array_path_get and array_path_exists functions by carlos-granados · Pull Request #21637 · php/php-src

carlos-granados · 2026-04-04T14:01:59Z

This PR proposes adding two small, focused array functions to ext/standard for retrieving and checking nested array elements using a path, in a single step.

See the related RFC:
https://wiki.php.net/rfc/array_get_and_array_has

ndossche · 2026-04-04T21:59:39Z

+
+	if (dot == NULL) {
+		/* No dot found, this is a simple key lookup */
+		zend_string *zkey = zend_string_init(key, key_len, 0);


If you have a zend_string already here, then you can avoid the allocation.
In general, use zend_symtable_str_find (also below).

ndossche · 2026-04-04T22:00:22Z

+
+	/* If key is null, return the whole array */
+	if (key == NULL || Z_TYPE_P(key) == IS_NULL) {
+		ZVAL_ARR(return_value, zend_array_dup(ht));


Avoid a duplication, just return a copy with the refcount. Sadly, ZVAL_ARR sucks as it doesn't take into account the mutability in the type info; so you'd have to use Z_PARAM_ARRAY so you have a zval that you can use RETURN_COPY on.

ndossche · 2026-04-04T22:00:28Z

+		result = array_get_nested(ht, Z_STRVAL_P(key), Z_STRLEN_P(key));
+
+		if (result != NULL) {
+			ZVAL_COPY(return_value, result);


RETURN_COPY

ndossche · 2026-04-04T22:00:35Z

+		result = zend_hash_index_find(ht, Z_LVAL_P(key));
+
+		if (result != NULL) {
+			ZVAL_COPY(return_value, result);


RETURN_COPY

ndossche · 2026-04-04T22:00:42Z

+
+	/* Key not found, return default value */
+	if (default_value != NULL) {
+		ZVAL_COPY(return_value, default_value);


RETURN_COPY

ndossche · 2026-04-04T22:00:49Z

+	if (default_value != NULL) {
+		ZVAL_COPY(return_value, default_value);
+	} else {
+		RETVAL_NULL();


Not necessary, the return value is NULL implicitly

ndossche · 2026-04-04T22:01:03Z

+	}
+
+	/* Invalid key type */
+	RETURN_FALSE;


This case is impossible by using the type int|string, use ZEND_UNREACHABLE();

Not yet impossible, because Z_PARAM_ZVAL, but yes, it should be properly typed in ZPP.

ndossche · 2026-04-04T22:02:38Z

+	}
+
+	/* Recurse into the nested array with the remaining key */
+	return array_get_nested(Z_ARRVAL_P(current), dot + 1, key_len - segment_len - 1);


Unless this gets tailcall eliminated, you have primitive recursion here which can overflow. Best to write an explicit loop.

ndossche · 2026-04-04T22:04:35Z

+	current = zend_symtable_find(ht, segment);
+	zend_string_release(segment);
+
+	if (current == NULL || Z_TYPE_P(current) != IS_ARRAY) {


Does this handle references properly?

carlos-granados · 2026-04-06T14:05:18Z

@ndossche thanks for your review. Following some comments on the RFC discussion I have updated the implementation to also accept an array in the $key parameter. I also looked into all the other issues that you raised (though some do not apply any longer)

ndossche · 2026-04-06T17:22:18Z

+	}
+
+	/* Handle integer keys (simple lookup) */
+	ZEND_ASSERT(Z_TYPE_P(key) == IS_LONG);


This is insufficient, the ZPP parsing still accepts any argument. The stubs don't enforce the argument type check in any way.

ndossche · 2026-04-06T17:22:34Z

+		/* Use php_explode to split the string by '.' */
+		zend_string *delim = ZSTR_CHAR('.');
+		array_init(&segments_array);
+		php_explode(delim, Z_STR_P(key), &segments_array, ZEND_LONG_MAX);


This is a bit of an inefficient way to go about it.

ndossche · 2026-04-06T17:23:17Z

+		result = zend_hash_index_find(ht, Z_LVAL_P(key));
+
+		if (result != NULL) {
+			RETURN_COPY(result);


This won't work properly with references I believe.
The function stub doesn't seem to return a reference, so this needs to be RETURN_COPY_DEREF (same issue is also present in other places).

# Conflicts: # ext/standard/basic_functions_arginfo.h # ext/standard/basic_functions_decl.h

carlos-granados · 2026-04-11T10:27:09Z

After some discussions, I have decided to remove the dot notation functionality and just leave the array paths option. This makes the code simpler

ndossche · 2026-04-11T10:29:10Z

+	num_segments = zend_hash_num_elements(path);
+
+	/* Iterate through each segment in the path array */
+	for (idx = 0; idx < num_segments; idx++) {


The way you wrote this implies something about the design of the function/API (this is important to note for your RFC).

