_sort_ex(), and finally find zend_hash.c: 2497

ZEND_API void ZEND_FASTCALL zend_hash_sort_ex(HashTable *ht, sort_func_t sort, bucket_compare_func_t compar, zend_bool renumber)
{
	Bucket *p;
	uint32_t i, j;

	IS_CONSISTENT(ht);
	HT_ASSERT_RC1(ht);

	if (!(ht->nNumOfElements>1) && !(renumber && ht->nNumOfElements>0)) {
		/* Doesn't require sorting */
		return;
	}

	// 这里的hole指数组元素被unset掉产生的洞
	if (HT_IS_WITHOUT_HOLES(ht)) {
		/* Store original order of elements in extra space to allow stable sorting. */
		for (i = 0; i < ht->nNumUsed; i++) {
			Z_EXTRA(ht->arData[i].val) = i;
		}
	} else {
		/* Remove holes and store original order. */
		for (j = 0, i = 0; j < ht->nNumUsed; j++) {
			p = ht->arData + j;
			if (UNEXPECTED(Z_TYPE(p->val) == IS_UNDEF)) continue;
			if (i != j) {
				ht->arData[i] = *p;
			}
			Z_EXTRA(ht->arData[i].val) = i;
			i++;
		}
		ht->nNumUsed = i;
	}

	sort((void *)ht->arData, ht->nNumUsed, sizeof(Bucket), (compare_func_t) compar,
			(swap_func_t)(renumber? zend_hash_bucket_renum_swap :
				((HT_FLAGS(ht) & HASH_FLAG_PACKED) ? zend_hash_bucket_packed_swap : zend_hash_bucket_swap)));
	...

Okay! , the official comments explain how to achieve the stability of sorting. Before sorting, use this Z_EXTRA to retain the mapping of original array elements to subscripts.

But when I searched for the submission information of this code, I found a problem: the rfc related to stable sorting is at https://wiki.php.net/rfc/stable_sorting. You can see that it happened in May this year. And for php8.0.

?? What happened to the previous php7 sorting stability?

Immediately construct an example:

$count = 10;
$cc = [];
for ($i=0; $i<$count; $i++) {
    $cc[] = [
        &#39;id&#39; => $i,
        &#39;default&#39; => rand(0, 10),
    ];
}
$cc = Arr::sort($cc, function ($i) {
   return $i[&#39;default&#39;];
});
dd($cc);

After many tests under php7, it was found that when $count is relatively small, the sorting is stable, but$countThe sorting becomes unstable again in larger cases. That is to say, online sorting is unstable and cannot be reproduced locally because the array length of the use case is too short.

Seeing this, it can be determined that it is a problem of quick sort length threshold optimization, and finally checked the relevant information. sort in php7 is based on the std::sort implementation of libc in LLVM. There is special optimization when the number of elements is less than or equal to 16. If the number of elements is greater than 16, quick sorting will be performed, while php5 will directly use quick sorting.

<?php

$count = 100;
$cc = [];
for ($i=0; $i<$count; $i++) {
    $cc[] = [
        &#39;id&#39; => $i,
        &#39;default&#39; => rand(0, 10),
    ];
}
usort($cc, function($a, $b){
    if ($a[&#39;default&#39;] == $b[&#39;default&#39;]) return 0;
    return ($a[&#39;default&#39;] < $b[&#39;default&#39;]) ? 1 : -1;
});
print_r($cc);

Finally tried it in php8 environment, the sorting is absolutely stable

The above is the detailed content of Thoughts on the stability of PHP sorting!. For more information, please follow other related articles on the PHP Chinese website!