Current File : //home/tradevaly/www/phpmy/libraries/classes/Charsets/Collation.php |
<?php
/**
* Value object class for a collation
*/
declare(strict_types=1);
namespace PhpMyAdmin\Charsets;
use function __;
use function _pgettext;
use function count;
use function explode;
use function implode;
/**
* Value object class for a collation
*/
final class Collation
{
/**
* The collation name
*
* @var string
*/
private $name;
/**
* A description of the collation
*
* @var string
*/
private $description;
/**
* The name of the character set with which the collation is associated
*
* @var string
*/
private $charset;
/**
* The collation ID
*
* @var int
*/
private $id;
/**
* Whether the collation is the default for its character set
*
* @var bool
*/
private $isDefault;
/**
* Whether the character set is compiled into the server
*
* @var bool
*/
private $isCompiled;
/**
* Used for determining the memory used to sort strings in this collation
*
* @var int
*/
private $sortLength;
/**
* The collation pad attribute
*
* @var string
*/
private $padAttribute;
/**
* @param string $name Collation name
* @param string $charset Related charset
* @param int $id Collation ID
* @param bool $isDefault Whether is the default
* @param bool $isCompiled Whether the charset is compiled
* @param int $sortLength Sort length
* @param string $padAttribute Pad attribute
*/
private function __construct(
string $name,
string $charset,
int $id,
bool $isDefault,
bool $isCompiled,
int $sortLength,
string $padAttribute
) {
$this->name = $name;
$this->charset = $charset;
$this->id = $id;
$this->isDefault = $isDefault;
$this->isCompiled = $isCompiled;
$this->sortLength = $sortLength;
$this->padAttribute = $padAttribute;
$this->description = $this->buildDescription();
}
/**
* @param string[] $state State obtained from the database server
*/
public static function fromServer(array $state): self
{
return new self(
$state['Collation'] ?? '',
$state['Charset'] ?? '',
(int) ($state['Id'] ?? 0),
isset($state['Default']) && ($state['Default'] === 'Yes' || $state['Default'] === '1'),
isset($state['Compiled']) && ($state['Compiled'] === 'Yes' || $state['Compiled'] === '1'),
(int) ($state['Sortlen'] ?? 0),
$state['Pad_attribute'] ?? ''
);
}
public function getName(): string
{
return $this->name;
}
public function getDescription(): string
{
return $this->description;
}
public function getCharset(): string
{
return $this->charset;
}
public function getId(): int
{
return $this->id;
}
public function isDefault(): bool
{
return $this->isDefault;
}
public function isCompiled(): bool
{
return $this->isCompiled;
}
public function getSortLength(): int
{
return $this->sortLength;
}
public function getPadAttribute(): string
{
return $this->padAttribute;
}
/**
* Returns description for given collation
*
* @return string collation description
*/
private function buildDescription(): string
{
$parts = explode('_', $this->getName());
$name = __('Unknown');
$variant = null;
$suffixes = [];
$unicode = false;
$unknown = false;
$level = 0;
foreach ($parts as $part) {
if ($level === 0) {
/* Next will be language */
$level = 1;
/* First should be charset */
[$name, $unicode, $unknown, $variant] = $this->getNameForLevel0($unicode, $unknown, $part, $variant);
continue;
}
if ($level === 1) {
/* Next will be variant unless changed later */
$level = 4;
/* Locale name or code */
$found = true;
[$name, $level, $found] = $this->getNameForLevel1($unicode, $unknown, $part, $name, $level, $found);
if ($found) {
continue;
}
// Not parsed token, fall to next level
}
if ($level === 2) {
/* Next will be variant */
$level = 4;
/* Germal variant */
if ($part === 'pb') {
$name = _pgettext('Collation', 'German (phone book order)');
continue;
}
$name = _pgettext('Collation', 'German (dictionary order)');
