chr(0), "\x1B 1" => chr(1), "\x1B 2" => chr(2), "\x1B 3" => chr(3), "\x1B 4" => chr(4), "\x1B 5" => chr(5), "\x1B 6" => chr(6), "\x1B 7" => chr(7), "\x1B 8" => chr(8), "\x1B 9" => chr(9), "\x1B :" => chr(10), "\x1B ;" => chr(11), "\x1B <" => chr(12), "\x1B :" => chr(13), "\x1B >" => chr(14), "\x1B ?" => chr(15), "\x1B!0" => chr(16), "\x1B!1" => chr(17), "\x1B!2" => chr(18), "\x1B!3" => chr(19), "\x1B!4" => chr(20), "\x1B!5" => chr(21), "\x1B!6" => chr(22), "\x1B!7" => chr(23), "\x1B!8" => chr(24), "\x1B!9" => chr(25), "\x1B!:" => chr(26), "\x1B!;" => chr(27), "\x1B!<" => chr(28), "\x1B!=" => chr(29), "\x1B!>" => chr(30), "\x1B!?" => chr(31), "\x1B'?" => chr(127), "\x1B(0" => '€', // 128 in CP1252 "\x1B(2" => '‚', // 130 in CP1252 "\x1B(3" => 'ƒ', // 131 in CP1252 "\x1B(4" => '„', // 132 in CP1252 "\x1B(5" => '…', // 133 in CP1252 "\x1B(6" => '†', // 134 in CP1252 "\x1B(7" => '‡', // 135 in CP1252 "\x1B(8" => 'ˆ', // 136 in CP1252 "\x1B(9" => '‰', // 137 in CP1252 "\x1B(:" => 'Š', // 138 in CP1252 "\x1B(;" => '‹', // 139 in CP1252 "\x1BNj" => 'Œ', // 140 in CP1252 "\x1B(>" => 'Ž', // 142 in CP1252 "\x1B)1" => '‘', // 145 in CP1252 "\x1B)2" => '’', // 146 in CP1252 "\x1B)3" => '“', // 147 in CP1252 "\x1B)4" => '”', // 148 in CP1252 "\x1B)5" => '•', // 149 in CP1252 "\x1B)6" => '–', // 150 in CP1252 "\x1B)7" => '—', // 151 in CP1252 "\x1B)8" => '˜', // 152 in CP1252 "\x1B)9" => '™', // 153 in CP1252 "\x1B):" => 'š', // 154 in CP1252 "\x1B);" => '›', // 155 in CP1252 "\x1BNz" => 'œ', // 156 in CP1252 "\x1B)>" => 'ž', // 158 in CP1252 "\x1B)?" => 'Ÿ', // 159 in CP1252 "\x1B*0" => ' ', // 160 in CP1252 "\x1BN!" => '¡', // 161 in CP1252 "\x1BN\"" => '¢', // 162 in CP1252 "\x1BN#" => '£', // 163 in CP1252 "\x1BN(" => '¤', // 164 in CP1252 "\x1BN%" => '¥', // 165 in CP1252 "\x1B*6" => '¦', // 166 in CP1252 "\x1BN'" => '§', // 167 in CP1252 "\x1BNH " => '¨', // 168 in CP1252 "\x1BNS" => '©', // 169 in CP1252 "\x1BNc" => 'ª', // 170 in CP1252 "\x1BN+" => '«', // 171 in CP1252 "\x1B*<" => '¬', // 172 in CP1252 "\x1B*=" => '­', // 173 in CP1252 "\x1BNR" => '®', // 174 in CP1252 "\x1B*?" => '¯', // 175 in CP1252 "\x1BN0" => '°', // 176 in CP1252 "\x1BN1" => '±', // 177 in CP1252 "\x1BN2" => '²', // 178 in CP1252 "\x1BN3" => '³', // 179 in CP1252 "\x1BNB " => '´', // 180 in CP1252 "\x1BN5" => 'µ', // 181 in CP1252 "\x1BN6" => '¶', // 182 in CP1252 "\x1BN7" => '·', // 183 in CP1252 "\x1B+8" => '¸', // 184 in CP1252 "\x1BNQ" => '¹', // 185 in CP1252 "\x1BNk" => 'º', // 186 in CP1252 "\x1BN;" => '»', // 187 in CP1252 "\x1BN<" => '¼', // 188 in CP1252 "\x1BN=" => '½', // 189 in CP1252 "\x1BN>" => '¾', // 190 in CP1252 "\x1BN?" => '¿', // 191 in CP1252 "\x1BNAA" => 'À', // 192 in CP1252 "\x1BNBA" => 'Á', // 193 in CP1252 "\x1BNCA" => 'Â', // 194 in CP1252 "\x1BNDA" => 'Ã', // 195 in CP1252 "\x1BNHA" => 'Ä', // 196 in CP1252 "\x1BNJA" => 'Å', // 197 in CP1252 "\x1BNa" => 'Æ', // 198 in CP1252 "\x1BNKC" => 'Ç', // 199 in CP1252 "\x1BNAE" => 'È', // 200 in CP1252 "\x1BNBE" => 'É', // 201 in CP1252 "\x1BNCE" => 'Ê', // 202 in CP1252 "\x1BNHE" => 'Ë', // 203 in CP1252 "\x1BNAI" => 'Ì', // 204 in CP1252 "\x1BNBI" => 'Í', // 205 in CP1252 "\x1BNCI" => 'Î', // 206 in CP1252 "\x1BNHI" => 'Ï', // 207 in CP1252 "\x1BNb" => 'Ð', // 208 in CP1252 "\x1BNDN" => 'Ñ', // 209 in CP1252 "\x1BNAO" => 'Ò', // 210 in CP1252 "\x1BNBO" => 'Ó', // 211 in CP1252 "\x1BNCO" => 'Ô', // 212 in CP1252 "\x1BNDO" => 'Õ', // 213 in CP1252 "\x1BNHO" => 'Ö', // 214 in CP1252 "\x1B-7" => '×', // 215 in CP1252 "\x1BNi" => 'Ø', // 216 in CP1252 "\x1BNAU" => 'Ù', // 217 in CP1252 "\x1BNBU" => 'Ú', // 218 in CP1252 "\x1BNCU" => 'Û', // 219 in CP1252 "\x1BNHU" => 