<?php
/**
 * WordPress implementation for PHP functions either missing from older PHP versions or not included by default.
 *
 * @package PHP
 * @access private
 */

// If gettext isn't available.
if ( ! function_exists( '_' ) ) {
	function _( $string ) {
		return $string;
	}
}

/**
 * Returns whether PCRE/u (PCRE_UTF8 modifier) is available for use.
 *
 * @ignore
 * @since 4.2.2
 * @access private
 *
 * @param bool $set - Used for testing only
 *             null   : default - get PCRE/u capability
 *             false  : Used for testing - return false for future calls to this function
 *             'reset': Used for testing - restore default behavior of this function
 */
function _wp_can_use_pcre_u( $set = null ) {
	static $utf8_pcre = 'reset';

	if ( null !== $set ) {
		$utf8_pcre = $set;
	}

	if ( 'reset' === $utf8_pcre ) {
		// phpcs:ignore WordPress.PHP.NoSilencedErrors.Discouraged -- intentional error generated to detect PCRE/u support.
		$utf8_pcre = @preg_match( '/^./u', 'a' );
	}

	return $utf8_pcre;
}

if ( ! function_exists( 'mb_substr' ) ) :
	/**
	 * Compat function to mimic mb_substr().
	 *
	 * @ignore
	 * @since 3.2.0
	 *
	 * @see _mb_substr()
	 *
	 * @param string      $str      The string to extract the substring from.
	 * @param int         $start    Position to being extraction from in `$str`.
	 * @param int|null    $length   Optional. Maximum number of characters to extract from `$str`.
	 *                              Default null.
	 * @param string|null $encoding Optional. Character encoding to use. Default null.
	 * @return string Extracted substring.
	 */
	function mb_substr( $str, $start, $length = null, $encoding = null ) {
		return _mb_substr( $str, $start, $length, $encoding );
	}
endif;

/**
 * Internal compat function to mimic mb_substr().
 *
 * Only understands UTF-8 and 8bit.  All other character sets will be treated as 8bit.
 * For $encoding === UTF-8, the $str input is expected to be a valid UTF-8 byte sequence.
 * The behavior of this function for invalid inputs is undefined.
 *
 * @ignore
 * @since 3.2.0
 *
 * @param string      $str      The string to extract the substring from.
 * @param int         $start    Position to being extraction from in `$str`.
 * @param int|null    $length   Optional. Maximum number of characters to extract from `$str`.
 *                              Default null.
 * @param string|null $encoding Optional. Character encoding to use. Default null.
 * @return string Extracted substring.
 */
function _mb_substr( $str, $start, $length = null, $encoding = null ) {
	if ( null === $str ) {
		return '';
	}

	if ( null === $encoding ) {
		$encoding = get_option( 'blog_charset' );
	}

	/*
	 * The solution below works only for UTF-8, so in case of a different
	 * charset just use built-in substr().
	 */
	if ( ! in_array( $encoding, array( 'utf8', 'utf-8', 'UTF8', 'UTF-8' ), true ) ) {
		return is_null( $length ) ? substr( $str, $start ) : substr( $str, $start, $length );
	}

	if ( _wp_can_use_pcre_u() ) {
		// Use the regex unicode support to separate the UTF-8 characters into an array.
		preg_match_all( '/./us', $str, $match );
		$chars = is_null( $length ) ? array_slice( $match[0], $start ) : array_slice( $match[0], $start, $length );
		return implode( '', $chars );
	}

	$regex = '/(
		[\x00-\x7F]                  # single-byte sequences   0xxxxxxx
		| [\xC2-\xDF][\x80-\xBF]       # double-byte sequences   110xxxxx 10xxxxxx
		| \xE0[\xA0-\xBF][\x80-\xBF]   # triple-byte sequences   1110xxxx 10xxxxxx * 2
		| [\xE1-\xEC][\x80-\xBF]{2}
		| \xED[\x80-\x9F][\x80-\xBF]
		| [\xEE-\xEF][\x80-\xBF]{2}
		| \xF0[\x90-\xBF][\x80-\xBF]{2} # four-byte sequences   11110xxx 10xxxxxx * 3
		| [\xF1-\xF3][\x80-\xBF]{3}
		| \xF4[\x80-\x8F][\x80-\xBF]{2}
	)/x';

