PluginProbe
Search Atlas SEO – OTTO AI SEO Automation for WordPress / 2.7.2
Search Atlas SEO – OTTO AI SEO Automation for WordPress v2.7.2
2.7.1 2.7.2 2.7.0 2.6.26 2.6.25 2.6.24 2.6.23 2.6.22 2.6.21 2.6.20 2.6.19 2.6.18 2.6.17 2.6.16 2.6.15 2.6.14 2.6.13 2.6.12 2.6.11 2.6.10 2.6.9 2.6.8 2.6.7 2.6.6 2.6.5 All 141 releases
← All changes | otto/Otto_html_class.php +1848 -211 2.6.13 → 2.7.2 View file →
@@ -14,8 +14,13 @@
14 14 */
15 15
16 16 Class Metasync_otto_html{
17 17
18 + # Marker announcing that OTTO processed this <head>. Deliberately a comment
19 + # rather than an attribute on the tag — see mark_head_optimized().
20 + const HEAD_OPTIMIZED_TOKEN = 'metasync_optimized';
21 + const HEAD_OPTIMIZED_MARKER = '<!--metasync_optimized-->';
22 +
18 23 # html dom
19 24 private $dom;
20 25
21 26 # the file path to save to
@@ -62,9 +67,9 @@
62 67 return $html;
63 68 }
64 69
65 70 /**
66 - * WP-355: Fix malformed self-closing non-void HTML tags before DOM parsing.
71 + * Fix malformed self-closing non-void HTML tags before DOM parsing.
67 72 *
68 73 * Some themes (e.g. Bootstrap Component Blox) output malformed tags like:
69 74 * <ul/ class="sub-menu dropdown-menu" />
70 75 * SimpleHtmlDom interprets <ul/ as a void element, strips all attributes,
@@ -126,10 +131,197 @@
126 131 return str_ireplace(array_keys($map), array_values($map), $html);
127 132 }
128 133
129 134 /**
130 - * WP-355 / WP-315: Capture ALL <style> blocks before DOM processing.
135 + * Quote unquoted attribute values before DOM parsing.
131 136 *
137 + * Per the HTML spec an unquoted attribute value ends at whitespace, so
138 + * browsers read this correctly:
139 + *
140 + * data-autoplay-viewport=80%
141 + * data-mousewheel="false"
142 + * data-type="carousel"
143 + *
144 + * SimpleHtmlDom does not stop at the whitespace. It runs to the next quote
145 + * character, so the value becomes
146 + * `80%\n data-mousewheel="false"\n data-type="carousel"...` and EVERY
147 + * following attribute on the tag is swallowed into it and lost.
148 + *
149 + * Supreme Modules' image carousel emits exactly that markup. The attributes
150 + * lost after it include data-type, data-slider-orientation and
151 + * data-slideshow-to-show, so Swiper initialises on defaults, stacks all
152 + * slides and grows the page by one slide-set per autoplay tick until it is
153 + * tens of thousands of pixels tall.
154 + *
155 + * Normalising to `data-autoplay-viewport="80%"` before parsing makes the
156 + * markup unambiguous, so no attribute is lost. Already-quoted values are
157 + * left untouched, and <script>/<style> bodies are shielded so their
158 + * contents are never mistaken for tags.
159 + *
160 + * @param string $html Raw HTML before DOM parsing.
161 + * @return string HTML with every unquoted attribute value quoted.
162 + */
163 + private function quote_unquoted_attribute_values($html) {
164 + # Shield <script>/<style> bodies — their contents can contain '<' and '='
165 + # and must never be treated as markup.
166 + $protected = array();
167 + $html = preg_replace_callback(
168 + '/<(script|style)\b([^>]*)>(.*?)<\/\1>/is',
169 + function ($m) use (&$protected) {
170 + $token = '<!--METASYNC_ATTRQ_' . count($protected) . '-->';
171 + $protected[$token] = $m[0];
172 + return $token;
173 + },
174 + $html
175 + );
176 +
177 + # Walk real tags only. The alternation lets a quoted value contain '>'
178 + # so the tag boundary is found correctly. Possessive quantifiers (*+ / ++)
179 + # are deliberate: the three branches are mutually exclusive on their first
180 + # character, so backtracking can never produce a different match, and
181 + # forbidding it removes the catastrophic-backtracking risk this shape
182 + # otherwise carries on markup with many quotes and no '>'.
183 + $result = preg_replace_callback(
184 + '/<([a-zA-Z][a-zA-Z0-9:-]*+)((?:[^>"\']++|"[^"]*+"|\'[^\']*+\')*+)>/s',
185 + function ($m) {
186 + if (strpos($m[2], '=') === false) {
187 + return $m[0];
188 + }
189 +
190 + $attrs = $this->quote_unquoted_values_in_attr_region($m[2]);
191 +
192 + # null means nothing needed changing — return the ORIGINAL tag so
193 + # a document of well-formed HTML comes back byte-identical rather
194 + # than merely equivalent.
195 + return $attrs === null ? $m[0] : '<' . $m[1] . $attrs . '>';
196 + },
197 + $html
198 + );
199 +
200 + if ($result === null) {
201 + $result = $html;
202 + }
203 +
204 + if (!empty($protected)) {
205 + $result = str_replace(array_keys($protected), array_values($protected), $result);
206 + }
207 +
208 + return $result;
209 + }
210 +
211 + /**
212 + * Quote the unquoted values inside one tag's attribute region.
213 + *
214 + * Must be a tokeniser, not a regex. A flat pattern over the region cannot
215 + * tell an attribute from ordinary text inside an already-quoted value, and
216 + * would wreck common markup — `content="width=device-width,
217 + * initial-scale=1"` looks like it contains an `initial-scale=1` attribute.
218 + * Quoted values are therefore copied through verbatim and never inspected.
219 + *
220 + * @param string $attrs Attribute region: everything between the tag name and '>'.
221 + * @return string|null Rewritten region, or null when nothing needed changing.
222 + */
223 + private function quote_unquoted_values_in_attr_region($attrs) {
224 + $len = strlen($attrs);
225 + $out = '';
226 + $i = 0;
227 + $changed = false;
228 +
229 + while ($i < $len) {
230 + $c = $attrs[$i];
231 +
232 + # Whitespace between attributes — preserve exactly.
233 + if ($c === ' ' || $c === "\t" || $c === "\n" || $c === "\r" || $c === "\f" || $c === "\v") {
234 + $out .= $c;
235 + $i++;
236 + continue;
237 + }
238 +
239 + # Attribute name.
240 + if (preg_match('/\G[a-zA-Z_:][a-zA-Z0-9_:.\-]*+/', $attrs, $nm, 0, $i)) {
241 + $out .= $nm[0];
242 + $i += strlen($nm[0]);
243 +
244 + # Look past optional whitespace for '='. No '=' means a valueless
245 + # boolean attribute; leave it alone.
246 + $j = $i;
247 + while ($j < $len && strpos(" \t\n\r\f\v", $attrs[$j]) !== false) {
248 + $j++;
249 + }
250 + if ($j >= $len || $attrs[$j] !== '=') {
251 + continue;
252 + }
253 +
254 + $out .= substr($attrs, $i, $j - $i) . '=';
255 + $i = $j + 1;
256 +
257 + # Whitespace after '='.
258 + while ($i < $len && strpos(" \t\n\r\f\v", $attrs[$i]) !== false) {
259 + $out .= $attrs[$i];
260 + $i++;
261 + }
262 +
263 + if ($i >= $len) {
264 + continue;
265 + }
266 +
267 + # Quoted value — copy verbatim through the closing quote so its
268 + # contents are never treated as attributes.
269 + if ($attrs[$i] === '"' || $attrs[$i] === "'") {
270 + $quote = $attrs[$i];
271 + $end = strpos($attrs, $quote, $i + 1);
272 + if ($end === false) {
273 + $out .= substr($attrs, $i);
274 + $i = $len;
275 + } else {
276 + $out .= substr($attrs, $i, $end - $i + 1);
277 + $i = $end + 1;
278 + }
279 + continue;
280 + }
281 +
282 + # Unquoted value — runs to the next whitespace. ('>' cannot occur:
283 + # it would have ended the tag before this region.)