'Ü', // 220 in CP1252 "\x1B-=" => 'Ý', // 221 in CP1252 "\x1BNl" => 'Þ', // 222 in CP1252 "\x1BN{" => 'ß', // 223 in CP1252 "\x1BNAa" => 'à', // 224 in CP1252 "\x1BNBa" => 'á', // 225 in CP1252 "\x1BNCa" => 'â', // 226 in CP1252 "\x1BNDa" => 'ã', // 227 in CP1252 "\x1BNHa" => 'ä', // 228 in CP1252 "\x1BNJa" => 'å', // 229 in CP1252 "\x1BNq" => 'æ', // 230 in CP1252 "\x1BNKc" => 'ç', // 231 in CP1252 "\x1BNAe" => 'è', // 232 in CP1252 "\x1BNBe" => 'é', // 233 in CP1252 "\x1BNCe" => 'ê', // 234 in CP1252 "\x1BNHe" => 'ë', // 235 in CP1252 "\x1BNAi" => 'ì', // 236 in CP1252 "\x1BNBi" => 'í', // 237 in CP1252 "\x1BNCi" => 'î', // 238 in CP1252 "\x1BNHi" => 'ï', // 239 in CP1252 "\x1BNs" => 'ð', // 240 in CP1252 "\x1BNDn" => 'ñ', // 241 in CP1252 "\x1BNAo" => 'ò', // 242 in CP1252 "\x1BNBo" => 'ó', // 243 in CP1252 "\x1BNCo" => 'ô', // 244 in CP1252 "\x1BNDo" => 'õ', // 245 in CP1252 "\x1BNHo" => 'ö', // 246 in CP1252 "\x1B/7" => '÷', // 247 in CP1252 "\x1BNy" => 'ø', // 248 in CP1252 "\x1BNAu" => 'ù', // 249 in CP1252 "\x1BNBu" => 'ú', // 250 in CP1252 "\x1BNCu" => 'û', // 251 in CP1252 "\x1BNHu" => 'ü', // 252 in CP1252 "\x1B/=" => 'ý', // 253 in CP1252 "\x1BN|" => 'þ', // 254 in CP1252 "\x1BNHy" => 'ÿ', // 255 in CP1252 ); } /** * Get whether mbstring extension is available * * @return boolean */ public static function getIsMbstringEnabled() { if (isset(self::$isMbstringEnabled)) { return self::$isMbstringEnabled; } self::$isMbstringEnabled = function_exists('mb_convert_encoding') ? true : false; return self::$isMbstringEnabled; } /** * Get whether iconv extension is available * * @return boolean */ public static function getIsIconvEnabled() { if (isset(self::$isIconvEnabled)) { return self::$isIconvEnabled; } // Fail if iconv doesn't exist if (!function_exists('iconv')) { self::$isIconvEnabled = false; return false; } // Sometimes iconv is not working, and e.g. iconv('UTF-8', 'UTF-16LE', 'x') just returns false, if (!@iconv('UTF-8', 'UTF-16LE', 'x')) { self::$isIconvEnabled = false; return false; } // Sometimes iconv_substr('A', 0, 1, 'UTF-8') just returns false in PHP 5.2.0 // we cannot use iconv in that case either (http://bugs.php.net/bug.php?id=37773) if (!@iconv_substr('A', 0, 1, 'UTF-8')) { self::$isIconvEnabled = false; return false; } // CUSTOM: IBM AIX iconv() does not work if (defined('PHP_OS') && @stristr(PHP_OS, 'AIX') && defined('ICONV_IMPL') && (@strcasecmp(ICONV_IMPL, 'unknown') == 0) && defined('ICONV_VERSION') && (@strcasecmp(ICONV_VERSION, 'unknown') == 0)) { self::$isIconvEnabled = false; return false; } // If we reach here no problems were detected with iconv self::$isIconvEnabled = true; return true; } public static function buildCharacterSets() { if (empty(self::$controlCharacters)) { self::buildControlCharacters(); } if (empty(self::$SYLKCharacters)) { self::buildSYLKCharacters(); } } /** * Convert from OpenXML escaped control character to PHP control character * * Excel 2007 team: * ---------------- * That's correct, control characters are stored directly in the shared-strings table. * We do encode characters that cannot be represented in XML using the following escape sequence: * _xHHHH_ where H represents a hexadecimal character in the character's value... * So you could end up with something like _x0008_ in a string (either in a cell value () * element or in the shared string element. * * @param string $value Value to unescape * @return string */ public static function ControlCharacterOOXML2PHP($value = '') { return str_replace(array_keys(self::$controlCharacters), array_values(self::$controlCharacters), $value); } /** * Convert from PHP control character to OpenXML escaped