summaryrefslogtreecommitdiff
path: root/vendor/ezyang/htmlpurifier/library/HTMLPurifier/UnitConverter.php
diff options
context:
space:
mode:
Diffstat (limited to 'vendor/ezyang/htmlpurifier/library/HTMLPurifier/UnitConverter.php')
-rw-r--r--vendor/ezyang/htmlpurifier/library/HTMLPurifier/UnitConverter.php307
1 files changed, 307 insertions, 0 deletions
diff --git a/vendor/ezyang/htmlpurifier/library/HTMLPurifier/UnitConverter.php b/vendor/ezyang/htmlpurifier/library/HTMLPurifier/UnitConverter.php
new file mode 100644
index 0000000..166f3bf
--- /dev/null
+++ b/vendor/ezyang/htmlpurifier/library/HTMLPurifier/UnitConverter.php
@@ -0,0 +1,307 @@
+<?php
+
+/**
+ * Class for converting between different unit-lengths as specified by
+ * CSS.
+ */
+class HTMLPurifier_UnitConverter
+{
+
+ const ENGLISH = 1;
+ const METRIC = 2;
+ const DIGITAL = 3;
+
+ /**
+ * Units information array. Units are grouped into measuring systems
+ * (English, Metric), and are assigned an integer representing
+ * the conversion factor between that unit and the smallest unit in
+ * the system. Numeric indexes are actually magical constants that
+ * encode conversion data from one system to the next, with a O(n^2)
+ * constraint on memory (this is generally not a problem, since
+ * the number of measuring systems is small.)
+ */
+ protected static $units = array(
+ self::ENGLISH => array(
+ 'px' => 3, // This is as per CSS 2.1 and Firefox. Your mileage may vary
+ 'pt' => 4,
+ 'pc' => 48,
+ 'in' => 288,
+ self::METRIC => array('pt', '0.352777778', 'mm'),
+ ),
+ self::METRIC => array(
+ 'mm' => 1,
+ 'cm' => 10,
+ self::ENGLISH => array('mm', '2.83464567', 'pt'),
+ ),
+ );
+
+ /**
+ * Minimum bcmath precision for output.
+ * @type int
+ */
+ protected $outputPrecision;
+
+ /**
+ * Bcmath precision for internal calculations.
+ * @type int
+ */
+ protected $internalPrecision;
+
+ /**
+ * Whether or not BCMath is available.
+ * @type bool
+ */
+ private $bcmath;
+
+ public function __construct($output_precision = 4, $internal_precision = 10, $force_no_bcmath = false)
+ {
+ $this->outputPrecision = $output_precision;
+ $this->internalPrecision = $internal_precision;
+ $this->bcmath = !$force_no_bcmath && function_exists('bcmul');
+ }
+
+ /**
+ * Converts a length object of one unit into another unit.
+ * @param HTMLPurifier_Length $length
+ * Instance of HTMLPurifier_Length to convert. You must validate()
+ * it before passing it here!
+ * @param string $to_unit
+ * Unit to convert to.
+ * @return HTMLPurifier_Length|bool
+ * @note
+ * About precision: This conversion function pays very special
+ * attention to the incoming precision of values and attempts
+ * to maintain a number of significant figure. Results are
+ * fairly accurate up to nine digits. Some caveats:
+ * - If a number is zero-padded as a result of this significant
+ * figure tracking, the zeroes will be eliminated.
+ * - If a number contains less than four sigfigs ($outputPrecision)
+ * and this causes some decimals to be excluded, those
+ * decimals will be added on.
+ */
+ public function convert($length, $to_unit)
+ {
+ if (!$length->isValid()) {
+ return false;
+ }
+
+ $n = $length->getN();
+ $unit = $length->getUnit();
+
+ if ($n === '0' || $unit === false) {
+ return new HTMLPurifier_Length('0', false);
+ }
+
+ $state = $dest_state = false;
+ foreach (self::$units as $k => $x) {
+ if (isset($x[$unit])) {
+ $state = $k;
+ }
+ if (isset($x[$to_unit])) {
+ $dest_state = $k;
+ }
+ }
+ if (!$state || !$dest_state) {
+ return false;
+ }
+
+ // Some calculations about the initial precision of the number;
+ // this will be useful when we need to do final rounding.
+ $sigfigs = $this->getSigFigs($n);
+ if ($sigfigs < $this->outputPrecision) {
+ $sigfigs = $this->outputPrecision;
+ }
+
+ // BCMath's internal precision deals only with decimals. Use
+ // our default if the initial number has no decimals, or increase
+ // it by how ever many decimals, thus, the number of guard digits
+ // will always be greater than or equal to internalPrecision.
+ $log = (int)floor(log(abs($n), 10));
+ $cp = ($log < 0) ? $this->internalPrecision - $log : $this->internalPrecision; // internal precision
+
+ for ($i = 0; $i < 2; $i++) {
+
+ // Determine what unit IN THIS SYSTEM we need to convert to
+ if ($dest_state === $state) {
+ // Simple conversion
+ $dest_unit = $to_unit;
+ } else {
+ // Convert to the smallest unit, pending a system shift
+ $dest_unit = self::$units[$state][$dest_state][0];
+ }
+
+ // Do the conversion if necessary
+ if ($dest_unit !== $unit) {
+ $factor = $this->div(self::$units[$state][$unit], self::$units[$state][$dest_unit], $cp);
+ $n = $this->mul($n, $factor, $cp);
+ $unit = $dest_unit;
+ }
+
+ // Output was zero, so bail out early. Shouldn't ever happen.