284 + $start = $i;
285 + while ($i < $len && strpos(" \t\n\r\f\v", $attrs[$i]) === false) {
286 + $i++;
287 + }
288 + $value = substr($attrs, $start, $i - $start);
289 +
290 + if ($value === '') {
291 + continue;
292 + }
293 +
294 + # Only rewrite when it actually prevents a loss. SimpleHtmlDom
295 + # mis-parses an unquoted value by running from it to the NEXT
296 + # QUOTE CHARACTER, swallowing everything between. With no quote
297 + # left in this region there is nothing to swallow and the value
298 + # already parses correctly — so leave that markup byte-identical
299 + # rather than churn it. (OceanWP's `style=color:#;` as the final
300 + # attribute is the common benign case.)
301 + $rest = substr($attrs, $i);
302 + if (strpos($rest, '"') === false && strpos($rest, "'") === false) {
303 + $out .= $value;
304 + continue;
305 + }
306 +
307 + $out .= '"' . str_replace('"', '&quot;', $value) . '"';
308 + $changed = true;
309 + continue;
310 + }
311 +
312 + # Anything else (stray punctuation, a '/' before '>', malformed input)
313 + # — copy through untouched.
314 + $out .= $c;
315 + $i++;
316 + }
317 +
318 + return $changed ? $out : null;
319 + }
320 +
321 + /**
322 + * / Capture ALL <style> blocks before DOM processing.
323 + *
132 324 * SimpleHtmlDom can lose or corrupt <style> blocks during parse/serialize,
133 325 * especially those without an id attribute (common in custom themes).
134 326 * This captures every <style> block so we can re-inject any that are lost.
135 327 *
@@ -151,9 +343,9 @@
151 343 return $blocks;
152 344 }
153 345
154 346 /**
155 - * WP-355 / WP-315: Re-inject any <style> blocks lost during DOM processing.
347 + * / Re-inject any <style> blocks lost during DOM processing.
156 348 *
157 349 * Compares the captured blocks against the final HTML and re-injects any
158 350 * that disappeared. Named blocks (with id) are checked by id; anonymous
159 351 * blocks are checked by exact content match.
@@ -185,10 +377,68 @@
185 377 return $html;
186 378 }
187 379
188 380 /**
189 - * WP-465: Protect entity-encoded `srcdoc` attribute values before DOM parsing.
381 + * Build the new inner HTML for a heading whose text OTTO is replacing,
382 + * preserving the author's inline formatting.
190 383 *
384 + * Headings are matched by their stripped text, so the recommended value is
385 + * plain text with no way to know how it maps onto the original markup.
386 + * Writing it back directly would discard whatever was inside — a coloured
387 + * <span>, <em>, <br> line breaks — and visibly restyle the page.
388 + *
389 + * Two cases are safe:
390 + * - the heading is already plain text -> replace it
391 + * - the heading is ONE nested inline wrapper -> keep the wrappers,
392 + * chain around a single run of text swap only the text
393 + *
394 + * Anything else is ambiguous: sibling elements, <br>-separated lines, or
395 + * text mixed with elements have no single place the new sentence belongs.
396 + * Those return null so the caller leaves the heading alone — an unapplied
397 + * suggestion is better than silently restyling the customer's content.
398 + *
399 + * @param string $inner_html Current inner HTML of the heading.
400 + * @param string $replacement_text New text, already escaped by the caller.
401 + * @return string|null New inner HTML, or null to skip this heading.
402 + */
403 + public static function build_heading_inner_html($inner_html, $replacement_text) {
404 + $inner = trim($inner_html);
405 +
406 + # Plain text — nothing to preserve.
407 + if (strpos($inner, '<') === false) {
408 + return $replacement_text;
409 + }
410 +
411 + # Inline elements whose wrapper carries the styling worth keeping.
412 + # Deliberately excludes <br>: it encodes line structure, not styling,
413 + # and its presence means the heading has more than one text run.
414 + $inline = 'span|em|strong|b|i|u|a|small|mark|abbr|cite|q|s|sub|sup|font';
415 +
416 + $open = '';
417 + $close = '';
418 +
419 + # Peel wrappers outermost-in. The greedy (.*) means a heading with
420 + # sibling elements (e.g. <em>A</em><br><em>B</em>) leaves stray tags in
421 + # $inner, which the ambiguity check below then rejects.
422 + while (preg_match('#^<(' . $inline . ')(\s[^>]*)?>(.*)</\1\s*>$#is', $inner, $m)) {
423 + # Group 2 (the attribute list) is optional but always captured as ''
424 + # when absent, because group 3 follows it.
425 + $open .= '<' . $m[1] . $m[2] . '>';
426 + $close = '</' . $m[1] . '>' . $close;
427 + $inner = trim($m[3]);
428 +
429 + if (strpos($inner, '<') === false) {
430 + return $open . $replacement_text . $close;
431 + }
432 + }
433 +
434 + # Still markup left after peeling — ambiguous, so don't touch it.
435 + return null;
436 + }
437 +
438 + /**
439 + * / Protect entity-encoded `srcdoc` attribute values before DOM parsing.
440 + *
191 441 * Lazy-loaded YouTube/video facades embed a full HTML document inside an
192 442 * iframe's `srcdoc` attribute, entity-encoded:
193 443 * <iframe srcdoc="&lt;style&gt;*{overflow:hidden}...&lt;/style&gt;&lt;a&gt;..."></iframe>
194 444 * SimpleHtmlDom decodes that value and re-emits the inner <style>/<a>/<img>
@@ -217,9 +467,9 @@
217 467 );
218 468 }
219 469
220 470 /**
221 - * WP-465: Restore srcdoc values tokenized by protect_srcdoc_attributes().
471 + * Restore srcdoc values tokenized by protect_srcdoc_attributes().
222 472 * Tokens are unique and never altered by SimpleHtmlDom, so a plain
223 473 * str_replace on the final output is safe.
224 474 *
225 475 * @param string $html Processed HTML.
@@ -233,9 +483,9 @@
233 483 return str_replace(array_keys($store), array_values($store), $html);
234 484 }
235 485
236 486 /**
237 - * WP-465: Safety net — strip any lazy-video FACADE <style> that leaked into the
487 + * Safety net — strip any lazy-video FACADE <style> that leaked into the
238 488 * document as a real stylesheet.
239 489 *
240 490 * Lazy YouTube/video facades carry this exact CSS reset inside their iframe
241 491 * srcdoc: *{...overflow:hidden} html,body{height:100%} img,span{position:absolute;width:100%;...}
@@ -250,19 +500,212 @@
250 500 *
251 501 * @param string $html
252 502 * @return string
253 503 */
504 + /**
505 + * Crash-free guard for full-document string transforms.
506 + *
507 + * PCRE operations (preg_replace / preg_replace_callback) return NULL when they
508 + * hit a limit — most notably "JIT stack limit exhausted" on Divi/page-builder
509 + * pages whose large inline <style> blocks defeat tempered-greedy patterns.
510 + * Historically that NULL propagated through the rest of the pipeline, so
511 + * process_html_directly() returned empty HTML, the render strategy's 50%-size
512 + * sanity check rejected it, and the ORIGINAL un-optimised page was served —
513 + * OTTO silently did nothing.
514 + *
515 + * This guard keeps the last-good HTML whenever a transform yields NULL (or an
516 + * unexpectedly empty string), so a single failing regex can no longer discard
517 + * every OTTO optimisation. It is defence-in-depth: the specific pattern that
518 + * triggered (strip_hoisted_facade_styles) has been rewritten to be
519 + * non-backtracking, but this guard still protects the rest of the pipeline.
520 + *
521 + * @param string|null $new Result of the transform.
522 + * @param string $prev HTML before the transform (fallback value).
523 + * @param string $where Transform name, for the log line.
524 + * @return string
525 + */
526 + private function otto_guard_html($new, $prev, $where = '') {
527 + if ($new === null || (is_string($new) && $new === '' && $prev !== '')) {
528 + # A full-document transform ($where) returned NULL (e.g. a PCRE
529 + # limit such as "JIT stack limit exhausted"). Keep the last-good
530 + # HTML so one failed step can't discard every OTTO optimisation.
531 + # Intentionally silent — not logged, to keep customer logs clean.
532 + return $prev;
533 + }
534 + return $new;
535 + }
536 +
254 537 private function strip_hoisted_facade_styles($html) {
255 - return preg_replace(
256 - '#<style\b[^>]*>(?:(?!</style>).)*?\bimg\s*,\s*span\s*\{[^}]*position\s*:\s*absolute[^}]*\}(?:(?!</style>).)*?</style>#is',
257 - '',
538 + # Non-backtracking implementation.
