Current File : //home/tradevaly/public_html/phpmy/libraries/classes/Linter.php |
<?php
/**
* Analyzes a query and gives user feedback.
*/
declare(strict_types=1);
namespace PhpMyAdmin;
use PhpMyAdmin\SqlParser\Lexer;
use PhpMyAdmin\SqlParser\Parser;
use PhpMyAdmin\SqlParser\UtfString;
use PhpMyAdmin\SqlParser\Utils\Error as ParserError;
use function __;
use function defined;
use function htmlspecialchars;
use function mb_strlen;
use function sprintf;
use function strlen;
/**
* The linter itself.
*/
class Linter
{
/**
* Gets the starting position of each line.
*
* @param string|UtfString $str String to be analyzed.
*
* @return array
*/
public static function getLines($str)
{
if ((! ($str instanceof UtfString)) && defined('USE_UTF_STRINGS') && USE_UTF_STRINGS) {
// If the lexer uses UtfString for processing then the position will
// represent the position of the character and not the position of
// the byte.
$str = new UtfString($str);
}
// The reason for using the strlen is that the length
// required is the length in bytes, not characters.
//
// Given the following string: `????+`, where `?` represents a
// multi-byte character (lets assume that every `?` is a 2-byte
// character) and `+` is a newline, the first value of `$i` is `0`
// and the last one is `4` (because there are 5 characters). Bytes
// `$str[0]` and `$str[1]` are the first character, `$str[2]` and
// `$str[3]` are the second one and `$str[4]` is going to be the
// first byte of the third character. The fourth and the last one
// (which is actually a new line) aren't going to be processed at
// all.
$len = $str instanceof UtfString ?
$str->length() : strlen($str);
$lines = [0];
for ($i = 0; $i < $len; ++$i) {
if ($str[$i] !== "\n") {
continue;
}
$lines[] = $i + 1;
}
return $lines;
}
/**
* Computes the number of the line and column given an absolute position.
*
* @param array $lines The starting position of each line.
* @param int $pos The absolute position
*
* @return array
*/
public static function findLineNumberAndColumn(array $lines, $pos)
{
$line = 0;
foreach ($lines as $lineNo => $lineStart) {
if ($lineStart > $pos) {
break;
}
$line = $lineNo;
}
return [
$line,
$pos - $lines[$line],
];
}
/**
* Runs the linting process.
*
* @param string $query The query to be checked.
*
* @return array
*/
public static function lint($query)
{
// Disabling lint for huge queries to save some resources.
if (mb_strlen($query) > 10000) {
return [
[
'message' => __('Linting is disabled for this query because it exceeds the maximum length.'),
'fromLine' => 0,
'fromColumn' => 0,
'toLine' => 0,
'toColumn' => 0,
'severity' => 'warning',
],
];
}
/**
* Lexer used for tokenizing the query.
*/
$lexer = new Lexer($query);
/**
* Parsed used for analysing the query.
*/
$parser = new Parser($lexer->list);
/**
* Array containing all errors.
*/
$errors = ParserError::get([$lexer, $parser]);
/**
* The response containing of all errors.
*
* @var array
*/
$response = [];
/**
* The starting position for each line.
*
* CodeMirror requires relative position to line, but the parser stores
* only the absolute position of the character in string.
*/
$lines = static::getLines($query);
// Building the response.
foreach ($errors as $error) {
// Starting position of the string that caused the error.
[$fromLine, $fromColumn] = static::findLineNumberAndColumn($lines, $error[3]);
// Ending position of the string that caused the error.
[$toLine, $toColumn] = static::findLineNumberAndColumn(
$lines,
$error[3] + mb_strlen((string) $error[2])
);
// Building the response.
$response[] = [
'message' => sprintf(
__('%1$s (near <code>%2$s</code>)'),
htmlspecialchars((string) $error[0]),
htmlspecialchars((string) $error[2])
),
'fromLine' => $fromLine,
'fromColumn' => $fromColumn,
'toLine' => $toLine,
'toColumn' => $toColumn,
'severity' => 'error',
];
}
// Sending back the answer.
return $response;
}
}