+ if ($n === '') {
+ $n = '0';
+ $unit = $to_unit;
+ break;
+ }
+
+ // It was a simple conversion, so bail out
+ if ($dest_state === $state) {
+ break;
+ }
+
+ if ($i !== 0) {
+ // Conversion failed! Apparently, the system we forwarded
+ // to didn't have this unit. This should never happen!
+ return false;
+ }
+
+ // Pre-condition: $i == 0
+
+ // Perform conversion to next system of units
+ $n = $this->mul($n, self::$units[$state][$dest_state][1], $cp);
+ $unit = self::$units[$state][$dest_state][2];
+ $state = $dest_state;
+
+ // One more loop around to convert the unit in the new system.
+
+ }
+
+ // Post-condition: $unit == $to_unit
+ if ($unit !== $to_unit) {
+ return false;
+ }
+
+ // Useful for debugging:
+ //echo "<pre>n";
+ //echo "$n\nsigfigs = $sigfigs\nnew_log = $new_log\nlog = $log\nrp = $rp\n</pre>\n";
+
+ $n = $this->round($n, $sigfigs);
+ if (strpos($n, '.') !== false) {
+ $n = rtrim($n, '0');
+ }
+ $n = rtrim($n, '.');
+
+ return new HTMLPurifier_Length($n, $unit);
+ }
+
+ /**
+ * Returns the number of significant figures in a string number.
+ * @param string $n Decimal number
+ * @return int number of sigfigs
+ */
+ public function getSigFigs($n)
+ {
+ $n = ltrim($n, '0+-');
+ $dp = strpos($n, '.'); // decimal position
+ if ($dp === false) {
+ $sigfigs = strlen(rtrim($n, '0'));
+ } else {
+ $sigfigs = strlen(ltrim($n, '0.')); // eliminate extra decimal character
+ if ($dp !== 0) {
+ $sigfigs--;
+ }
+ }
+ return $sigfigs;
+ }
+
+ /**
+ * Adds two numbers, using arbitrary precision when available.
+ * @param string $s1
+ * @param string $s2
+ * @param int $scale
+ * @return string
+ */
+ private function add($s1, $s2, $scale)
+ {
+ if ($this->bcmath) {
+ return bcadd($s1, $s2, $scale);
+ } else {
+ return $this->scale((float)$s1 + (float)$s2, $scale);
+ }
+ }
+
+ /**
+ * Multiples two numbers, using arbitrary precision when available.
+ * @param string $s1
+ * @param string $s2
+ * @param int $scale
+ * @return string
+ */
+ private function mul($s1, $s2, $scale)
+ {
+ if ($this->bcmath) {
+ return bcmul($s1, $s2, $scale);
+ } else {
+ return $this->scale((float)$s1 * (float)$s2, $scale);
+ }
+ }
+
+ /**
+ * Divides two numbers, using arbitrary precision when available.
+ * @param string $s1
+ * @param string $s2
+ * @param int $scale
+ * @return string
+ */
+ private function div($s1, $s2, $scale)
+ {
+ if ($this->bcmath) {
+ return bcdiv($s1, $s2, $scale);
+ } else {
+ return $this->scale((float)$s1 / (float)$s2, $scale);
+ }
+ }
+
+ /**
+ * Rounds a number according to the number of sigfigs it should have,
+ * using arbitrary precision when available.
+ * @param float $n
+ * @param int $sigfigs
+ * @return string
+ */
+ private function round($n, $sigfigs)
+ {
+ $new_log = (int)floor(log(abs($n), 10)); // Number of digits left of decimal - 1
+ $rp = $sigfigs - $new_log - 1; // Number of decimal places needed
+ $neg = $n < 0 ? '-' : ''; // Negative sign
+ if ($this->bcmath) {
+ if ($rp >= 0) {
+ $n = bcadd($n, $neg . '0.' . str_repeat('0', $rp) . '5', $rp + 1);
+ $n = bcdiv($n, '1', $rp);
+ } else {
+ // This algorithm partially depends on the standardized
+ // form of numbers that comes out of bcmath.
+ $n = bcadd($n, $neg . '5' . str_repeat('0', $new_log - $sigfigs), 0);
+ $n = substr($n, 0, $sigfigs + strlen($neg)) . str_repeat('0', $new_log - $sigfigs + 1);
+ }
+ return $n;
+ } else {
+ return $this->scale(round($n, $sigfigs - $new_log - 1), $rp + 1);
+ }
+ }
+
+ /**
+ * Scales a float to $scale digits right of decimal point, like BCMath.
+ * @param float $r
+ * @param int $scale
+ * @return string
+ */
+ private function scale($r, $scale)
+ {
+ if ($scale < 0) {
+ // The f sprintf type doesn't support negative numbers, so we
+ // need to cludge things manually. First get the string.
+ $r = sprintf('%.0f', (float)$r);
+ // Due to floating point precision loss, $r will more than likely
+ // look something like 4652999999999.9234. We grab one more digit
+ // than we need to precise from $r and then use that to round
+ // appropriately.
+ $precise = (string)round(substr($r, 0, strlen($r) + $scale), -1);
+ // Now we return it, truncating the zero that was rounded off.
+ return substr($precise, 0, -1) . str_repeat('0', -$scale + 1);
+ }
+ return sprintf('%.' . $scale . 'f', (float)$r);
+ }
+}
+
+// vim: et sw=4 sts=4