539 + #
540 + # The previous single-regex approach used two tempered-greedy segments
541 + # ((?:(?!</style>).)*?) to keep the match inside one <style> block. That
542 + # runs a negative lookahead for every character, and on Divi/page-builder
543 + # pages whose inline <style> blocks are tens of KB it overflows the PCRE
544 + # JIT stack ("JIT stack limit exhausted"). preg_replace() then returns
545 + # NULL, which discarded the entire OTTO-modified document.
546 + #
547 + # Instead, isolate each <style>…</style> block with a single lazy .*?
548 + # (cheap, no per-char lookahead) and test only that bounded block for the
549 + # facade signature. The signature check runs on one small block at a time,
550 + # so neither pattern can strain the JIT stack.
551 + if (!is_string($html) || stripos($html, '<style') === false) {
552 + return $html;
553 + }
554 +
555 + $out = preg_replace_callback(
556 + '#<style\b[^>]*>.*?</style>#is',
557 + static function ($m) {
558 + # Facade reset: "img,span{ … position:absolute … }" inside this block.
559 + if (preg_match('#\bimg\s*,\s*span\s*\{[^}]*position\s*:\s*absolute[^}]*\}#is', $m[0])) {
560 + return '';
561 + }
562 + return $m[0];
563 + },
258 564 $html
259 565 );
566 +
567 + # Belt-and-suspenders: if PCRE still fails for any reason, keep the input.
568 + return $out === null ? $html : $out;
260 569 }
261 570
262 571 /**
263 - * WP-465: Keep the charset declaration within the first 1024 bytes of <head>.
572 + * Mark the <head> as OTTO-processed without altering the <head> tag itself.
264 573 *
574 + * This marker used to be an attribute — <head metasync_optimized> — which
575 + * removed the literal string "<head>" from the document. Anything that
576 + * injects into the head with a literal str_replace('<head>', ...) then found
577 + * no match on the real tag and silently matched the *next* literal "<head>"
578 + * further down the page. That next one can sit inside a JS comment, a <pre>
579 + * block or escaped documentation text; injecting a <script>...</script> there
580 + * closes the enclosing script early, strands the rest of its code as a text
581 + * node in <head>, and the browser therefore ends <head> at that point —
582 + * pushing title, canonical, OG/Twitter and JSON-LD into <body>, where
583 + * crawlers and social scrapers ignore them.
584 + *
585 + * A comment placed immediately after the tag carries the same signal and
586 + * leaves the tag byte-identical. Note that an HTML minifier may strip
587 + * comments, so treat the <meta name="otto"> tag as the durable marker; this
588 + * one is a convenience for support and QA.
589 + *
590 + * Only the first <head> is marked. The real tag is always the first one in
591 + * the document, so this pass cannot wander the way a bare literal match can.
592 + *
593 + * @param string $html Serialized HTML.
594 + * @return string
595 + */
596 + private function mark_head_optimized($html)
597 + {
598 + # Idempotent, and also a no-op on pages still carrying the legacy
599 + # attribute form (e.g. HTML already in a page cache).
600 + if ($html === '' || strpos($html, self::HEAD_OPTIMIZED_TOKEN) !== false) {
601 + return $html;
602 + }
603 +
604 + $out = preg_replace(
605 + '/(<head(?:\s[^>]*)?>)/i',
606 + '$1' . self::HEAD_OPTIMIZED_MARKER,
607 + $html,
608 + 1
609 + );
610 +
611 + # Belt-and-suspenders: if PCRE fails for any reason, keep the input.
612 + return $out === null ? $html : $out;
613 + }
614 +
615 + /**
616 + * Apply OTTO's canonical via string replacement (reliable fallback).
617 + *
618 + * Why the DOM path fails: do_header_replacements() sets the canonical with a
619 + * DOM attribute edit ($link->href = …). That edit works in isolation, but
620 + * insert_header_html() runs *earlier* in the same pipeline and reassigns
621 + * $head->outertext to a raw string (to inject OTTO's schema). In SimpleHtmlDom,
622 + * assigning ->outertext freezes that node — <head> now serializes as that
623 + * literal string and no longer re-renders from its child node tree. The
624 + * canonical <link> is captured inside that frozen string with its OLD href, so
625 + * the later $link->href edit is never reflected in the output and the SEO
626 + * plugin's (Yoast/Rank Math/AIOSEO) canonical always wins. This is the same
627 + * head-freeze that forced title/meta onto string-replacement fallbacks;
628 + * canonical was the one case that never got one.
629 + *
630 + * This pass runs on the final serialized HTML (after the freeze), so it is
631 + * immune to the ordering problem. It guarantees exactly one canonical —
632 + * OTTO's: it removes every existing <link rel="canonical"> and inserts OTTO's
633 + * recommended value (marked data-otto="true") right after <head>. Manual
634 + * canonicals (_metasync_canonical_url / meta_canonical) still take priority,
635 + * matching the DOM path's protection.
636 + *
637 + * @param string $html Serialized HTML.
638 + * @param array $replacement_data OTTO suggestions.
639 + * @return string
640 + */
641 + private function apply_canonical_via_string($html, $replacement_data) {
642 + if (!is_string($html) || empty($replacement_data['header_replacements']) || !is_array($replacement_data['header_replacements'])) {
643 + return $html;
644 + }
645 +
646 + # Find OTTO's canonical recommendation.
647 + $canonical = '';
648 + foreach ($replacement_data['header_replacements'] as $item) {
649 + if (($item['type'] ?? '') === 'link' && ($item['rel'] ?? '') === 'canonical') {
650 + $canonical = $item['recommended_value'] ?? $item['value'] ?? '';
651 + break;
652 + }
653 + }
654 + if (empty($canonical) || !is_string($canonical)) {
655 + return $html;
656 + }
657 +
658 + # Validate OTTO's own suggestion too: if the platform payload
659 + # carries a corrupted/non-URL canonical, leave the document untouched.
660 + $canonical = Metasync_Canonical_Sanitizer::sanitize($canonical);
661 + if ($canonical === '') {
662 + return $html;
663 + }
664 +
665 + # Respect a manually-set canonical (same protection as the DOM path).
666 + # Validated: a legacy row corrupted to "Array" must not count
667 + # as a manual canonical — that would suppress OTTO's correct value.
668 + if (function_exists('is_singular') && is_singular()) {
669 + $post_id = function_exists('get_the_ID') ? get_the_ID() : 0;
670 + if ($post_id) {
671 + $custom = Metasync_Canonical_Sanitizer::sanitize(get_post_meta($post_id, '_metasync_canonical_url', true));
672 + if ($custom === '') {
673 + $custom = Metasync_Canonical_Sanitizer::sanitize(get_post_meta($post_id, 'meta_canonical', true));
674 + }
675 + if ($custom !== '') {
676 + return $html; # manual canonical wins — leave the document untouched
677 + }
678 + }
679 + }
680 +
681 + # Only proceed if there is a <head> to place the tag in (never end up with zero canonical).
682 + if (!preg_match('#<head\b[^>]*>#i', $html)) {
683 + return $html;
684 + }
685 +
686 + $tag = '<link rel="canonical" href="' . htmlspecialchars($canonical, ENT_QUOTES, 'UTF-8') . '" data-otto="true" />';
687 +
688 + # Remove every existing canonical link (bounded per-tag pattern; null-safe).
689 + $stripped = preg_replace('#<link\b[^>]*\brel=(["\'])canonical\1[^>]*>\s*#i', '', $html);
690 + if (is_string($stripped)) {
691 + $html = $stripped;
692 + }
693 +
694 + # Insert OTTO's canonical right after <head> (callback avoids $/\ interpolation from the URL).
695 + $inserted = preg_replace_callback('#(<head\b[^>]*>)#i', function ($m) use ($tag) {
696 + return $m[1] . "\n" . $tag;
697 + }, $html, 1);
698 + if (is_string($inserted)) {
699 + $html = $inserted;
700 + }
701 +
702 + return $html;
703 + }
704 +
705 + /**
706 + * Keep the charset declaration within the first 1024 bytes of <head>.