	// Start with 1 element instead of 0 since the first thing we do is pop.
	$chars = array( '' );
	do {
		// We had some string left over from the last round, but we counted it in that last round.
		array_pop( $chars );

		/*
		 * Split by UTF-8 character, limit to 1000 characters (last array element will contain
		 * the rest of the string).
		 */
		$pieces = preg_split( $regex, $str, 1000, PREG_SPLIT_DELIM_CAPTURE | PREG_SPLIT_NO_EMPTY );

		$chars = array_merge( $chars, $pieces );

		// If there's anything left over, repeat the loop.
	} while ( count( $pieces ) > 1 && $str = array_pop( $pieces ) );

	return implode( '', array_slice( $chars, $start, $length ) );
}

if ( ! function_exists( 'mb_strlen' ) ) :
	/**
	 * Compat function to mimic mb_strlen().
	 *
	 * @ignore
	 * @since 4.2.0
	 *
	 * @see _mb_strlen()
	 *
	 * @param string      $str      The string to retrieve the character length from.
	 * @param string|null $encoding Optional. Character encoding to use. Default null.
	 * @return int String length of `$str`.
	 */
	function mb_strlen( $str, $encoding = null ) {
		return _mb_strlen( $str, $encoding );
	}
endif;

/**
 * Internal compat function to mimic mb_strlen().
 *
 * Only understands UTF-8 and 8bit.  All other character sets will be treated as 8bit.
 * For $encoding === UTF-8, the `$str` input is expected to be a valid UTF-8 byte
 * sequence. The behavior of this function for invalid inputs is undefined.
 *
 * @ignore
 * @since 4.2.0
 *
 * @param string      $str      The string to retrieve the character length from.
 * @param string|null $encoding Optional. Character encoding to use. Default null.
 * @return int String length of `$str`.
 */
function _mb_strlen( $str, $encoding = null ) {
	if ( null === $encoding ) {
		$encoding = get_option( 'blog_charset' );
	}

	/*
	 * The solution below works only for UTF-8, so in case of a different charset
	 * just use built-in strlen().
	 */
	if ( ! in_array( $encoding, array( 'utf8', 'utf-8', 'UTF8', 'UTF-8' ), true ) ) {
		return strlen( $str );
	}

	if ( _wp_can_use_pcre_u() ) {
		// Use the regex unicode support to separate the UTF-8 characters into an array.
		preg_match_all( '/./us', $str, $match );
		return count( $match[0] );
	}

	$regex = '/(?:
		[\x00-\x7F]                  # single-byte sequences   0xxxxxxx
		| [\xC2-\xDF][\x80-\xBF]       # double-byte sequences   110xxxxx 10xxxxxx
		| \xE0[\xA0-\xBF][\x80-\xBF]   # triple-byte sequences   1110xxxx 10xxxxxx * 2
		| [\xE1-\xEC][\x80-\xBF]{2}
		| \xED[\x80-\x9F][\x80-\xBF]
		| [\xEE-\xEF][\x80-\xBF]{2}
		| \xF0[\x90-\xBF][\x80-\xBF]{2} # four-byte sequences   11110xxx 10xxxxxx * 3
		| [\xF1-\xF3][\x80-\xBF]{3}
		| \xF4[\x80-\x8F][\x80-\xBF]{2}
	)/x';

	// Start at 1 instead of 0 since the first thing we do is decrement.
	$count = 1;
	do {
		// We had some string left over from the last round, but we counted it in that last round.
		$count--;

		/*
		 * Split by UTF-8 character, limit to 1000 characters (last array element will contain
		 * the rest of the string).
		 */
		$pieces = preg_split( $regex, $str, 1000 );

		// Increment.
		$count += count( $pieces );

		// If there's anything left over, repeat the loop.
	} while ( $str = array_pop( $pieces ) );