// Not parsed token, fall to next level
}
if ($level === 3) {
/* Next will be variant */
$level = 4;
/* Spanish variant */
if ($part === 'trad') {
$name = _pgettext('Collation', 'Spanish (traditional)');
continue;
}
$name = _pgettext('Collation', 'Spanish (modern)');
// Not parsed token, fall to next level
}
if ($level === 4) {
/* Next will be suffix */
$level = 5;
/* Variant */
$found = true;
$variantFound = $this->getVariant($part);
if ($variantFound === null) {
$found = false;
} else {
$variant = $variantFound;
}
if ($found) {
continue;
}
// Not parsed token, fall to next level
}
if ($level < 5) {
continue;
}
/* Suffixes */
$suffixes = $this->addSuffixes($suffixes, $part);
}
return $this->buildName($name, $variant, $suffixes);
}
/**
* @param string[] $suffixes
*/
private function buildName(string $result, ?string $variant, array $suffixes): string
{
if ($variant !== null) {
$result .= ' (' . $variant . ')';
}
if (count($suffixes) > 0) {
$result .= ', ' . implode(', ', $suffixes);
}
return $result;
}
private function getVariant(string $part): ?string
{
switch ($part) {
case '0900':
return 'UCA 9.0.0';
case '520':
return 'UCA 5.2.0';
case 'mysql561':
return 'MySQL 5.6.1';
case 'mysql500':
return 'MySQL 5.0.0';
default:
return null;
}
}
/**
* @param string[] $suffixes
*
* @return string[]
*/
private function addSuffixes(array $suffixes, string $part): array
{
switch ($part) {
case 'ci':
$suffixes[] = _pgettext('Collation variant', 'case-insensitive');
break;
case 'cs':
$suffixes[] = _pgettext('Collation variant', 'case-sensitive');
break;
case 'ai':
$suffixes[] = _pgettext('Collation variant', 'accent-insensitive');
break;
case 'as':
$suffixes[] = _pgettext('Collation variant', 'accent-sensitive');
break;
case 'ks':
$suffixes[] = _pgettext('Collation variant', 'kana-sensitive');
break;
case 'w2':
case 'l2':
$suffixes[] = _pgettext('Collation variant', 'multi-level');
break;
case 'bin':
$suffixes[] = _pgettext('Collation variant', 'binary');
break;
case 'nopad':
$suffixes[] = _pgettext('Collation variant', 'no-pad');
break;
}
return $suffixes;
}
/**
* @return array<int, bool|string|null>
* @psalm-return array{string, bool, bool, string|null}
*/
private function getNameForLevel0(
bool $unicode,
bool $unknown,
string $part,
?string $variant
): array {
switch ($part) {
case 'binary':
$name = _pgettext('Collation', 'Binary');
break;
// Unicode charsets
case 'utf8mb4':
$variant = 'UCA 4.0.0';
// Fall through to other unicode
case 'ucs2':
case 'utf8':
case 'utf16':
case 'utf16le':
case 'utf16be':
case 'utf32':
$name = _pgettext('Collation', 'Unicode');
$unicode = true;
break;
// West European charsets
case 'ascii':
case 'cp850':
case 'dec8':
case 'hp8':
case 'latin1':
case 'macroman':
$name = _pgettext('Collation', 'West European');
break;
// Central European charsets
case 'cp1250':
case 'cp852':
case 'latin2':
case 'macce':
$name = _pgettext('Collation', 'Central European');
break;
// Russian charsets
case 'cp866':
case 'koi8r':
$name = _pgettext('Collation', 'Russian');
break;
// Chinese charsets
case 'gb2312':
case 'gbk':
$name = _pgettext('Collation', 'Simplified Chinese');
break;
case 'big5':
$name = _pgettext('Collation', 'Traditional Chinese');
break;
case 'gb18030':
$name = _pgettext('Collation', 'Chinese');
$unicode = true;
break;
// Japanese charsets
case 'sjis':
case 'ujis':
case 'cp932':
case 'eucjpms':
$name = _pgettext('Collation', 'Japanese');
break;
// Baltic charsets
case 'cp1257':
case 'latin7':
$name = _pgettext('Collation', 'Baltic');
break;
// Other
case 'armscii8':
case 'armscii':
$name = _pgettext('Collation', 'Armenian');
break;
case 'cp1251':
$name = _pgettext('Collation', 'Cyrillic');
break;
case 'cp1256':
$name = _pgettext('Collation', 'Arabic');
break;
case 'euckr':
