Blame | Letzte Änderung | Log anzeigen | RSS feed
<?php/** This file is part of the Symfony package.** (c) Fabien Potencier <fabien@symfony.com>** For the full copyright and license information, please view the LICENSE* file that was distributed with this source code.*/namespace Symfony\Component\String;use Symfony\Component\String\Exception\ExceptionInterface;use Symfony\Component\String\Exception\InvalidArgumentException;use Symfony\Component\String\Exception\RuntimeException;/*** Represents a binary-safe string of bytes.** @author Nicolas Grekas <p@tchwork.com>* @author Hugo Hamon <hugohamon@neuf.fr>** @throws ExceptionInterface*/class ByteString extends AbstractString{private const ALPHABET_ALPHANUMERIC = '123456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefghijkmnopqrstuvwxyz';public function __construct(string $string = ''){$this->string = $string;}/** The following method was derived from code of the Hack Standard Library (v4.40 - 2020-05-03)** https://github.com/hhvm/hsl/blob/80a42c02f036f72a42f0415e80d6b847f4bf62d5/src/random/private.php#L16** Code subject to the MIT license (https://github.com/hhvm/hsl/blob/master/LICENSE).** Copyright (c) 2004-2020, Facebook, Inc. (https://www.facebook.com/)*/public static function fromRandom(int $length = 16, string $alphabet = null): self{if ($length <= 0) {throw new InvalidArgumentException(sprintf('A strictly positive length is expected, "%d" given.', $length));}$alphabet ??= self::ALPHABET_ALPHANUMERIC;$alphabetSize = \strlen($alphabet);$bits = (int) ceil(log($alphabetSize, 2.0));if ($bits <= 0 || $bits > 56) {throw new InvalidArgumentException('The length of the alphabet must in the [2^1, 2^56] range.');}$ret = '';while ($length > 0) {$urandomLength = (int) ceil(2 * $length * $bits / 8.0);$data = random_bytes($urandomLength);$unpackedData = 0;$unpackedBits = 0;for ($i = 0; $i < $urandomLength && $length > 0; ++$i) {// Unpack 8 bits$unpackedData = ($unpackedData << 8) | \ord($data[$i]);$unpackedBits += 8;// While we have enough bits to select a character from the alphabet, keep// consuming the random datafor (; $unpackedBits >= $bits && $length > 0; $unpackedBits -= $bits) {$index = ($unpackedData & ((1 << $bits) - 1));$unpackedData >>= $bits;// Unfortunately, the alphabet size is not necessarily a power of two.// Worst case, it is 2^k + 1, which means we need (k+1) bits and we// have around a 50% chance of missing as k gets largerif ($index < $alphabetSize) {$ret .= $alphabet[$index];--$length;}}}}return new static($ret);}public function bytesAt(int $offset): array{$str = $this->string[$offset] ?? '';return '' === $str ? [] : [\ord($str)];}public function append(string ...$suffix): static{$str = clone $this;$str->string .= 1 >= \count($suffix) ? ($suffix[0] ?? '') : implode('', $suffix);return $str;}public function camel(): static{$str = clone $this;$parts = explode(' ', trim(ucwords(preg_replace('/[^a-zA-Z0-9\x7f-\xff]++/', ' ', $this->string))));$parts[0] = 1 !== \strlen($parts[0]) && ctype_upper($parts[0]) ? $parts[0] : lcfirst($parts[0]);$str->string = implode('', $parts);return $str;}public function chunk(int $length = 1): array{if (1 > $length) {throw new InvalidArgumentException('The chunk length must be greater than zero.');}if ('' === $this->string) {return [];}$str = clone $this;$chunks = [];foreach (str_split($this->string, $length) as $chunk) {$str->string = $chunk;$chunks[] = clone $str;}return $chunks;}public function endsWith(string|iterable|AbstractString $suffix): bool{if ($suffix instanceof AbstractString) {$suffix = $suffix->string;} elseif (!\is_string($suffix)) {return parent::endsWith($suffix);}return '' !