<?php
/**
* Checks that end of line characters are correct.
*
* @author Greg Sherwood <gsherwood@squiz.net>
* @copyright 2006-2015 Squiz Pty Ltd (ABN 77 084 670 600)
* @license https://github.com/squizlabs/PHP_CodeSniffer/blob/master/licence.txt BSD Licence
*/
namespace PHP_CodeSniffer\Standards\Generic\Sniffs\Files;
use PHP_CodeSniffer\Files\File;
use PHP_CodeSniffer\Sniffs\Sniff;
class LineEndingsSniff implements Sniff
{
/**
* A list of tokenizers this sniff supports.
*
* @var array
*/
public $supportedTokenizers = [
'PHP',
'JS',
'CSS',
];
/**
* The valid EOL character.
*
* @var string
*/
public $eolChar = '\n';
/**
* Returns an array of tokens this test wants to listen for.
*
* @return array
*/
public function register()
{
return [T_OPEN_TAG];
}//end register()
/**
* Processes this sniff, when one of its tokens is encountered.
*
* @param \PHP_CodeSniffer\Files\File $phpcsFile The file being scanned.
* @param int $stackPtr The position of the current token in
* the stack passed in $tokens.
*
* @return int
*/
public function process(File $phpcsFile, $stackPtr)
{
$found = $phpcsFile->eolChar;
$found = str_replace("\n", '\n', $found);
$found = str_replace("\r", '\r', $found);
$phpcsFile->recordMetric($stackPtr, 'EOL char', $found);
if ($found === $this->eolChar) {
// Ignore the rest of the file.
return ($phpcsFile->numTokens + 1);
}
// Check for single line files without an EOL. This is a very special
// case and the EOL char is set to \n when this happens.
if ($found === '\n') {
$tokens = $phpcsFile->getTokens();
$lastToken = ($phpcsFile->numTokens - 1);
if ($tokens[$lastToken]['line'] === 1
&& $tokens[$lastToken]['content'] !== "\n"
) {
return;
}
}
$error = 'End of line character is invalid; expected "%s" but found "%s"';
$expected = $this->eolChar;
$expected = str_replace("\n", '\n', $expected);
$expected = str_replace("\r", '\r', $expected);
$data = [
$expected,
$found,
];
// Errors are always reported on line 1, no matter where the first PHP tag is.
$fix = $phpcsFile->addFixableError($error, 0, 'InvalidEOLChar', $data);
if ($fix === true) {
$tokens = $phpcsFile->getTokens();
switch ($this->eolChar) {
case '\n':
$eolChar = "\n";
break;
case '\r':
$eolChar = "\r";
break;
case '\r\n':
$eolChar = "\r\n";
break;
default:
$eolChar = $this->eolChar;
break;
}
for ($i = 0; $i < $phpcsFile->numTokens; $i++) {
if (isset($tokens[($i + 1)]) === true
&& $tokens[($i + 1)]['line'] <= $tokens[$i]['line']
) {
continue;
}
// Token is the last on a line.
if (isset($tokens[$i]['orig_content']) === true) {
$tokenContent = $tokens[$i]['orig_content'];
} else {
$tokenContent = $tokens[$i]['content'];
}
if ($tokenContent === '') {
// Special case for JS/CSS close tag.
continue;
}
$newContent = rtrim($tokenContent, "\r\n");
$newContent .= $eolChar;
if ($tokenContent !== $newContent) {
$phpcsFile->fixer->replaceToken($i, $newContent);
}
}//end for
}//end if
// Ignore the rest of the file.
return ($phpcsFile->numTokens + 1);
}//end process()
}//end class