$name = _pgettext('Collation', 'Korean');
break;
case 'hebrew':
$name = _pgettext('Collation', 'Hebrew');
break;
case 'geostd8':
$name = _pgettext('Collation', 'Georgian');
break;
case 'greek':
$name = _pgettext('Collation', 'Greek');
break;
case 'keybcs2':
$name = _pgettext('Collation', 'Czech-Slovak');
break;
case 'koi8u':
$name = _pgettext('Collation', 'Ukrainian');
break;
case 'latin5':
$name = _pgettext('Collation', 'Turkish');
break;
case 'swe7':
$name = _pgettext('Collation', 'Swedish');
break;
case 'tis620':
$name = _pgettext('Collation', 'Thai');
break;
default:
$name = _pgettext('Collation', 'Unknown');
$unknown = true;
break;
}
return [$name, $unicode, $unknown, $variant];
}
/**
* @return array<int, bool|int|string>
* @psalm-return array{string, int, bool}
*/
private function getNameForLevel1(
bool $unicode,
bool $unknown,
string $part,
string $name,
int $level,
bool $found
): array {
switch ($part) {
case 'general':
break;
case 'bulgarian':
case 'bg':
$name = _pgettext('Collation', 'Bulgarian');
break;
case 'chinese':
case 'cn':
case 'zh':
if ($unicode) {
$name = _pgettext('Collation', 'Chinese');
}
break;
case 'croatian':
case 'hr':
$name = _pgettext('Collation', 'Croatian');
break;
case 'czech':
case 'cs':
$name = _pgettext('Collation', 'Czech');
break;
case 'danish':
case 'da':
$name = _pgettext('Collation', 'Danish');
break;
case 'english':
case 'en':
$name = _pgettext('Collation', 'English');
break;
case 'esperanto':
case 'eo':
$name = _pgettext('Collation', 'Esperanto');
break;
case 'estonian':
case 'et':
$name = _pgettext('Collation', 'Estonian');
break;
case 'german1':
$name = _pgettext('Collation', 'German (dictionary order)');
break;
case 'german2':
$name = _pgettext('Collation', 'German (phone book order)');
break;
case 'german':
case 'de':
/* Name is set later */
$level = 2;
break;
case 'hungarian':
case 'hu':
$name = _pgettext('Collation', 'Hungarian');
break;
case 'icelandic':
case 'is':
$name = _pgettext('Collation', 'Icelandic');
break;
case 'japanese':
case 'ja':
$name = _pgettext('Collation', 'Japanese');
break;
case 'la':
$name = _pgettext('Collation', 'Classical Latin');
break;
case 'latvian':
case 'lv':
$name = _pgettext('Collation', 'Latvian');
break;
case 'lithuanian':
case 'lt':
$name = _pgettext('Collation', 'Lithuanian');
break;
case 'korean':
case 'ko':
$name = _pgettext('Collation', 'Korean');
break;
case 'myanmar':
case 'my':
$name = _pgettext('Collation', 'Burmese');
break;
case 'persian':
$name = _pgettext('Collation', 'Persian');
break;
case 'polish':
case 'pl':
$name = _pgettext('Collation', 'Polish');
break;
case 'roman':
$name = _pgettext('Collation', 'West European');
break;
case 'romanian':
case 'ro':
$name = _pgettext('Collation', 'Romanian');
break;
case 'ru':
$name = _pgettext('Collation', 'Russian');
break;
case 'si':
case 'sinhala':
$name = _pgettext('Collation', 'Sinhalese');
break;
case 'slovak':
case 'sk':
$name = _pgettext('Collation', 'Slovak');
break;
case 'slovenian':
case 'sl':
$name = _pgettext('Collation', 'Slovenian');
break;
case 'spanish':
$name = _pgettext('Collation', 'Spanish (modern)');
break;
case 'es':
/* Name is set later */
$level = 3;
break;
case 'spanish2':
$name = _pgettext('Collation', 'Spanish (traditional)');
break;
case 'swedish':
case 'sv':
$name = _pgettext('Collation', 'Swedish');
break;
case 'thai':
case 'th':
$name = _pgettext('Collation', 'Thai');
break;
case 'turkish':
case 'tr':
$name = _pgettext('Collation', 'Turkish');
break;
case 'ukrainian':
case 'uk':
$name = _pgettext('Collation', 'Ukrainian');
break;
case 'vietnamese':
case 'vi':
$name = _pgettext('Collation', 'Vietnamese');
break;
case 'unicode':
if ($unknown) {
$name = _pgettext('Collation', 'Unicode');
}
break;
default:
$found = false;
}
return [$name, $level, $found];
}
}