== $suffix && \strlen($this->string) >= \strlen($suffix) && 0 === substr_compare($this->string, $suffix, -\strlen($suffix), null, $this->ignoreCase);}public function equalsTo(string|iterable|AbstractString $string): bool{if ($string instanceof AbstractString) {$string = $string->string;} elseif (!\is_string($string)) {return parent::equalsTo($string);}if ('' !== $string && $this->ignoreCase) {return 0 === strcasecmp($string, $this->string);}return $string === $this->string;}public function folded(): static{$str = clone $this;$str->string = strtolower($str->string);return $str;}public function indexOf(string|iterable|AbstractString $needle, int $offset = 0): ?int{if ($needle instanceof AbstractString) {$needle = $needle->string;} elseif (!\is_string($needle)) {return parent::indexOf($needle, $offset);}if ('' === $needle) {return null;}$i = $this->ignoreCase ? stripos($this->string, $needle, $offset) : strpos($this->string, $needle, $offset);return false === $i ? null : $i;}public function indexOfLast(string|iterable|AbstractString $needle, int $offset = 0): ?int{if ($needle instanceof AbstractString) {$needle = $needle->string;} elseif (!\is_string($needle)) {return parent::indexOfLast($needle, $offset);}if ('' === $needle) {return null;}$i = $this->ignoreCase ? strripos($this->string, $needle, $offset) : strrpos($this->string, $needle, $offset);return false === $i ? null : $i;}public function isUtf8(): bool{return '' === $this->string || preg_match('//u', $this->string);}public function join(array $strings, string $lastGlue = null): static{$str = clone $this;$tail = null !== $lastGlue && 1 < \count($strings) ? $lastGlue.array_pop($strings) : '';$str->string = implode($this->string, $strings).$tail;return $str;}public function length(): int{return \strlen($this->string);}public function lower(): static{$str = clone $this;$str->string = strtolower($str->string);return $str;}public function match(string $regexp, int $flags = 0, int $offset = 0): array{$match = ((\PREG_PATTERN_ORDER | \PREG_SET_ORDER) & $flags) ? 'preg_match_all' : 'preg_match';if ($this->ignoreCase) {$regexp .= 'i';}set_error_handler(static function ($t, $m) { throw new InvalidArgumentException($m); });try {if (false === $match($regexp, $this->string, $matches, $flags | \PREG_UNMATCHED_AS_NULL, $offset)) {throw new RuntimeException('Matching failed with error: '.preg_last_error_msg());}} finally {restore_error_handler();}return $matches;}public function padBoth(int $length, string $padStr = ' '): static{$str = clone $this;$str->string = str_pad($this->string, $length, $padStr, \STR_PAD_BOTH);return $str;}public function padEnd(int $length, string $padStr = ' '): static{$str = clone $this;$str->string = str_pad($this->string, $length, $padStr, \STR_PAD_RIGHT);return $str;}public function padStart(int $length, string $padStr = ' '): static{$str = clone $this;$str->string = str_pad($this->string, $length, $padStr, \STR_PAD_LEFT);return $str;}public function prepend(string ...$prefix): static{$str = clone $this;$str->string = (1 >= \count($prefix) ? ($prefix[0] ?? '') : implode('', $prefix)).$str->string;return $str;}public function replace(string $from, string $to): static{$str = clone $this;if ('' !== $from) {$str->string = $this->ignoreCase ? str_ireplace($from, $to, $this->string) : str_replace($from, $to, $this->string);}return $str;}public function replaceMatches(string $fromRegexp, string|callable $to): static{if ($this->ignoreCase) {$fromRegexp .= 'i';}$replace = \is_array($to) || $to instanceof \Closure ? 'preg_replace_callback' : 'preg_replace';set_error_handler(static function ($t, $m) { throw new InvalidArgumentException($m); });try {if (null === $string = $replace($fromRegexp, $to, $this->string)) {$lastError = preg_last_error();foreach (get_defined_constants(true)['pcre'] as $k => $v) {if ($lastError === $v && str_ends_with($k, '_ERROR')) {throw new RuntimeException('Matching failed with '.