257 lines
11 KiB
PHP
257 lines
11 KiB
PHP
<?php
|
|
|
|
function wpautop( $pee, $br = true ) {
|
|
$pre_tags = array();
|
|
|
|
if ( trim( $pee ) === '' ) {
|
|
return '';
|
|
}
|
|
|
|
// Just to make things a little easier, pad the end.
|
|
$pee = $pee . "\n";
|
|
|
|
/*
|
|
* Pre tags shouldn't be touched by autop.
|
|
* Replace pre tags with placeholders and bring them back after autop.
|
|
*/
|
|
if ( strpos( $pee, '<pre' ) !== false ) {
|
|
$pee_parts = explode( '</pre>', $pee );
|
|
$last_pee = array_pop( $pee_parts );
|
|
$pee = '';
|
|
$i = 0;
|
|
|
|
foreach ( $pee_parts as $pee_part ) {
|
|
$start = strpos( $pee_part, '<pre' );
|
|
|
|
// Malformed html?
|
|
if ( $start === false ) {
|
|
$pee .= $pee_part;
|
|
continue;
|
|
}
|
|
|
|
$name = "<pre wp-pre-tag-$i></pre>";
|
|
$pre_tags[ $name ] = substr( $pee_part, $start ) . '</pre>';
|
|
|
|
$pee .= substr( $pee_part, 0, $start ) . $name;
|
|
$i++;
|
|
}
|
|
|
|
$pee .= $last_pee;
|
|
}
|
|
// Change multiple <br>s into two line breaks, which will turn into paragraphs.
|
|
$pee = preg_replace( '|<br\s*/?>\s*<br\s*/?>|', "\n\n", $pee );
|
|
|
|
$allblocks = '(?:table|thead|tfoot|caption|col|colgroup|tbody|tr|td|th|div|dl|dd|dt|ul|ol|li|pre|form|map|area|blockquote|address|math|style|p|h[1-6]|hr|fieldset|legend|section|article|aside|hgroup|header|footer|nav|figure|figcaption|details|menu|summary)';
|
|
|
|
// Add a double line break above block-level opening tags.
|
|
$pee = preg_replace( '!(<' . $allblocks . '[\s/>])!', "\n\n$1", $pee );
|
|
|
|
// Add a double line break below block-level closing tags.
|
|
$pee = preg_replace( '!(</' . $allblocks . '>)!', "$1\n\n", $pee );
|
|
|
|
// Add a double line break after hr tags, which are self closing.
|
|
$pee = preg_replace( '!(<hr\s*?/?>)!', "$1\n\n", $pee );
|
|
|
|
// Standardize newline characters to "\n".
|
|
$pee = str_replace( array( "\r\n", "\r" ), "\n", $pee );
|
|
|
|
// Find newlines in all elements and add placeholders.
|
|
$pee = wp_replace_in_html_tags( $pee, array( "\n" => ' <!-- wpnl --> ' ) );
|
|
|
|
// Collapse line breaks before and after <option> elements so they don't get autop'd.
|
|
if ( strpos( $pee, '<option' ) !== false ) {
|
|
$pee = preg_replace( '|\s*<option|', '<option', $pee );
|
|
$pee = preg_replace( '|</option>\s*|', '</option>', $pee );
|
|
}
|
|
|
|
/*
|
|
* Collapse line breaks inside <object> elements, before <param> and <embed> elements
|
|
* so they don't get autop'd.
|
|
*/
|
|
if ( strpos( $pee, '</object>' ) !== false ) {
|
|
$pee = preg_replace( '|(<object[^>]*>)\s*|', '$1', $pee );
|
|
$pee = preg_replace( '|\s*</object>|', '</object>', $pee );
|
|
$pee = preg_replace( '%\s*(</?(?:param|embed)[^>]*>)\s*%', '$1', $pee );
|
|
}
|
|
|
|
/*
|
|
* Collapse line breaks inside <audio> and <video> elements,
|
|
* before and after <source> and <track> elements.
|
|
*/
|
|
if ( strpos( $pee, '<source' ) !== false || strpos( $pee, '<track' ) !== false ) {
|
|
$pee = preg_replace( '%([<\[](?:audio|video)[^>\]]*[>\]])\s*%', '$1', $pee );
|
|
$pee = preg_replace( '%\s*([<\[]/(?:audio|video)[>\]])%', '$1', $pee );
|
|
$pee = preg_replace( '%\s*(<(?:source|track)[^>]*>)\s*%', '$1', $pee );
|
|
}
|
|
|
|
// Collapse line breaks before and after <figcaption> elements.