707 + *
265 708 * OTTO injects meta tags + JSON-LD schema at the top of <head>, which can push
266 709 * the theme's <meta charset="utf-8"> past the 1024-byte limit that browsers
267 710 * enforce for in-document charset detection (HTML spec). When that happens —
268 711 * and a cached/CDN response is served without an HTTP charset header — the
@@ -342,9 +785,9 @@
342 785 $response = wp_remote_get($url_with_params, $args);
343 786
344 787 # Check for errors
345 788 if (is_wp_error($response)) {
346 - error_log('MetaSync ' . Metasync::get_whitelabel_otto_name() . ': API call failed - ' . $response->get_error_message());
789 + error_log('MetaSync ' . Metasync::get_whitelabel_otto_name() . ': API call failed - ' . $response->get_error_message()); // phpcs:ignore WordPress.PHP.DevelopmentFunctions.error_log_error_log -- unreachable in production
347 790 return false;
348 791 }
349 792
350 793 # get the response body
@@ -354,9 +797,9 @@
354 797 $response_code = wp_remote_retrieve_response_code($response);
355 798
356 799 # if no change data skip
357 800 if (empty($body) || $response_code !== 200){
358 - error_log('MetaSync ' . Metasync::get_whitelabel_otto_name() . ': API returned empty or non-200. Code: ' . $response_code);
801 + error_log('MetaSync ' . Metasync::get_whitelabel_otto_name() . ': API returned empty or non-200. Code: ' . $response_code); // phpcs:ignore WordPress.PHP.DevelopmentFunctions.error_log_error_log -- unreachable in production
359 802 return false;
360 803 }
361 804
362 805 # set the html file path
@@ -549,9 +992,11 @@
549 992
550 993 # Check for timeout or connection errors
551 994 if (is_wp_error($route_html)) {
552 995 $error_msg = $route_html->get_error_message();
553 - error_log('MetaSync ' . Metasync::get_whitelabel_otto_name() . ' DEBUG: FAILED - wp_remote_get error: ' . $error_msg . ' for route: ' . $route);
996 + if (function_exists('metasync_otto_report_render_failure')) {
997 + metasync_otto_report_render_failure('HTTP_FETCH_FAILED', 'Internal fetch failed: ' . $error_msg, [], 'warning');
998 + }
554 999 return false;
555 1000 }
556 1001
557 1002 # get body
@@ -559,25 +1004,72 @@
559 1004
560 1005 # Get the response code
561 1006 $response_code = wp_remote_retrieve_response_code($route_html);
562 1007
1008 + // Requests follows redirects and returns the final body with its final
1009 + // status. Do not put that body on the original route when a redirect
1010 + // changed the URL: doing so creates a soft redirect with mismatched
1011 + // content and SEO metadata. The normal WordPress request can handle
1012 + // the redirect, and OTTO will process the destination on its own URL.
1013 + $effective_url = $this->get_effective_fetch_url($route_html);
1014 + if ($effective_url !== '' && !$this->fetch_url_matches_route($effective_url, $route, $request_body)) {
1015 + return false;
1016 + }
563 1017
1018 + # Capture the internal fetch's `Set-Cookie`. On the HTTP path OTTO keeps only the
1019 + # body below and would otherwise discard this — but it is both the form/session
1020 + # signal for the caching decision AND (same-origin only) the session the visitor
1021 + # needs for their forms to work. $request_body is the URL actually fetched (it may
1022 + # be rewritten to localhost for tunnels), so only allow pass-through when its
1023 + # scheme+host match the visitor's; otherwise we would set a wrong-domain cookie.
1024 + if (class_exists('Metasync_Otto_Render_Strategy')) {
1025 + $fetch_set_cookie = wp_remote_retrieve_header($route_html, 'set-cookie');
1026 + $fetch_parts = wp_parse_url($request_body);
1027 + $visitor_host = preg_replace('/:\d+$/', '', $_SERVER['HTTP_HOST'] ?? '');
1028 + $same_origin = !empty($fetch_parts['host'])
1029 + && strcasecmp($fetch_parts['host'], $visitor_host) === 0
1030 + && ((($fetch_parts['scheme'] ?? '') === 'https') === (bool) is_ssl());
1031 + Metasync_Otto_Render_Strategy::set_http_fetch_cookie($fetch_set_cookie, $same_origin);
1032 +
1033 + # Also record the internal response's own `Cache-Control`. When the page's
1034 + # render starts a PHP session, PHP emits its session cache-limiter headers
1035 + # (`no-store, no-cache, must-revalidate`) — and it does so whether the session
1036 + # is new OR resumed, so this catches session pages that set no NEW cookie and
1037 + # would otherwise look cacheable. It is read as a signal only and never
1038 + # forwarded to the visitor: our own no-cache request header is not echoed into
1039 + # the response, so this cannot make ordinary pages uncacheable.
1040 + Metasync_Otto_Render_Strategy::set_http_fetch_cache_control(
1041 + wp_remote_retrieve_header($route_html, 'cache-control')
1042 + );
1043 + }
1044 +
1045 +
564 1046 # check not empty
565 1047 if(empty($html_body) || $response_code !== 200){
566 - error_log('MetaSync ' . Metasync::get_whitelabel_otto_name() . ' DEBUG: FAILED - Empty body or non-200 status for route: ' . $route);
1048 + if (function_exists('metasync_otto_report_render_failure')) {
1049 + metasync_otto_report_render_failure('HTTP_FETCH_EMPTY', 'Internal fetch returned empty body or non-200 status (HTTP ' . intval($response_code) . ')', [], 'warning');
1050 + }
567 1051 return false;
568 1052 }
569 1053
1054 + # Skip DOM processing on oversized documents to avoid fatal OOM.
1055 + if (class_exists('Metasync_Otto_Render_Strategy')
1056 + && !Metasync_Otto_Render_Strategy::is_document_processable(strlen($html_body))
1057 + ) {
1058 + Metasync_Otto_Render_Strategy::log_oversized_skip('handle_route_html', strlen($html_body));
1059 + return false;
1060 + }
1061 +
570 1062 # Remove XML declaration
571 1063 $html_body = preg_replace('/<\?xml[^?]*\?>\s*/i', '', $html_body);
572 1064
573 - # WP-355 / WP-315: Save ALL original <style> blocks before DOM processing.
1065 + # / Save ALL original <style> blocks before DOM processing.
574 1066 # SimpleHtmlDom can lose or corrupt style blocks during parse/serialize —
575 1067 # not only Divi's id-tagged ones but also anonymous <style> blocks used by
576 1068 # custom themes for nav collapse CSS, responsive breakpoints, etc.
577 1069 $original_style_blocks = $this->capture_style_blocks($html_body);
578 1070
579 - # WP-465: Shield entity-encoded srcdoc values (lazy YouTube/video facades)
1071 + # Shield entity-encoded srcdoc values (lazy YouTube/video facades)
580 1072 # so SimpleHtmlDom can't decode them and hoist their global <style> into <head>.
581 1073 $srcdoc_store = array();
582 1074 $html_body = $this->protect_srcdoc_attributes($html_body, $srcdoc_store);
583 1075
@@ -583,12 +1075,16 @@
583 1075
584 1076 # Escape bare < in text content (e.g. "<4 microns") before DOM parsing
585 1077 $html_body = $this->sanitize_text_less_than($html_body);
586 1078
587 - # WP-355: Fix malformed self-closing non-void tags (e.g. <ul/ class="...">)
1079 + # Fix malformed self-closing non-void tags (e.g. <ul/ class="...">)
588 1080 # before DOM parsing — SimpleHtmlDom strips attributes from these.
589 1081 $html_body = $this->fix_malformed_self_closing_tags($html_body);
590 1082
1083 + # Quote unquoted attribute values (e.g. data-autoplay-viewport=80%) —
1084 + # SimpleHtmlDom swallows every following attribute into such a value.
1085 + $html_body = $this->quote_unquoted_attribute_values($html_body);
1086 +
591 1087 # now that the html is not empty
592 1088 # load it into the simple html dom
593 1089 $this->dom->load($html_body, true, false);
594 1090
@@ -716,13 +1212,19 @@
716 1212
717 1213 $result_html = preg_replace_callback(
718 1214 '/(<' . $heading_type . '(?:\s[^>]*)?>)(.*?)(<\/' . $heading_type . '>)/is',
719 1215 function ($m) use ($current_value, $recommended_value) {
720 - $inner_text = trim(preg_replace('/\s+/', ' ', html_entity_decode(strip_tags($m[2]), ENT_QUOTES, 'UTF-8')));
721 - if ($inner_text === $current_value) {
722 - return $m[1] . $recommended_value . $m[3];
1216 + $inner_text = trim(preg_replace('/\s+/', ' ', html_entity_decode(wp_strip_all_tags($m[2]), ENT_QUOTES, 'UTF-8')));
1217 + if ($inner_text !== $current_value) {
1218 + return $m[0];
723 1219 }
724 - return $m[0];
1220 + # Keep the author's inline formatting; skip when it can't
1221 + # be preserved rather than flattening the heading.
