diff options
Diffstat (limited to 'includes/vendor/rmccue/requests/library/Requests/IDNAEncoder.php')
| -rw-r--r-- | includes/vendor/rmccue/requests/library/Requests/IDNAEncoder.php | 388 |
1 files changed, 388 insertions, 0 deletions
diff --git a/includes/vendor/rmccue/requests/library/Requests/IDNAEncoder.php b/includes/vendor/rmccue/requests/library/Requests/IDNAEncoder.php new file mode 100644 index 0000000..ebbe211 --- /dev/null +++ b/includes/vendor/rmccue/requests/library/Requests/IDNAEncoder.php @@ -0,0 +1,388 @@ +<?php + +/** + * IDNA URL encoder + * + * Note: Not fully compliant, as nameprep does nothing yet. + * + * @package Requests + * @subpackage Utilities + * @see https://tools.ietf.org/html/rfc3490 IDNA specification + * @see https://tools.ietf.org/html/rfc3492 Punycode/Bootstrap specification + */ +class Requests_IDNAEncoder { + /** + * ACE prefix used for IDNA + * + * @see https://tools.ietf.org/html/rfc3490#section-5 + * @var string + */ + const ACE_PREFIX = 'xn--'; + + /**#@+ + * Bootstrap constant for Punycode + * + * @see https://tools.ietf.org/html/rfc3492#section-5 + * @var int + */ + const BOOTSTRAP_BASE = 36; + const BOOTSTRAP_TMIN = 1; + const BOOTSTRAP_TMAX = 26; + const BOOTSTRAP_SKEW = 38; + const BOOTSTRAP_DAMP = 700; + const BOOTSTRAP_INITIAL_BIAS = 72; + const BOOTSTRAP_INITIAL_N = 128; + /**#@-*/ + + /** + * Encode a hostname using Punycode + * + * @param string $string Hostname + * @return string Punycode-encoded hostname + */ + public static function encode($string) { + $parts = explode('.', $string); + foreach ($parts as &$part) { + $part = self::to_ascii($part); + } + return implode('.', $parts); + } + + /** + * Convert a UTF-8 string to an ASCII string using Punycode + * + * @throws Requests_Exception Provided string longer than 64 ASCII characters (`idna.provided_too_long`) + * @throws Requests_Exception Prepared string longer than 64 ASCII characters (`idna.prepared_too_long`) + * @throws Requests_Exception Provided string already begins with xn-- (`idna.provided_is_prefixed`) + * @throws Requests_Exception Encoded string longer than 64 ASCII characters (`idna.encoded_too_long`) + * + * @param string $string ASCII or UTF-8 string (max length 64 characters) + * @return string ASCII string + */ + public static function to_ascii($string) { + // Step 1: Check if the string is already ASCII + if (self::is_ascii($string)) { + // Skip to step 7 + if (strlen($string) < 64) { + return $string; + } + + throw new Requests_Exception('Provided string is too long', 'idna.provided_too_long', $string); + } + + // Step 2: nameprep + $string = self::nameprep($string); + + // Step 3: UseSTD3ASCIIRules is false, continue + // Step 4: Check if it's ASCII now + if (self::is_ascii($string)) { + // Skip to step 7 + if (strlen($string) < 64) { + return $string; + } + + throw new Requests_Exception('Prepared string is too long', 'idna.prepared_too_long', $string); + } + + // Step 5: Check ACE prefix + if (strpos($string, self::ACE_PREFIX) === 0) { + throw new Requests_Exception('Provided string begins with ACE prefix', 'idna.provided_is_prefixed', $string); + } + + // Step 6: Encode with Punycode + $string = self::punycode_encode($string); + + // Step 7: Prepend ACE prefix + $string = self::ACE_PREFIX . $string; + + // Step 8: Check size + if (strlen($string) < 64) { + return $string; + } + + throw new Requests_Exception('Encoded string is too long', 'idna.encoded_too_long', $string); + } + + /** + * Check whether a given string contains only ASCII characters + * + * @internal (Testing found regex was the fastest implementation) + * + * @param string $string + * @return bool Is the string ASCII-only? + */ + protected static function is_ascii($string) { + return (preg_match('/(?:[^\x00-\x7F])/', $string) !