|
|
if ( strpos( $pee, '<figcaption' ) !== false ) {
|
|
$pee = preg_replace( '|\s*(<figcaption[^>]*>)|', '$1', $pee );
|
|
$pee = preg_replace( '|</figcaption>\s*|', '</figcaption>', $pee );
|
|
}
|
|
|
|
// Remove more than two contiguous line breaks.
|
|
$pee = preg_replace( "/\n\n+/", "\n\n", $pee );
|
|
|
|
// Split up the contents into an array of strings, separated by double line breaks.
|
|
$pees = preg_split( '/\n\s*\n/', $pee, -1, PREG_SPLIT_NO_EMPTY );
|
|
|
|
// Reset $pee prior to rebuilding.
|
|
$pee = '';
|
|
|
|
// Rebuild the content as a string, wrapping every bit with a <p>.
|
|
foreach ( $pees as $tinkle ) {
|
|
$pee .= '<p>' . trim( $tinkle, "\n" ) . "</p>\n";
|
|
}
|
|
|
|
// Under certain strange conditions it could create a P of entirely whitespace.
|
|
$pee = preg_replace( '|<p>\s*</p>|', '', $pee );
|
|
|
|
// Add a closing <p> inside <div>, <address>, or <form> tag if missing.
|
|
$pee = preg_replace( '!<p>([^<]+)</(div|address|form)>!', '<p>$1</p></$2>', $pee );
|
|
|
|
// If an opening or closing block element tag is wrapped in a <p>, unwrap it.
|
|
$pee = preg_replace( '!<p>\s*(</?' . $allblocks . '[^>]*>)\s*</p>!', '$1', $pee );
|
|
|
|
// In some cases <li> may get wrapped in <p>, fix them.
|
|
$pee = preg_replace( '|<p>(<li.+?)</p>|', '$1', $pee );
|
|
|
|
// If a <blockquote> is wrapped with a <p>, move it inside the <blockquote>.
|
|
$pee = preg_replace( '|<p><blockquote([^>]*)>|i', '<blockquote$1><p>', $pee );
|
|
$pee = str_replace( '</blockquote></p>', '</p></blockquote>', $pee );
|
|
|
|
// If an opening or closing block element tag is preceded by an opening <p> tag, remove it.
|
|
$pee = preg_replace( '!<p>\s*(</?' . $allblocks . '[^>]*>)!', '$1', $pee );
|
|
|
|
// If an opening or closing block element tag is followed by a closing <p> tag, remove it.
|
|
$pee = preg_replace( '!(</?' . $allblocks . '[^>]*>)\s*</p>!', '$1', $pee );
|
|
|
|
// Optionally insert line breaks.
|
|
if ( $br ) {
|
|
// Replace newlines that shouldn't be touched with a placeholder.
|
|
$pee = preg_replace_callback( '/<(script|style|svg).*?<\/\\1>/s', '_autop_newline_preservation_helper', $pee );
|
|
|
|
// Normalize <br>
|
|
$pee = str_replace( array( '<br>', '<br/>' ), '<br />', $pee );
|
|
|
|
// Replace any new line characters that aren't preceded by a <br /> with a <br />.
|
|
$pee = preg_replace( '|(?<!<br />)\s*\n|', "<br />\n", $pee );
|
|
|
|
// Replace newline placeholders with newlines.
|
|
$pee = str_replace( '<WPPreserveNewline />', "\n", $pee );
|
|
}
|
|
|
|
// If a <br /> tag is after an opening or closing block tag, remove it.
|
|
$pee = preg_replace( '!(</?' . $allblocks . '[^>]*>)\s*<br />!', '$1', $pee );
|
|
|
|
// If a <br /> tag is before a subset of opening or closing block tags, remove it.
|
|
$pee = preg_replace( '!<br />(\s*</?(?:p|li|div|dl|dd|dt|th|pre|td|ul|ol)[^>]*>)!', '$1', $pee );
|
|
$pee = preg_replace( "|\n</p>$|", '</p>', $pee );
|
|
|
|
// Replace placeholder <pre> tags with their original content.