1222 + $new_inner = self::build_heading_inner_html($m[2], $recommended_value);
1223 + if ($new_inner === null) {
1224 + return $m[0];
1225 + }
1226 + return $m[1] . $new_inner . $m[3];
725 1227 },
726 1228 $result_html,
727 1229 -1
728 1230 );
@@ -730,19 +1232,20 @@
730 1232 }
731 1233
732 1234 # CRITICAL FIX: Apply image alt text manually via string replacement
733 1235 # DOM changes via SimpleHtmlDom don't persist on Oxygen/page-builder sites using HTTP render path
734 - $result_html = $this->apply_image_alt_text_via_string($result_html, $replacement_data);
1236 + $result_html = $this->otto_guard_html($this->apply_image_alt_text_via_string($result_html, $replacement_data), $result_html, 'apply_image_alt_text_via_string');
735 1237
736 1238 # Apply insertions — only if DOM insertion didn't already apply it
737 1239 if (!empty($replacement_data['header_html_insertion'])) {
738 - $header_html_check = trim($replacement_data['header_html_insertion']);
1240 + # Escape JSON-LD payload closers and stamp OTTO's blocks first; the DOM
1241 + # path inserts the same bytes, so the already-applied check must compare
1242 + # against the stamped form or it would insert a second copy.
1243 + $header_html_insertion = $this->stamp_otto_json_ld(
1244 + metasync_escape_json_ld_blocks_in_html($replacement_data['header_html_insertion'])
1245 + );
1246 + $header_html_check = trim($header_html_insertion);
739 1247 if (strpos($result_html, $header_html_check) === false) {
740 - $header_html_insertion = preg_replace(
741 - '/<script(\s[^>]*)type\s*=\s*(["\'])application\/ld\+json\2/i',
742 - '<script$1type=$2application/ld+json$2 data-otto="true"',
743 - $replacement_data['header_html_insertion']
744 - );
745 1248 $safe_header = str_replace(array('\\', '$'), array('\\\\', '\\$'), $header_html_insertion);
746 1249 $result_html = preg_replace('/(<\/head>)/i', $safe_header . "\n" . '$1', $result_html, 1);
747 1250 }
748 1251 }
@@ -764,44 +1267,70 @@
764 1267 $safe_footer = str_replace(array('\\', '$'), array('\\\\', '\\$'), $replacement_data['footer_html_insertion']);
765 1268 $result_html = preg_replace('/(<\/html>)/i', $safe_footer . "\n" . '$1', $result_html, 1);
766 1269 }
767 1270
768 - # DEDUPLICATION: Remove duplicate <title>, OG, Twitter tags, canonical, and JSON-LD schema
769 - $result_html = $this->deduplicate_title_tags($result_html);
770 - $result_html = $this->deduplicate_og_twitter_tags($result_html);
771 - $result_html = $this->deduplicate_schema_tags($result_html);
772 - $result_html = $this->deduplicate_canonical_tags($result_html);
1271 + # Apply OTTO's canonical via string replacement (DOM edit is clobbered by the insert_header_html head-freeze; see apply_canonical_via_string).
1272 + $result_html = $this->apply_canonical_via_string($result_html, $replacement_data);
773 1273
774 - # Ensure metasync_optimized attribute on <head> (post-serialization so dom->clear() can't wipe it)
775 - if (!$this->is_amp_page() && strpos($result_html, 'metasync_optimized') === false) {
776 - $result_html = preg_replace('/<head(\s|>)/i', '<head metasync_optimized$1', $result_html, 1);
1274 + # DEDUPLICATION: Remove duplicate <title>, meta description, OG, Twitter tags, canonical, and JSON-LD schema
1275 + $result_html = $this->otto_guard_html($this->deduplicate_title_tags($result_html), $result_html, 'deduplicate_title_tags');
1276 + $result_html = $this->otto_guard_html($this->deduplicate_description_tags($result_html), $result_html, 'deduplicate_description_tags');
1277 + $result_html = $this->otto_guard_html($this->deduplicate_keywords_tags($result_html), $result_html, 'deduplicate_keywords_tags');
1278 + $result_html = $this->otto_guard_html($this->deduplicate_og_twitter_tags($result_html), $result_html, 'deduplicate_og_twitter_tags');
1279 + $result_html = $this->otto_guard_html($this->apply_metabox_og_precedence($result_html), $result_html, 'apply_metabox_og_precedence');
1280 + $result_html = $this->otto_guard_html($this->apply_custom_seo_precedence($result_html), $result_html, 'apply_custom_seo_precedence');
1281 + $result_html = $this->otto_guard_html($this->deduplicate_schema_tags($result_html), $result_html, 'deduplicate_schema_tags');
1282 + $result_html = $this->otto_guard_html($this->deduplicate_canonical_tags($result_html), $result_html, 'deduplicate_canonical_tags');
1283 +
1284 + # Mark the <head> as processed (post-serialization so dom->clear() can't wipe it)
1285 + if (!$this->is_amp_page()) {
1286 + $result_html = $this->mark_head_optimized($result_html);
777 1287 }
778 1288
779 - $result_html = $this->restore_case_sensitive_attributes($result_html);
1289 + $result_html = $this->otto_guard_html($this->restore_case_sensitive_attributes($result_html), $result_html, 'restore_case_sensitive_attributes');
780 1290
781 - # WP-355 / WP-315: Re-inject any <style> blocks lost during processing.
782 - $result_html = $this->restore_lost_style_blocks($result_html, $original_style_blocks);
1291 + # / Re-inject any <style> blocks lost during processing.
1292 + $result_html = $this->otto_guard_html($this->restore_lost_style_blocks($result_html, $original_style_blocks), $result_html, 'restore_lost_style_blocks');
783 1293
784 - # WP-315: Fix Divi 5 shortcode framework class renumbering (HTTP path)
785 - $result_html = $this->fix_divi_class_renumbering($result_html);
1294 + # Fix Divi 5 shortcode framework class renumbering (HTTP path)
1295 + $result_html = $this->otto_guard_html($this->fix_divi_class_renumbering($result_html), $result_html, 'fix_divi_class_renumbering');
786 1296
787 - # WP-315: Clean OTTO internal fetch params from HTML output.
1297 + # Clean OTTO internal fetch params from HTML output.
788 1298 # The HTTP render uses wp_remote_get with ?is_otto_page_fetch=1&otto_block_title=1&otto_block_desc=1
789 1299 # These leak into form action URLs, canonical links, etc. in the rendered HTML.
790 - $result_html = $this->clean_otto_fetch_params($result_html);
1300 + $result_html = $this->otto_guard_html($this->clean_otto_fetch_params($result_html), $result_html, 'clean_otto_fetch_params');
791 1301
792 - # WP-465: Remove any lazy-video facade <style> hoisted into the page
1302 + # Remove any lazy-video facade <style> hoisted into the page
793 1303 # (runs while srcdoc is still tokenized, so the iframe's own copy is safe).
1304 + $result_html = $this->otto_guard_html($this->strip_hoisted_facade_styles($result_html), $result_html, 'strip_hoisted_facade_styles');
1305 +
1306 + # Restore the original encoded srcdoc values (undo tokenization).
1307 + $result_html = $this->otto_guard_html($this->restore_srcdoc_attributes($result_html, $srcdoc_store), $result_html, 'restore_srcdoc_attributes');
1308 +
1309 + # Keep <meta charset> within the first 1024 bytes so external CSS
1310 + # (e.g. checkmark content:"✓") doesn't mojibake on cached responses.
1311 + $result_html = $this->otto_guard_html($this->ensure_early_charset_meta($result_html), $result_html, 'ensure_early_charset_meta');
1312 +
1313 + # undo double-encoded numeric/hex character references produced by the
1314 + # bundled simplehtmldom serializer from attributes like data-x-icon-s="&#xf3c5".