	// Fencepost: preg_split() always returns one extra item in the array.
	return --$count;
}

if ( ! function_exists( 'hash_hmac' ) ) :
	/**
	 * Compat function to mimic hash_hmac().
	 *
	 * The Hash extension is bundled with PHP by default since PHP 5.1.2.
	 * However, the extension may be explicitly disabled on select servers.
	 * As of PHP 7.4.0, the Hash extension is a core PHP extension and can no
	 * longer be disabled.
	 * I.e. when PHP 7.4.0 becomes the minimum requirement, this polyfill
	 * and the associated `_hash_hmac()` function can be safely removed.
	 *
	 * @ignore
	 * @since 3.2.0
	 *
	 * @see _hash_hmac()
	 *
	 * @param string $algo       Hash algorithm. Accepts 'md5' or 'sha1'.
	 * @param string $data       Data to be hashed.
	 * @param string $key        Secret key to use for generating the hash.
	 * @param bool   $raw_output Optional. Whether to output raw binary data (true),
	 *                           or lowercase hexits (false). Default false.
	 * @return string|false The hash in output determined by `$raw_output`. False if `$algo`
	 *                      is unknown or invalid.
	 */
	function hash_hmac( $algo, $data, $key, $raw_output = false ) {
		return _hash_hmac( $algo, $data, $key, $raw_output );
	}
endif;

/**
 * Internal compat function to mimic hash_hmac().
 *
 * @ignore
 * @since 3.2.0
 *
 * @param string $algo       Hash algorithm. Accepts 'md5' or 'sha1'.
 * @param string $data       Data to be hashed.
 * @param string $key        Secret key to use for generating the hash.
 * @param bool   $raw_output Optional. Whether to output raw binary data (true),
 *                           or lowercase hexits (false). Default false.
 * @return string|false The hash in output determined by `$raw_output`. False if `$algo`
 *                      is unknown or invalid.
 */
function _hash_hmac( $algo, $data, $key, $raw_output = false ) {
	$packs = array(
		'md5'  => 'H32',
		'sha1' => 'H40',
	);

	if ( ! isset( $packs[ $algo ] ) ) {
		return false;
	}

	$pack = $packs[ $algo ];

	if ( strlen( $key ) > 64 ) {
		$key = pack( $pack, $algo( $key ) );
	}

	$key = str_pad( $key, 64, chr( 0 ) );

	$ipad = ( substr( $key, 0, 64 ) ^ str_repeat( chr( 0x36 ), 64 ) );
	$opad = ( substr( $key, 0, 64 ) ^ str_repeat( chr( 0x5C ), 64 ) );

	$hmac = $algo( $opad . pack( $pack, $algo( $ipad . $data ) ) );

	if ( $raw_output ) {
		return pack( $pack, $hmac );
	}
	return $hmac;
}

if ( ! function_exists( 'hash_equals' ) ) :
	/**
	 * Timing attack safe string comparison
	 *
	 * Compares two strings using the same time whether they're equal or not.
	 *
	 * Note: It can leak the length of a string when arguments of differing length are supplied.
	 *
	 * This function was added in PHP 5.6.
	 * However, the Hash extension may be explicitly disabled on select servers.
	 * As of PHP 7.4.0, the Hash extension is a core PHP extension and can no
	 * longer be disabled.
	 * I.e. when PHP 7.4.0 becomes the minimum requirement, this polyfill
	 * can be safely removed.
	 *
	 * @since 3.9.2
	 *
	 * @param string $a Expected string.
	 * @param string $b Actual, user supplied, string.
	 * @return bool Whether strings are equal.
	 */
	function hash_equals( $a, $b ) {
		$a_length = strlen( $a );
		if ( strlen( $b ) !== $a_length ) {
			return false;
		}
		$result = 0;

		// Do not attempt to "optimize" this.
		for ( $i = 0; $i < $a_length; $i++ ) {
			$result |= ord( $a[ $i ] ) ^ ord( $b[ $i ] );
		}

		return 0 === $result;
	}
endif;