control character * * Excel 2007 team: * ---------------- * That's correct, control characters are stored directly in the shared-strings table. * We do encode characters that cannot be represented in XML using the following escape sequence: * _xHHHH_ where H represents a hexadecimal character in the character's value... * So you could end up with something like _x0008_ in a string (either in a cell value () * element or in the shared string element. * * @param string $value Value to escape * @return string */ public static function ControlCharacterPHP2OOXML($value = '') { return str_replace(array_values(self::$controlCharacters), array_keys(self::$controlCharacters), $value); } /** * Try to sanitize UTF8, stripping invalid byte sequences. Not perfect. Does not surrogate characters. * * @param string $value * @return string */ public static function SanitizeUTF8($value) { if (self::getIsIconvEnabled()) { $value = @iconv('UTF-8', 'UTF-8', $value); return $value; } if (self::getIsMbstringEnabled()) { $value = mb_convert_encoding($value, 'UTF-8', 'UTF-8'); return $value; } // else, no conversion return $value; } /** * Check if a string contains UTF8 data * * @param string $value * @return boolean */ public static function IsUTF8($value = '') { return $value === '' || preg_match('/^./su', $value) === 1; } /** * Formats a numeric value as a string for output in various output writers forcing * point as decimal separator in case locale is other than English. * * @param mixed $value * @return string */ public static function FormatNumber($value) { if (is_float($value)) { return str_replace(',', '.', $value); } return (string) $value; } /** * Converts a UTF-8 string into BIFF8 Unicode string data (8-bit string length) * Writes the string using uncompressed notation, no rich text, no Asian phonetics * If mbstring extension is not available, ASCII is assumed, and compressed notation is used * although this will give wrong results for non-ASCII strings * see OpenOffice.org's Documentation of the Microsoft Excel File Format, sect. 2.5.3 * * @param string $value UTF-8 encoded string * @param mixed[] $arrcRuns Details of rich text runs in $value * @return string */ public static function UTF8toBIFF8UnicodeShort($value, $arrcRuns = array()) { // character count $ln = self::CountCharacters($value, 'UTF-8'); // option flags if (empty($arrcRuns)) { $opt = (self::getIsIconvEnabled() || self::getIsMbstringEnabled()) ? 0x0001 : 0x0000; $data = pack('CC', $ln, $opt); // characters $data .= self::ConvertEncoding($value, 'UTF-16LE', 'UTF-8'); } else { $data = pack('vC', $ln, 0x09); $data .= pack('v', count($arrcRuns)); // characters $data .= self::ConvertEncoding($value, 'UTF-16LE', 'UTF-8'); foreach ($arrcRuns as $cRun) { $data .= pack('v', $cRun['strlen']); $data .= pack('v', $cRun['fontidx']); } } return $data; } /** * Converts a UTF-8 string into BIFF8 Unicode string data (16-bit string length) * Writes the string using uncompressed notation, no rich text, no Asian phonetics * If mbstring extension is not available, ASCII is assumed, and compressed notation is used * although this will give wrong results for non-ASCII strings * see OpenOffice.org's Documentation of the Microsoft Excel File Format, sect. 2.5.3 * * @param string $value UTF-8 encoded string * @return string */ public static function UTF8toBIFF8UnicodeLong($value) { // character count $ln = self::CountCharacters($value, 'UTF-8'); // option flags $opt = (self::getIsIconvEnabled() || self::getIsMbstringEnabled()) ? 