|
|
if ( ! empty( $pre_tags ) ) {
|
|
$pee = str_replace( array_keys( $pre_tags ), array_values( $pre_tags ), $pee );
|
|
}
|
|
|
|
// Restore newlines in all elements.
|
|
if ( false !== strpos( $pee, '<!-- wpnl -->' ) ) {
|
|
$pee = str_replace( array( ' <!-- wpnl --> ', '<!-- wpnl -->' ), "\n", $pee );
|
|
}
|
|
|
|
return $pee;
|
|
}
|
|
|
|
function wp_replace_in_html_tags( $haystack, $replace_pairs ) {
|
|
// Find all elements.
|
|
$textarr = wp_html_split( $haystack );
|
|
$changed = false;
|
|
|
|
// Optimize when searching for one item.
|
|
if ( 1 === count( $replace_pairs ) ) {
|
|
// Extract $needle and $replace.
|
|
foreach ( $replace_pairs as $needle => $replace ) {
|
|
}
|
|
|
|
// Loop through delimiters (elements) only.
|
|
for ( $i = 1, $c = count( $textarr ); $i < $c; $i += 2 ) {
|
|
if ( false !== strpos( $textarr[ $i ], $needle ) ) {
|
|
$textarr[ $i ] = str_replace( $needle, $replace, $textarr[ $i ] );
|
|
$changed = true;
|
|
}
|
|
}
|
|
} else {
|
|
// Extract all $needles.
|
|
$needles = array_keys( $replace_pairs );
|
|
|
|
// Loop through delimiters (elements) only.
|
|
for ( $i = 1, $c = count( $textarr ); $i < $c; $i += 2 ) {
|
|
foreach ( $needles as $needle ) {
|
|
if ( false !== strpos( $textarr[ $i ], $needle ) ) {
|
|
$textarr[ $i ] = strtr( $textarr[ $i ], $replace_pairs );
|
|
$changed = true;
|
|
// After one strtr() break out of the foreach loop and look at next element.
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if ( $changed ) {
|
|
$haystack = implode( $textarr );
|
|
}
|
|
|
|
return $haystack;
|
|
}
|
|
|
|
function wp_html_split( $input ) {
|
|
return preg_split( get_html_split_regex(), $input, -1, PREG_SPLIT_DELIM_CAPTURE );
|
|
}
|
|
|
|
function get_html_split_regex() {
|
|
static $regex;
|
|
|
|
if ( ! isset( $regex ) ) {
|
|
// phpcs:disable Squiz.Strings.ConcatenationSpacing.PaddingFound -- don't remove regex indentation
|
|
$comments =
|
|
'!' // Start of comment, after the <.
|
|
. '(?:' // Unroll the loop: Consume everything until --> is found.
|
|
. '-(?!->)' // Dash not followed by end of comment.
|
|
. '[^\-]*+' // Consume non-dashes.
|
|
. ')*+' // Loop possessively.
|
|
. '(?:-->)?'; // End of comment. If not found, match all input.
|
|
|
|
$cdata =
|
|
'!\[CDATA\[' // Start of comment, after the <.
|
|
. '[^\]]*+' // Consume non-].
|
|
. '(?:' // Unroll the loop: Consume everything until ]]> is found.
|
|
. '](?!]>)' // One ] not followed by end of comment.
|
|
. '[^\]]*+' // Consume non-].
|
|
. ')*+' // Loop possessively.
|
|
. '(?:]]>)?'; // End of comment. If not found, match all input.
|
|
|
|
$escaped =
|
|
'(?=' // Is the element escaped?
|
|
. '!--'
|
|
. '|'
|
|
. '!\[CDATA\['
|
|
. ')'
|
|
. '(?(?=!-)' // If yes, which type?
|
|
. $comments
|
|
. '|'
|
|
. $cdata
|
|
. ')';
|
|
|
|
$regex =
|
|
'/(' // Capture the entire match.
|
|
. '<' // Find start of element.
|
|
. '(?' // Conditional expression follows.
|
|
. $escaped // Find end of escaped element.
|
|
. '|' // ... else ...
|
|
. '[^>]*>?' // Find end of normal element.
|
|
. ')'
|
|
. ')/';
|
|
// phpcs:enable
|
|
}
|
|
|
|
return $regex;
|
|
} |