// random_int() was introduced in PHP 7.0.
if ( ! function_exists( 'random_int' ) ) {
	require ABSPATH . WPINC . '/random_compat/random.php';
}
// sodium_crypto_box() was introduced in PHP 7.2.
if ( ! function_exists( 'sodium_crypto_box' ) ) {
	require ABSPATH . WPINC . '/sodium_compat/autoload.php';
}

if ( ! function_exists( 'is_countable' ) ) {
	/**
	 * Polyfill for is_countable() function added in PHP 7.3.
	 *
	 * Verify that the content of a variable is an array or an object
	 * implementing the Countable interface.
	 *
	 * @since 4.9.6
	 *
	 * @param mixed $var The value to check.
	 * @return bool True if `$var` is countable, false otherwise.
	 */
	function is_countable( $var ) {
		return ( is_array( $var )
			|| $var instanceof Countable
			|| $var instanceof SimpleXMLElement
			|| $var instanceof ResourceBundle
		);
	}
}

if ( ! function_exists( 'is_iterable' ) ) {
	/**
	 * Polyfill for is_iterable() function added in PHP 7.1.
	 *
	 * Verify that the content of a variable is an array or an object
	 * implementing the Traversable interface.
	 *
	 * @since 4.9.6
	 *
	 * @param mixed $var The value to check.
	 * @return bool True if `$var` is iterable, false otherwise.
	 */
	function is_iterable( $var ) {
		return ( is_array( $var ) || $var instanceof Traversable );
	}
}

if ( ! function_exists( 'array_key_first' ) ) {
	/**
	 * Polyfill for array_key_first() function added in PHP 7.3.
	 *
	 * Get the first key of the given array without affecting
	 * the internal array pointer.
	 *
	 * @since 5.9.0
	 *
	 * @param array $arr An array.
	 * @return string|int|null The first key of array if the array
	 *                         is not empty; `null` otherwise.
	 */
	function array_key_first( array $arr ) {
		foreach ( $arr as $key => $value ) {
			return $key;
		}
	}
}

if ( ! function_exists( 'array_key_last' ) ) {
	/**
	 * Polyfill for `array_key_last()` function added in PHP 7.3.
	 *
	 * Get the last key of the given array without affecting the
	 * internal array pointer.
	 *
	 * @since 5.9.0
	 *
	 * @param array $arr An array.
	 * @return string|int|null The last key of array if the array
	 *.                        is not empty; `null` otherwise.
	 */
	function array_key_last( array $arr ) {
		if ( empty( $arr ) ) {
			return null;
		}
		end( $arr );
		return key( $arr );
	}
}

if ( ! function_exists( 'str_contains' ) ) {
	/**
	 * Polyfill for `str_contains()` function added in PHP 8.0.
	 *
	 * Performs a case-sensitive check indicating if needle is
	 * contained in haystack.
	 *
	 * @since 5.9.0
	 *
	 * @param string $haystack The string to search in.
	 * @param string $needle   The substring to search for in the haystack.
	 * @return bool True if `$needle` is in `$haystack`, otherwise false.
	 */
	function str_contains( $haystack, $needle ) {
		return ( '' === $needle || false !== strpos( $haystack, $needle ) );
	}
}

if ( ! function_exists( 'str_starts_with' ) ) {
	/**
	 * Polyfill for `str_starts_with()` function added in PHP 8.0.
	 *
	 * Performs a case-sensitive check indicating if
	 * the haystack begins with needle.
	 *
	 * @since 5.9.0
	 *
	 * @param string $haystack The string to search in.
	 * @param string $needle   The substring to search for in the `$haystack`.
	 * @return bool True if `$haystack` starts with `$needle`, otherwise false.
	 */
	function str_starts_with( $haystack, $needle ) {
		if ( '' === $needle ) {
			return true;
		}
		return 0 === strpos( $haystack, $needle );
	}
}

if ( ! function_exists( 'str_ends_with' ) ) {
	/**
	 * Polyfill for `str_ends_with()` function added in PHP 8.0.
	 *
	 * Performs a case-sensitive check indicating if
	 * the haystack ends with needle.
	 *
	 * @since 5.9.0
	 *
	 * @param string $haystack The string to search in.
	 * @param string $needle   The substring to search for in the `$haystack`.
	 * @return bool True if `$haystack` ends with `$needle`, otherwise false.
	 */
	function str_ends_with( $haystack, $needle ) {
		if ( '' === $haystack && '' !== $needle ) {
			return false;
		}
		$len = strlen( $needle );
		return 0 === substr_compare( $haystack, $needle, -$len, $len );
	}
}