$k.'.');}}throw new RuntimeException('Matching failed with unknown error code.');}} finally {restore_error_handler();}$str = clone $this;$str->string = $string;return $str;}public function reverse(): static{$str = clone $this;$str->string = strrev($str->string);return $str;}public function slice(int $start = 0, int $length = null): static{$str = clone $this;$str->string = (string) substr($this->string, $start, $length ?? \PHP_INT_MAX);return $str;}public function snake(): static{$str = $this->camel();$str->string = strtolower(preg_replace(['/([A-Z]+)([A-Z][a-z])/', '/([a-z\d])([A-Z])/'], '\1_\2', $str->string));return $str;}public function splice(string $replacement, int $start = 0, int $length = null): static{$str = clone $this;$str->string = substr_replace($this->string, $replacement, $start, $length ?? \PHP_INT_MAX);return $str;}public function split(string $delimiter, int $limit = null, int $flags = null): array{if (1 > $limit ??= \PHP_INT_MAX) {throw new InvalidArgumentException('Split limit must be a positive integer.');}if ('' === $delimiter) {throw new InvalidArgumentException('Split delimiter is empty.');}if (null !== $flags) {return parent::split($delimiter, $limit, $flags);}$str = clone $this;$chunks = $this->ignoreCase? preg_split('{'.preg_quote($delimiter).'}iD', $this->string, $limit): explode($delimiter, $this->string, $limit);foreach ($chunks as &$chunk) {$str->string = $chunk;$chunk = clone $str;}return $chunks;}public function startsWith(string|iterable|AbstractString $prefix): bool{if ($prefix instanceof AbstractString) {$prefix = $prefix->string;} elseif (!\is_string($prefix)) {return parent::startsWith($prefix);}return '' !== $prefix && 0 === ($this->ignoreCase ? strncasecmp($this->string, $prefix, \strlen($prefix)) : strncmp($this->string, $prefix, \strlen($prefix)));}public function title(bool $allWords = false): static{$str = clone $this;$str->string = $allWords ? ucwords($str->string) : ucfirst($str->string);return $str;}public function toUnicodeString(string $fromEncoding = null): UnicodeString{return new UnicodeString($this->toCodePointString($fromEncoding)->string);}public function toCodePointString(string $fromEncoding = null): CodePointString{$u = new CodePointString();if (\in_array($fromEncoding, [null, 'utf8', 'utf-8', 'UTF8', 'UTF-8'], true) && preg_match('//u', $this->string)) {$u->string = $this->string;return $u;}set_error_handler(static function ($t, $m) { throw new InvalidArgumentException($m); });try {try {$validEncoding = false !== mb_detect_encoding($this->string, $fromEncoding ?? 'Windows-1252', true);} catch (InvalidArgumentException $e) {if (!\function_exists('iconv')) {throw $e;}$u->string = iconv($fromEncoding ?? 'Windows-1252', 'UTF-8', $this->string);return $u;}} finally {restore_error_handler();}if (!$validEncoding) {throw new InvalidArgumentException(sprintf('Invalid "%s" string.', $fromEncoding ?? 'Windows-1252'));}$u->string = mb_convert_encoding($this->string, 'UTF-8', $fromEncoding ?? 'Windows-1252');return $u;}public function trim(string $chars = " \t\n\r\0\x0B\x0C"): static{$str = clone $this;$str->string = trim($str->string, $chars);return $str;}public function trimEnd(string $chars = " \t\n\r\0\x0B\x0C"): static{$str = clone $this;$str->string = rtrim($str->string, $chars);return $str;}public function trimStart(string $chars = " \t\n\r\0\x0B\x0C"): static{$str = clone $this;$str->string = ltrim($str->string, $chars);return $str;}public function upper(): static{$str = clone $this;$str->string = strtoupper($str->string);return $str;}public function width(bool $ignoreAnsiDecoration = true): int{$string = preg_match('//u', $this->string) ? $this->string : preg_replace('/[\x80-\xFF]/', '?', $this->string);return (new CodePointString($string))->width($ignoreAnsiDecoration);}}