1315 + $result_html = $this->otto_guard_html($this->repair_double_encoded_entities($result_html), $result_html, 'repair_double_encoded_entities');
1316 +
1317 + # restore literal '&' query-string separators the serializer escaped to
1318 + # '&amp;', which silently truncates multi-family Google Fonts URLs.
1319 + $result_html = $this->otto_guard_html($this->repair_query_string_ampersands($result_html), $result_html, 'repair_query_string_ampersands');
1320 +
1321 + # Remove any lazy-video facade <style> hoisted into the page
1322 + # (runs while srcdoc is still tokenized, so the iframe's own copy is safe).
794 1323 $result_html = $this->strip_hoisted_facade_styles($result_html);
795 1324
796 - # WP-465: Restore the original encoded srcdoc values (undo tokenization).
1325 + # Restore the original encoded srcdoc values (undo tokenization).
797 1326 $result_html = $this->restore_srcdoc_attributes($result_html, $srcdoc_store);
798 1327
799 - # WP-465: Keep <meta charset> within the first 1024 bytes so external CSS
1328 + # Keep <meta charset> within the first 1024 bytes so external CSS
800 1329 # (e.g. checkmark content:"✓") doesn't mojibake on cached responses.
801 1330 $result_html = $this->ensure_early_charset_meta($result_html);
802 1331
803 - # WP-471: undo double-encoded numeric/hex character references produced by the
1332 + # undo double-encoded numeric/hex character references produced by the
804 1333 # bundled simplehtmldom serializer from attributes like data-x-icon-s="&#xf3c5".
805 1334 $result_html = $this->repair_double_encoded_entities($result_html);
806 1335
807 1336 return $result_html;
@@ -807,10 +1336,113 @@
807 1336 return $result_html;
808 1337 }
809 1338
810 1339 /**
811 - * WP-471: Repair double-encoded character references in serialized HTML.
1340 + * Read the final URL from the WordPress HTTP response when Requests has
1341 + * followed one or more redirects. Older HTTP transports may not expose it;
1342 + * in that case the empty result leaves the existing fail-open behavior.
812 1343 *
1344 + * @param mixed $response wp_remote_get() response.
1345 + * @return string Final URL, or an empty string when unavailable.
1346 + */
1347 + private function get_effective_fetch_url($response)
1348 + {
1349 + if (!is_array($response) || empty($response['http_response']) || !is_object($response['http_response'])) {
1350 + return '';
1351 + }
1352 +
1353 + $http_response = $response['http_response'];
1354 + if (!method_exists($http_response, 'get_response_object')) {
1355 + return '';
1356 + }
1357 +
1358 + $requests_response = $http_response->get_response_object();
1359 + if (!is_object($requests_response) || empty($requests_response->url) || !is_string($requests_response->url)) {
1360 + return '';
1361 + }
1362 +
1363 + return $requests_response->url;
1364 + }
1365 +
1366 + /**
1367 + * Determine whether a followed response still represents the requested
1368 + * route. The internal fetch may use a rewritten host for tunnel support,
1369 + * so both the original and transport hosts are accepted; the path and
1370 + * non-OTTO query parameters must remain identical.
1371 + *
1372 + * @param string $effective_url Final URL reported by Requests.
1373 + * @param string $route Original route supplied to OTTO.
1374 + * @param string $fetch_url URL actually passed to wp_remote_get().
1375 + * @return bool True when no meaningful redirect changed the route.
1376 + */
1377 + private function fetch_url_matches_route($effective_url, $route, $fetch_url)
1378 + {
1379 + $effective = wp_parse_url($effective_url);
1380 + $requested = wp_parse_url($route);
1381 + $transport = wp_parse_url($fetch_url);
1382 + if (!is_array($effective) || !is_array($requested)) {
1383 + return true;
1384 + }
1385 +
1386 + $effective_path = isset($effective['path']) ? $effective['path'] : '/';
1387 + $requested_path = isset($requested['path']) ? $requested['path'] : '/';
1388 + if ($effective_path !== $requested_path) {
1389 + return false;
1390 + }
1391 +
1392 + $allowed_hosts = [];
1393 + foreach ([$requested, $transport] as $parts) {
1394 + if (!empty($parts['host'])) {
1395 + $host = strtolower($parts['host']);
1396 + if (!empty($parts['port'])) {
1397 + $host .= ':' . $parts['port'];
1398 + }
1399 + $allowed_hosts[] = $host;
1400 + }
1401 + }
1402 + if (!empty($effective['host'])) {
1403 + $effective_host = strtolower($effective['host']);
1404 + if (!empty($effective['port'])) {
1405 + $effective_host .= ':' . $effective['port'];
1406 + }
1407 + if (!empty($allowed_hosts) && !in_array($effective_host, $allowed_hosts, true)) {
1408 + return false;
1409 + }
1410 + }
1411 +
1412 + return $this->without_otto_fetch_params($effective['query'] ?? '')
1413 + === $this->without_otto_fetch_params($requested['query'] ?? '');
1414 + }
1415 +
1416 + /**
1417 + * Remove parameters used only to mark OTTO's internal request.
1418 + *
1419 + * @param string $query Query string without the leading '?'.
1420 + * @return string Query with OTTO markers removed.
1421 + */
1422 + private function without_otto_fetch_params($query)
1423 + {
1424 + $query = (string) $query;
1425 + if ($query === '') {
1426 + return '';
1427 + }
1428 +
1429 + $parts = [];
1430 + foreach (explode('&', $query) as $part) {
1431 + $name = strstr($part, '=', true);
1432 + $name = $name === false ? $part : $name;
1433 + if (in_array($name, ['is_otto_page_fetch', 'otto_block_title', 'otto_block_desc'], true)) {
1434 + continue;
1435 + }
1436 + $parts[] = $part;
1437 + }
1438 +
1439 + return implode('&', $parts);
1440 + }
1441 +
1442 + /**
1443 + * Repair double-encoded character references in serialized HTML.
1444 + *
813 1445 * The bundled simplehtmldom serializer (HtmlNode::makeup(), reached via
814 1446 * $this->dom->save() and $root->outertext) runs every attribute value through
815 1447 * htmlentities(), which escapes the leading '&' of numeric/hex character
816 1448 * references. Theme icon attributes such as data-x-icon-s="&#xf3c5"
@@ -831,8 +1463,118 @@
831 1463 }
832 1464 return preg_replace('/&amp;(#x[0-9a-fA-F]+;?|#[0-9]+;?)/', '&$1', $html);
833 1465 }
834 1466
1467 + /**
1468 + * Restore literal '&' query-string separators inside URL attributes.
1469 + *
1470 + * HtmlNode::makeup() serialises every attribute through htmlentities(), so a
1471 + * separator that arrived as a bare '&' - or as '&#038;', which HtmlDocument
1472 + * decodes to '&' when it parses the attribute - is written back as '&amp;'.
1473 + * Both forms decode to '&' in a conforming HTML parser, so the browser is
1474 + * unaffected. A consumer that reads the raw attribute WITHOUT decoding
1475 + * entities sees a different URL, and Cloudflare Fonts is one of those: it
1476 + * refetches fonts.googleapis.com/css2?family=A&amp;family=B, Google treats
1477 + * only the first 'family=' as a real parameter (the rest arrive named
1478 + * 'amp;family'), and every font after the first is dropped from the response.
1479 + * The page keeps its font-family declarations but loses the @font-face rules,
1480 + * so text silently falls back to a system font.
1481 + *
1482 + * Rewrites only a '&amp;' that is immediately followed by a parameter name
1483 + * and '=', and only inside the listed URL attributes. A '&amp;' elsewhere is
1484 + * left escaped, and the mandatory '=' in the lookahead (with ';' excluded
1485 + * from the name class) means no ambiguous ampersand ('&' + alphanumerics +
1486 + * ';') is ever produced.
1487 + *
1488 + * All three delimiter forms the serializer can emit are handled: double
1489 + * quoted, single quoted, and unquoted - makeup() preserves HDOM_QUOTE_NO and
1490 + * emits no delimiter for a value with no whitespace, which is common on
1491 + * minified themes. ENT_COMPAT also leaves "'" unescaped, so a double-quoted
1492 + * value may legitimately contain an apostrophe.
1493 + *
1494 + * The attribute list stops at attributes whose value is always a URL.