== 1); + } + + /** + * Prepare a string for use as an IDNA name + * + * @todo Implement this based on RFC 3491 and the newer 5891 + * @param string $string + * @return string Prepared string + */ + protected static function nameprep($string) { + return $string; + } + + /** + * Convert a UTF-8 string to a UCS-4 codepoint array + * + * Based on Requests_IRI::replace_invalid_with_pct_encoding() + * + * @throws Requests_Exception Invalid UTF-8 codepoint (`idna.invalidcodepoint`) + * @param string $input + * @return array Unicode code points + */ + protected static function utf8_to_codepoints($input) { + $codepoints = array(); + + // Get number of bytes + $strlen = strlen($input); + + for ($position = 0; $position < $strlen; $position++) { + $value = ord($input[$position]); + + // One byte sequence: + if ((~$value & 0x80) === 0x80) { + $character = $value; + $length = 1; + $remaining = 0; + } + // Two byte sequence: + elseif (($value & 0xE0) === 0xC0) { + $character = ($value & 0x1F) << 6; + $length = 2; + $remaining = 1; + } + // Three byte sequence: + elseif (($value & 0xF0) === 0xE0) { + $character = ($value & 0x0F) << 12; + $length = 3; + $remaining = 2; + } + // Four byte sequence: + elseif (($value & 0xF8) === 0xF0) { + $character = ($value & 0x07) << 18; + $length = 4; + $remaining = 3; + } + // Invalid byte: + else { + throw new Requests_Exception('Invalid Unicode codepoint', 'idna.invalidcodepoint', $value); + } + + if ($remaining > 0) { + if ($position + $length > $strlen) { + throw new Requests_Exception('Invalid Unicode codepoint', 'idna.invalidcodepoint', $character); + } + for ($position++; $remaining > 0; $position++) { + $value = ord($input[$position]); + + // If it is invalid, count the sequence as invalid and reprocess the current byte: + if (($value & 0xC0) !== 0x80) { + throw new Requests_Exception('Invalid Unicode codepoint', 'idna.invalidcodepoint', $character); + } + + $character |= ($value & 0x3F) << (--$remaining * 6); + } + $position--; + } + + if ( + // Non-shortest form sequences are invalid + $length > 1 && $character <= 0x7F + || $length > 2 && $character <= 0x7FF + || $length > 3 && $character <= 0xFFFF + // Outside of range of ucschar codepoints + // Noncharacters + || ($character & 0xFFFE) === 0xFFFE + || $character >= 0xFDD0 && $character <= 0xFDEF + || ( + // Everything else not in ucschar + $character > 0xD7FF && $character < 0xF900 + || $character < 0x20 + || $character > 0x7E && $character < 0xA0 + || $character > 0xEFFFD + ) + ) { + throw new Requests_Exception('Invalid Unicode codepoint', 'idna.invalidcodepoint', $character); + } + + $codepoints[] = $character; + } + + return $codepoints; + } + + /** + * RFC3492-compliant encoder + * + * @internal Pseudo-code from Section 6.3 is commented with "#" next to relevant code + * @throws Requests_Exception On character outside of the domain (never happens with Punycode) (`idna.character_outside_domain`) + * + * @param string $input UTF-8 encoded string to encode + * @return string Punycode-encoded string + */ + public static function punycode_encode($input) { + $output = ''; +# let n = initial_n + $n = self::BOOTSTRAP_INITIAL_N; +# let delta = 0 + $delta = 0; +# let bias = initial_bias + $bias = self::BOOTSTRAP_INITIAL_BIAS; +# let h = b = the number of basic code points in the input + $h = $b = 0; // see loop +# copy them to the output in order + $codepoints = self::utf8_to_codepoints($input); + $extended = array(); + + foreach ($codepoints as $char) { + if ($char < 128) { + // Character is valid ASCII + // TODO: this should also check if it's valid for a URL + $output .= chr($char); + $h++; + } + // Check if the character is non-ASCII, but below initial n + // This never occurs for Punycode, so ignore in coverage + // @codeCoverageIgnoreStart + elseif ($char < $n) { + throw new Requests_Exception('Invalid character', 'idna.character_outside_domain', $char); + } + // @codeCoverageIgnoreEnd + else { + $extended[$char] = true; + } + } + $extended = array_keys($extended); + sort($extended); + $b = $h; +# [copy them] followed by a delimiter if b > 0 + if (strlen($output) > 0) { + $output .