This means that the array must be a list, not just a regular array.
e.g. the following will fail even though intuitively it should work:

<?php $path = ['hello', 1, 0]; unset($path[1]); print_r($path); /* Prints Array ( [0] => hello [2] => 0 ) */ var_dump(array_get([ 'hello' => ['world'], ], $path)); // Prints NULL

Either this should be fixed in the code or the RFC should explicitly spell this out.

One way to fix this is to use ZEND_HASH_FOREACH
EDIT: Looking at your RFC, you should indeed use ZEND_HASH_FOREACH

ndossche · 2026-04-11T10:30:24Z

+		}
+
+		/* Check if the segment exists and is an array for next iteration */
+		if (current == NULL || Z_TYPE_P(current) != IS_ARRAY) {


References are handled incorrectly:

<?php $path = ['hello', 0]; $array2 = ['world']; $array = ['hello' => &$array2]; var_dump(array_get($array, $path)); // Prints NULL, should print 'world'

ndossche · 2026-04-11T10:30:40Z

+
+	/* Path not found, return default value */
+	if (default_value != NULL) {
+		RETURN_COPY_DEREF(default_value);


I don't think this DEREF is necessary, as the function gets its default value passed by value.

ndossche · 2026-04-11T10:31:06Z

+/* {{{ Retrieves a value from a deeply nested array using an array path */
+PHP_FUNCTION(array_get)
+{
+	zval *array;


Where possible, please merge the declaration and assignment. Splitting them is bad practice nowadays because the scope of the variable is unclear.

ndossche · 2026-04-11T10:31:10Z

+/* {{{ Helper function to get a nested value from array using an array of path segments */
+static zval* array_get_nested(HashTable *ht, HashTable *path)
+{
+	zval *segment_val;


Where possible, please merge the declaration and assignment. Splitting them is bad practice nowadays because the scope of the variable is unclear.

ndossche · 2026-04-11T10:32:48Z

+		}
+
+		/* If this is the last segment, return the result */
+		if (idx == num_segments - 1) {


I think this check can be avoided by initializing current to NULL and returning current at the end of the function

ndossche · 2026-04-11T10:36:08Z

+		}
+
+		/* Segment must be a string or int */
+		if (Z_TYPE_P(segment_val) == IS_STRING) {


Oh and references are also not handled for segment_val. You can use zend_hash_index_find_deref, or if you switch to a ZEND_HASH_FOREACH loop use something like ZVAL_DEREF.

carlos-granados · 2026-04-11T16:26:43Z

@ndossche many thanks for your insightful comments, hopefully everything is OK now

ndossche

I didn't re-test, but my concerns are resolved and it looks fine from a technical PoV.
I'll leave the final review to someone else for after the RFC has been voted on; when it has you also need to add an entry to the UPGRADING file.

carlos-granados requested review from bukka and kocsismate as code owners April 4, 2026 14:02

github-actions bot added Extension: standard ABI break labels Apr 4, 2026

iluuu1994 added the RFC label Apr 4, 2026

ndossche reviewed Apr 4, 2026

View reviewed changes

ndossche reviewed Apr 6, 2026

View reviewed changes

carlos-granados added 5 commits April 11, 2026 11:58

Add array_get and array_has functions

cb40915

# Conflicts: # ext/standard/basic_functions_arginfo.h # ext/standard/basic_functions_decl.h

Add tests

3c71ecf

Accept an array of strings in the $key parameter

10037a7

changes after new PR review

cb4855d

Remove dot notation

4af39b9

carlos-granados force-pushed the array-get-array-has branch from f3248f2 to 4af39b9 Compare April 11, 2026 10:00

ndossche requested changes Apr 11, 2026

View reviewed changes

ndossche reviewed Apr 11, 2026

View reviewed changes

More changes after PR review

6f1bc06

ndossche approved these changes Apr 13, 2026

View reviewed changes

carlos-granados added 3 commits April 14, 2026 18:21

Throw TypeError exception if segment is not int or string

b947093

Add _path suffix to function names

01b4726

Update function names to array_path_get and array_path_exists

b3f7065

carlos-granados changed the title ~~Array_get and array_has functions~~ array_path_get and array_path_exists functions Apr 18, 2026

Conversation

carlos-granados commented Apr 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ndossche Apr 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

carlos-granados commented Apr 6, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

carlos-granados commented Apr 11, 2026

Uh oh!

ndossche Apr 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

carlos-granados commented Apr 11, 2026

Uh oh!

ndossche left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

carlos-granados commented Apr 4, 2026 •

edited

Loading

ndossche Apr 4, 2026 •

edited

Loading

ndossche Apr 11, 2026 •

edited

Loading