1495 + * 'content' is deliberately excluded because it also carries prose
1496 + * (<meta name="description" content="A &amp; B">), where unescaping would
1497 + * corrupt visible text. Attributes read only via JavaScript getAttribute()
1498 + * do not need it either - the DOM decodes entities for them; only consumers
1499 + * that scan raw HTML are affected.
1500 + *
1501 + * <script>/<style>/<textarea>/<noscript> bodies and comments are shielded
1502 + * first. Those are verbatim or RCDATA regions, not markup: rewriting bytes
1503 + * there would edit a JS string literal, or turn "&amp;copy=2" in a textarea
1504 + * into "(c)=2" once the browser decodes it.
1505 + *
1506 + * Plain negated character classes only - no tempered-greedy lookaheads, which
1507 + * blow the PCRE JIT stack on pages with very large inline blocks.
1508 + *
1509 + * @param string $html Serialized HTML from save()/outertext.
1510 + * @return string
1511 + */
1512 + private function repair_query_string_ampersands($html) {
1513 + # strpos guard: skip the regex entirely when there is nothing to repair
1514 + if (strpos($html, '&amp;') === false) {
1515 + return $html;
1516 + }
1517 +
1518 + # Shield verbatim / RCDATA regions and comments before touching anything.
1519 + # Masking also shrinks large inline blocks to a token, so the attribute
1520 + # pass never walks them.
1521 + $shielded = array();
1522 + $masked = preg_replace_callback(
1523 + '/<(script|style|textarea|noscript)\b[^>]*>.*?<\/\1>|<!--.*?-->/is',
1524 + function ($m) use (&$shielded) {
1525 + $token = '<!--METASYNC_QSA_' . count($shielded) . '-->';
1526 + $shielded[$token] = $m[0];
1527 + return $token;
1528 + },
1529 + $html
1530 + );
1531 +
1532 + # preg_* returns null on a PCRE failure; keep the original document.
1533 + if ($masked === null) {
1534 + return $html;
1535 + }
1536 +
1537 + $repaired = preg_replace_callback(
1538 + '/(\s(?:href|src|srcset|data-src|data-srcset|poster|action)\s*=\s*)(?:"([^"<>]*)"|\'([^\'<>]*)\'|([^\s"\'<>]+))/i',
1539 + function ($m) {
1540 + # Pick the delimiter branch that matched. An empty value carries
1541 + # no query string, so the ambiguity between an empty quoted value
1542 + # and an unset group cannot change the outcome.
1543 + if (isset($m[4]) && $m[4] !== '') { // @phpstan-ignore notIdentical.alwaysTrue (preg fills non-participating groups with '' - this check is what tells the unquoted branch from the quoted ones)
1544 + $quote = '';
1545 + $value = $m[4];
1546 + } elseif (isset($m[3]) && $m[3] !== '') {
1547 + $quote = "'";
1548 + $value = $m[3];
1549 + } else {
1550 + $quote = '"';
1551 + $value = isset($m[2]) ? $m[2] : '';
1552 + }
1553 +
1554 + # Only a value carrying a query string can hold separators.
1555 + if (strpos($value, '?') === false || strpos($value, '&amp;') === false) {
1556 + return $m[0];
1557 + }
1558 +
1559 + $fixed = preg_replace('/&amp;(?=[A-Za-z0-9_.\[\]%\-]+=)/', '&', $value);
1560 +
1561 + if ($fixed === null) {
1562 + return $m[0];
1563 + }
1564 +
1565 + return $m[1] . $quote . $fixed . $quote;
1566 + },
1567 + $masked
1568 + );
1569 +
1570 + if ($repaired === null) {
1571 + return $html;
1572 + }
1573 +
1574 + return empty($shielded) ? $repaired : strtr($repaired, $shielded);
1575 + }
1576 +
835 1577 # do the footer html insertion
836 1578 function do_footer_html_insertion($replacement_data){
837 1579
838 1580 # check that we have footer html
@@ -933,9 +1675,9 @@
933 1675 if (empty($images)) {
934 1676 return;
935 1677 }
936 1678
937 - # WP-425: index suggestions by URL path as well, so images authored with
1679 + # index suggestions by URL path as well, so images authored with
938 1680 # relative srcs still match OTTO's absolute-URL keys (and vice versa).
939 1681 $alt_lookup = $this->build_image_alt_lookup($image_data);
940 1682
941 1683 # PERFORMANCE OPTIMIZATION: O(n²) reduced to O(n)
@@ -974,9 +1716,9 @@
974 1716 }
975 1717 }
976 1718 }
977 1719
978 - # WP-425: memoized site host — home_url() runs its filter chain on every
1720 + # memoized site host — home_url() runs its filter chain on every
979 1721 # call and these helpers run per suggestion and per img.
980 1722 private ?string $site_host_cache = null;
981 1723
982 1724 /**
@@ -992,9 +1734,9 @@
992 1734 return $this->site_host_cache;
993 1735 }
994 1736
995 1737 /**
996 - * WP-425: Candidate lookup keys for an image URL — the URL itself plus,
1738 + * Candidate lookup keys for an image URL — the URL itself plus,
997 1739 * for relative URLs or absolute/protocol-relative URLs on this site's
998 1740 * host, the bare URL path. URLs on a different host only ever match
999 1741 * exactly, so a same path on a foreign host cannot produce a false match.
1000 1742 *
@@ -1022,9 +1764,9 @@
1022 1764 return $variants;
1023 1765 }
1024 1766
1025 1767 /**
1026 - * WP-425: Build the image alt lookup keyed by every URL variant of each
1768 + * Build the image alt lookup keyed by every URL variant of each
1027 1769 * OTTO suggestion, so relative img srcs match absolute suggestion URLs.
1028 1770 *
1029 1771 * @param array $image_data OTTO body_substitutions.images map (url => alt)
1030 1772 * @return array Expanded lookup (url-or-path => alt)
@@ -1046,9 +1788,9 @@
1046 1788 return $lookup;
1047 1789 }
1048 1790
1049 1791 /**
1050 - * WP-425: Resolve the OTTO alt text for an img src, trying the exact src
1792 + * Resolve the OTTO alt text for an img src, trying the exact src
1051 1793 * first and then its same-host path variant.
1052 1794 *
1053 1795 * @param array $lookup Expanded lookup from build_image_alt_lookup()
1054 1796 * @param string $image_src The img element's src attribute
@@ -1111,9 +1853,9 @@
1111 1853 *
1112 1854 * Known limitation (pre-existing): the `<img[^>]*src=` pattern can match a
1113 1855 * `data-src`/`data-lazy-src` attribute that ends with the target URL before
1114 1856 * the real `src`, mis-attributing the alt on lazy-loaded markup. Tracked as
1115 - * a follow-up to WP-425.
1857 + * a follow-up to.
1116 1858 *
1117 1859 * @param string $html The HTML to process
1118 1860 * @param array $replacement_data The replacement data containing body_substitutions.images
1119 1861 * @return string Modified HTML with alt text applied
@@ -1123,9 +1865,9 @@
1123 1865 return $html;
1124 1866 }
1125 1867
1126 1868 foreach ($replacement_data['body_substitutions']['images'] as $image_url => $alt_text) {
1127 - # WP-425: prefilter and match on the same-host path variant too, so
1869 + # prefilter and match on the same-host path variant too, so
1128 1870 # relative img srcs match OTTO's absolute suggestion URLs.
1129 1871 $src_pattern = $this->get_image_src_pattern($image_url);
1130 1872
1131 1873 if (empty($alt_text) || $src_pattern === null || strpos($html, $src_pattern['needle']) === false) {
@@ -1180,9 +1922,9 @@
1180 1922 return $html;
1181 1923 }
1182 1924
1183 1925 /**
1184 - * WP-425: Build the src match for an OTTO image suggestion URL.
1926 + * Build the src match for an OTTO image suggestion URL.
1185 1927 *
1186 1928 * For relative URLs or absolute/protocol-relative URLs on this site's
1187 1929 * host, the regex matches the URL path with an optional scheme+host
1188 1930 * prefix — so a relative img src matches an absolute suggestion URL and
@@ -1188,9 +1930,10 @@
1188 1930 * prefix — so a relative img src matches an absolute suggestion URL and
1189 1931 * vice versa. URLs on a foreign host keep the original exact match, so a
1190 1932 * same path on a different host cannot produce a false positive.