= '-'; + } +# {if the input contains a non-basic code point < n then fail} +# while h < length(input) do begin + while ($h < count($codepoints)) { +# let m = the minimum code point >= n in the input + $m = array_shift($extended); + //printf('next code point to insert is %s' . PHP_EOL, dechex($m)); +# let delta = delta + (m - n) * (h + 1), fail on overflow + $delta += ($m - $n) * ($h + 1); +# let n = m + $n = $m; +# for each code point c in the input (in order) do begin + for ($num = 0; $num < count($codepoints); $num++) { + $c = $codepoints[$num]; +# if c < n then increment delta, fail on overflow + if ($c < $n) { + $delta++; + } +# if c == n then begin + elseif ($c === $n) { +# let q = delta + $q = $delta; +# for k = base to infinity in steps of base do begin + for ($k = self::BOOTSTRAP_BASE; ; $k += self::BOOTSTRAP_BASE) { +# let t = tmin if k <= bias {+ tmin}, or +# tmax if k >= bias + tmax, or k - bias otherwise + if ($k <= ($bias + self::BOOTSTRAP_TMIN)) { + $t = self::BOOTSTRAP_TMIN; + } + elseif ($k >= ($bias + self::BOOTSTRAP_TMAX)) { + $t = self::BOOTSTRAP_TMAX; + } + else { + $t = $k - $bias; + } +# if q < t then break + if ($q < $t) { + break; + } +# output the code point for digit t + ((q - t) mod (base - t)) + $digit = $t + (($q - $t) % (self::BOOTSTRAP_BASE - $t)); + $output .= self::digit_to_char($digit); +# let q = (q - t) div (base - t) + $q = floor(($q - $t) / (self::BOOTSTRAP_BASE - $t)); +# end + } +# output the code point for digit q + $output .= self::digit_to_char($q); +# let bias = adapt(delta, h + 1, test h equals b?) + $bias = self::adapt($delta, $h + 1, $h === $b); +# let delta = 0 + $delta = 0; +# increment h + $h++; +# end + } +# end + } +# increment delta and n + $delta++; + $n++; +# end + } + + return $output; + } + + /** + * Convert a digit to its respective character + * + * @see https://tools.ietf.org/html/rfc3492#section-5 + * @throws Requests_Exception On invalid digit (`idna.invalid_digit`) + * + * @param int $digit Digit in the range 0-35 + * @return string Single character corresponding to digit + */ + protected static function digit_to_char($digit) { + // @codeCoverageIgnoreStart + // As far as I know, this never happens, but still good to be sure. + if ($digit < 0 || $digit > 35) { + throw new Requests_Exception(sprintf('Invalid digit %d', $digit), 'idna.invalid_digit', $digit); + } + // @codeCoverageIgnoreEnd + $digits = 'abcdefghijklmnopqrstuvwxyz0123456789'; + return substr($digits, $digit, 1); + } + + /** + * Adapt the bias + * + * @see https://tools.ietf.org/html/rfc3492#section-6.1 + * @param int $delta + * @param int $numpoints + * @param bool $firsttime + * @return int New bias + */ + protected static function adapt($delta, $numpoints, $firsttime) { +# function adapt(delta,numpoints,firsttime): +# if firsttime then let delta = delta div damp + if ($firsttime) { + $delta = floor($delta / self::BOOTSTRAP_DAMP); + } +# else let delta = delta div 2 + else { + $delta = floor($delta / 2); + } +# let delta = delta + (delta div numpoints) + $delta += floor($delta / $numpoints); +# let k = 0 + $k = 0; +# while delta > ((base - tmin) * tmax) div 2 do begin + $max = floor(((self::BOOTSTRAP_BASE - self::BOOTSTRAP_TMIN) * self::BOOTSTRAP_TMAX) / 2); + while ($delta > $max) { +# let delta = delta div (base - tmin) + $delta = floor($delta / (self::BOOTSTRAP_BASE - self::BOOTSTRAP_TMIN)); +# let k = k + base + $k += self::BOOTSTRAP_BASE; +# end + } +# return k + (((base - tmin + 1) * delta) div (delta + skew)) + return $k + floor(((self::BOOTSTRAP_BASE - self::BOOTSTRAP_TMIN + 1) * $delta) / ($delta + self::BOOTSTRAP_SKEW)); + } +}
\ No newline at end of file |