0x0001 : 0x0000; // characters $chars = self::ConvertEncoding($value, 'UTF-16LE', 'UTF-8'); $data = pack('vC', $ln, $opt) . $chars; return $data; } /** * Convert string from one encoding to another. First try mbstring, then iconv, finally strlen * * @param string $value * @param string $to Encoding to convert to, e.g. 'UTF-8' * @param string $from Encoding to convert from, e.g. 'UTF-16LE' * @return string */ public static function ConvertEncoding($value, $to, $from) { if (self::getIsIconvEnabled()) { return iconv($from, $to, $value); } if (self::getIsMbstringEnabled()) { return mb_convert_encoding($value, $to, $from); } if ($from == 'UTF-16LE') { return self::utf16_decode($value, false); } elseif ($from == 'UTF-16BE') { return self::utf16_decode($value); } // else, no conversion return $value; } /** * Decode UTF-16 encoded strings. * * Can handle both BOM'ed data and un-BOM'ed data. * Assumes Big-Endian byte order if no BOM is available. * This function was taken from http://php.net/manual/en/function.utf8-decode.php * and $bom_be parameter added. * * @param string $str UTF-16 encoded data to decode. * @return string UTF-8 / ISO encoded data. * @access public * @version 0.2 / 2010-05-13 * @author Rasmus Andersson {@link http://rasmusandersson.se/} * @author vadik56 */ public static function utf16_decode($str, $bom_be = true) { if (strlen($str) < 2) { return $str; } $c0 = ord($str[0]); $c1 = ord($str[1]); if ($c0 == 0xfe && $c1 == 0xff) { $str = substr($str, 2); } elseif ($c0 == 0xff && $c1 == 0xfe) { $str = substr($str, 2); $bom_be = false; } $len = strlen($str); $newstr = ''; for ($i=0; $i<$len; $i+=2) { if ($bom_be) { $val = ord($str[$i]) << 4; $val += ord($str[$i+1]); } else { $val = ord($str[$i+1]) << 4; $val += ord($str[$i]); } $newstr .= ($val == 0x228) ? "\n" : chr($val); } return $newstr; } /** * Get character count. First try mbstring, then iconv, finally strlen * * @param string $value * @param string $enc Encoding * @return int Character count */ public static function CountCharacters($value, $enc = 'UTF-8') { if (self::getIsMbstringEnabled()) { return mb_strlen($value, $enc); } if (self::getIsIconvEnabled()) { return iconv_strlen($value, $enc); } // else strlen return strlen($value); } /** * Get a substring of a UTF-8 encoded string. First try mbstring, then iconv, finally strlen * * @param string $pValue UTF-8 encoded string * @param int $pStart Start offset * @param int $pLength Maximum number of characters in substring * @return string */ public static function Substring($pValue = '', $pStart = 0, $pLength = 0) { if (self::getIsMbstringEnabled()) { return mb_substr($pValue, $pStart, $pLength, 'UTF-8'); } if (self::getIsIconvEnabled()) { return iconv_substr($pValue, $pStart, $pLength, 'UTF-8'); } // else substr return substr($pValue, $pStart, $pLength); } /** * Convert a UTF-8 encoded string to upper case * * @param string $pValue UTF-8 encoded string * @return string */ public static function StrToUpper($pValue = '') { if (function_exists('mb_convert_case')) { return mb_convert_case($pValue, MB_CASE_UPPER, "UTF-8"); } return strtoupper($pValue); } /** * Convert a UTF-8 encoded string to lower case * * @param string $pValue UTF-8 encoded string * @return string */ public static function StrToLower($pValue = '') { if (function_exists('mb_convert_case')) { return mb_convert_case($pValue, MB_CASE_LOWER, "UTF-8"); } return strtolower($pValue); } /** * Convert a UTF-8 encoded string to title/proper case * (uppercase every first character in each word, lower case all other characters) * * @param string $pValue UTF-8 encoded string * @return string */ public static function StrToTitle($pValue = '') { if (function_exists('mb_convert_case')) { return mb_convert_case($pValue, MB_CASE_TITLE, "UTF-8"); } return ucwords($pValue); } public static function mb_is_upper($char) { return mb_strtolower($char, "UTF-8") != $char; } public static function mb_str_split($string) { # Split at all position not after the start: ^ # and not before the end: $ return preg_split('/(?