// IMAGETYPE_WEBP constant is only defined in PHP 7.1 or later.
if ( ! defined( 'IMAGETYPE_WEBP' ) ) {
	define( 'IMAGETYPE_WEBP', 18 );
}

// IMG_WEBP constant is only defined in PHP 7.0.10 or later.
if ( ! defined( 'IMG_WEBP' ) ) {
	define( 'IMG_WEBP', IMAGETYPE_WEBP );
}
 

PHP Online Compiler

Write, Run & Share PHP code online using OneCompiler's PHP online compiler for free. It's one of the robust, feature-rich online compilers for PHP language, running on the latest version 7. Getting started with the OneCompiler's PHP compiler is simple and pretty fast. The editor shows sample boilerplate code when you choose language as PHP and start coding.

Taking inputs (stdin)

OneCompiler's PHP online editor supports stdin and users can give inputs to programs using the STDIN textbox under the I/O tab. Following is a sample PHP program which takes name as input and prints hello message with your name.

<?php
	fscanf(STDIN, "%s\n", $name);           
    echo "Hello ".$name.".\n";
?>

About PHP

PHP(Hypertext Preprocessor) is widely used server sripting language by Rasmus Lerdorf in the year 1994.

Key features

  • Free
  • powerful tool for making dynamic and interactive web pages
  • can integrate with almost all popular databases like MySQL, PostgreSQL, Oracle, Sybase, Informix, Microsoft SQL Server etc.
  • C like Syntax and easy to learn.
  • Object oriented scripting language.
  • easily embeddable into HTML
  • Loosely typed language.

Syntax help

Variables

In PHP, there is no need to explicitly declare variables to reserve memory space. When you assign a value to a variable, declaration happens automatically. Variables are case-sensitive in PHP.

$variable_name = value;  

Loops

1. IF Family:

If, If-else, Nested-Ifs are used when you want to perform a certain set of operations based on conditional expressions.

If

if(conditional-expression){    
//code    
} 

If-else

if(conditional-expression){  
//code if condition is true  
} else {  
//code if condition is false  
} 

Nested-If-else

if(condition-expression1) {  
    //code if above condition is true  
} elseif(condition-expression2){  
    //code if above condition is true  
}  
elseif(condition-expression3) {  
    //code if above condition is true  
}  
...  
else {  
    //code if all the conditions are false  
}  

2. Switch:

Switch is used to execute one set of statement from multiple conditions.

switch(conditional-expression) {    
case value1:    
 // code if the above value is matched    
 break;  // optional  
case value2:    
 // code if the above value is matched    
 break;  // optional  
...    
    
default:     
 // code to be executed when all the above cases are not matched;    
} 
 

3. For:

For loop is used to iterate a set of statements based on a condition.

for(Initialization; Condition; Increment/decrement){  
  // code  
} 

For-each:

// you can use any of the below syntax
foreach ($array as $element-value) {  
    //code  
}

foreach ($array as $key => $element-value) {   
    //code 
} 

4. While:

While is also used to iterate a set of statements based on a condition. Usually while is preferred when number of iterations are not known in advance.

while(condition) {  
 // code 
}  

5. Do-While:

Do-while is also used to iterate a set of statements based on a condition. It is mostly used when you need to execute the statements atleast once.

do {
  // code 
} while (condition); 

Functions

Function is a sub-routine which contains set of statements. Usually functions are written when multiple calls are required to same set of statements which increases re-usuability and modularity.

How to define a Function

function function_name(parameters) {  
  //code
}

How to call a Function

function_name (parameters)