_calculateFormulaValue($fractionFormula); return true; } return false; } // function convertToNumberIfFraction() /** * Get the decimal separator. If it has not yet been set explicitly, try to obtain number * formatting information from locale. * * @return string */ public static function getDecimalSeparator() { if (!isset(self::$decimalSeparator)) { $localeconv = localeconv(); self::$decimalSeparator = ($localeconv['decimal_point'] != '') ? $localeconv['decimal_point'] : $localeconv['mon_decimal_point']; if (self::$decimalSeparator == '') { // Default to . self::$decimalSeparator = '.'; } } return self::$decimalSeparator; } /** * Set the decimal separator. Only used by PHPExcel_Style_NumberFormat::toFormattedString() * to format output by PHPExcel_Writer_HTML and PHPExcel_Writer_PDF * * @param string $pValue Character for decimal separator */ public static function setDecimalSeparator($pValue = '.') { self::$decimalSeparator = $pValue; } /** * Get the thousands separator. If it has not yet been set explicitly, try to obtain number * formatting information from locale. * * @return string */ public static function getThousandsSeparator() { if (!isset(self::$thousandsSeparator)) { $localeconv = localeconv(); self::$thousandsSeparator = ($localeconv['thousands_sep'] != '') ? $localeconv['thousands_sep'] : $localeconv['mon_thousands_sep']; if (self::$thousandsSeparator == '') { // Default to . self::$thousandsSeparator = ','; } } return self::$thousandsSeparator; } /** * Set the thousands separator. Only used by PHPExcel_Style_NumberFormat::toFormattedString() * to format output by PHPExcel_Writer_HTML and PHPExcel_Writer_PDF * * @param string $pValue Character for thousands separator */ public static function setThousandsSeparator($pValue = ',') { self::$thousandsSeparator = $pValue; } /** * Get the currency code. If it has not yet been set explicitly, try to obtain the * symbol information from locale. * * @return string */ public static function getCurrencyCode() { if (!isset(self::$currencyCode)) { $localeconv = localeconv(); self::$currencyCode = ($localeconv['currency_symbol'] != '') ? $localeconv['currency_symbol'] : $localeconv['int_curr_symbol']; if (self::$currencyCode == '') { // Default to $ self::$currencyCode = '$'; } } return self::$currencyCode; } /** * Set the currency code. Only used by PHPExcel_Style_NumberFormat::toFormattedString() * to format output by PHPExcel_Writer_HTML and PHPExcel_Writer_PDF * * @param string $pValue Character for currency code */ public static function setCurrencyCode($pValue = '$') { self::$currencyCode = $pValue; } /** * Convert SYLK encoded string to UTF-8 * * @param string $pValue * @return string UTF-8 encoded string */ public static function SYLKtoUTF8($pValue = '') { // If there is no escape character in the string there is nothing to do if (strpos($pValue, '') === false) { return $pValue; } foreach (self::$SYLKCharacters as $k => $v) { $pValue = str_replace($k, $v, $pValue); } return $pValue; } /** * Retrieve any leading numeric part of a string, or return the full string if no leading numeric * (handles basic integer or float, but not exponent or non decimal) * * @param string $value * @return mixed string or only the leading numeric part of the string */ public static function testStringAsNumeric($value) { if (is_numeric($value)) { return $value; } $v = floatval($value); return (is_numeric(substr($value, 0, strlen($v)))) ? $v : $value; } }