1191 1933 *
1192 - * @param string $image_url OTTO suggestion URL
1934 + * @param string|int $image_url OTTO suggestion URL (may arrive as an int
1935 + * when it originates from an array key that PHP coerced)
1193 1936 * @return array|null ['needle' => strpos prefilter string,
1194 1937 * 'pattern' => regex fragment ('/' delimiter)] or null
1195 1938 */
1196 1939 private function get_image_src_pattern($image_url) {
@@ -1235,15 +1978,28 @@
1235 1978
1236 1979 # check matching text
1237 1980 if(trim($heading['current_value'] ?? '') == trim($text ?? '')){
1238 1981
1239 - # replace entire tag output — preserves attributes, removes all children (elements + text nodes)
1982 + # Rebuild the inner HTML, keeping the author's inline
1983 + # formatting. Returns null when the markup can't be preserved
1984 + # unambiguously — leave that heading alone rather than strip
1985 + # its <span style>/<em>/<br> and restyle the page.
1986 + $new_inner = self::build_heading_inner_html(
1987 + $heading_old->innertext,
1988 + $heading['recommended_value']
1989 + );
1990 +
1991 + if ($new_inner === null) {
1992 + continue;
1993 + }
1994 +
1995 + # replace the tag output, preserving the opening tag's attributes
1240 1996 $outer = $heading_old->outertext;
1241 1997 $open_end = strpos($outer, '>');
1242 1998 if ($open_end !== false) {
1243 1999 $open_tag = substr($outer, 0, $open_end + 1);
1244 2000 $close_tag = '</' . $heading['type'] . '>';
1245 - $heading_old->outertext = $open_tag . $heading['recommended_value'] . $close_tag;
2001 + $heading_old->outertext = $open_tag . $new_inner . $close_tag;
1246 2002 }
1247 2003
1248 2004 }
1249 2005 }
@@ -1345,10 +2101,10 @@
1345 2101 }
1346 2102
1347 2103 # If it's a protocol-relative URL (//example.com), check if it matches our domain
1348 2104 if (strpos($url, '//') === 0) {
1349 - $parsed_home = parse_url($home_url);
1350 - $parsed_link = parse_url($url);
2105 + $parsed_home = wp_parse_url($home_url);
2106 + $parsed_link = wp_parse_url($url);
1351 2107
1352 2108 if (isset($parsed_home['host']) && isset($parsed_link['host'])) {
1353 2109 if (strtolower($parsed_home['host']) === strtolower($parsed_link['host'])) {
1354 2110 return false;
@@ -1448,8 +2204,194 @@
1448 2204 # now do the actual html replacements
1449 2205 $body->outertext = '<body' . $attributes_string . '>'.$insert_data['body_top_html_insertion'].$body->innertext . '</body>';
1450 2206 }
1451 2207
2208 + /**
2209 + * Resolve the current singular post id reliably inside the OTTO output
2210 + * buffer / direct-render context.
2211 + *
2212 + * get_the_ID() depends on the global $post/loop state, which is not reliable
2213 + * once WordPress has finished rendering and the buffer/shutdown pass runs.
2214 + * get_queried_object_id() reads the parsed main query and stays correct there
2215 + * (same approach as apply_metabox_og_precedence()). Restricted to
2216 + * singular views: on archives/home the queried object id can be a term or
2217 + * user id, which must not be read as post meta.
2218 + *
2219 + * @return int Post id, or 0 when not applicable.
2220 + */
2221 + private function get_buffer_post_id() {
2222 + if (!function_exists('is_singular') || !is_singular()) {
2223 + return 0;
2224 + }
2225 + $post_id = function_exists('get_queried_object_id') ? get_queried_object_id() : 0;
2226 + return $post_id ? (int) $post_id : 0;
2227 + }
2228 +
2229 + /**
2230 + * Replace the first <meta $attr="$val"> content with $content_escaped.
2231 + *
2232 + * @param string $html
2233 + * @param string $attr "name" or "property"
2234 + * @param string $val e.g. "description", "og:description"
2235 + * @param string $content_escaped Already HTML-escaped content value
2236 + * @param string $marker Provenance marker for the winning tier
2237 + * @return string
2238 + */
2239 + private function force_custom_meta($html, $attr, $val, $content_escaped, $marker = 'custom') {
2240 + $marker_attribute = $marker === 'otto'
2241 + ? ' data-metasync-otto="true"'
2242 + : ' data-metasync-seo="custom"';
2243 + $tag = '<meta ' . $attr . '="' . $val . '" content="' . $content_escaped . '"' . $marker_attribute . ' />';
2244 + # The optional prefix group must end on whitespace or a quote so the
2245 + # attribute matched is $attr itself, not something merely ending in it:
2246 + # a bare [^>]*name= also matches data-name="keywords", so a third party's
2247 + # unrelated tag was consumed and rewritten while the real one survived —
2248 + # destroying their value AND leaving the duplicate the dedup had just
2249 + # collapsed. Making the group optional keeps the common
2250 + # <meta name="..." ...> form matching, where <meta\s has already consumed
2251 + # the only separator. Same rule as deduplicate_keywords_tags(); this
2252 + # helper serves title, description and keywords, so all three are covered.
2253 + $pattern = '/<meta\s(?:[^>]*[\s"\'])?' . preg_quote($attr, '/') . '\s*=\s*["\']' . preg_quote($val, '/') . '["\'][^>]*\/?>/i';
2254 + $count = 0;
2255 + $new = preg_replace_callback($pattern, function ($m) use ($tag) {
2256 + return $tag;
2257 + }, $html, 1, $count);
2258 + if ($count > 0) {
2259 + return $new;
2260 + }
2261 + # None present — insert after <head>.
2262 + return preg_replace_callback('/(<head[^>]*>)/i', function ($m) use ($tag) {
2263 + return $m[1] . "\n" . $tag;
2264 + }, $html, 1);
2265 + }
2266 +
2267 + /**
2268 + * Make sure the precedence resolver is loaded before calling it.
2269 + *
2270 + * @return bool
2271 + */
2272 + private static function precedence_available() {
2273 + if (class_exists('Metasync_Seo_Precedence')) {
2274 + return true;
2275 + }
2276 +
2277 + $file = dirname(__DIR__) . '/includes/class-metasync-seo-precedence.php';
2278 + if (is_readable($file)) {
2279 + require_once $file;
2280 + }
2281 +
2282 + return class_exists('Metasync_Seo_Precedence');
2283 + }
2284 +
2285 + /**
2286 + * Enforce the resolved SEO title, description and keywords as the FINAL
2287 + * word over OTTO.
2288 + *
2289 + * The resolver owns the global custom-versus-OTTO order. This final buffer
2290 + * pass must use it too: otherwise the setting works through the sidebar
2291 + * and third-party filters but OTTO's later string pass puts the custom
2292 + * value back on the page.
2293 + *
2294 + * OG and Twitter descriptions remain outside this method deliberately;
2295 + * social precedence is a separate follow-up scope.
2296 + *
2297 + * @param string $html Full HTML document (post-dedup).
2298 + * @return string
2299 + */
2300 + private function apply_custom_seo_precedence($html) {
2301 + if ($html === '') {
2302 + return $html;
2303 + }
2304 + $post_id = $this->get_buffer_post_id();
2305 + if (!$post_id) {
2306 + return $html;
2307 + }
2308 +
2309 + # This runs inside the output buffer, where an unresolvable class is a
2310 + # fatal that takes the whole page down — otto_guard_html() only catches
2311 + # a NULL return, not a thrown Error. A partial update can leave newer
2312 + # PHP files beside an older committed autoload classmap, so resolve the
2313 + # file directly before trusting the class, and leave the document
2314 + # untouched if it is genuinely absent. Same reasoning as
2315 + # Metasync_SEO_Conflict_Handler::precedence_available().
2316 + if (!self::precedence_available()) {
2317 + return $html;
2318 + }
2319 +
2320 + # Title — replace the resolved page title into the (single, post-dedup) <title>.
2321 + $resolved_title = Metasync_Seo_Precedence::resolve(
2322 + $post_id,
2323 + Metasync_Seo_Precedence::FIELD_TITLE
2324 + );
2325 + if (!empty($resolved_title['value'])) {
2326 + $title_tag = '<title>' . htmlspecialchars($resolved_title['value'], ENT_QUOTES, 'UTF-8') . '</title>';
2327 + $count = 0;
2328 + $replaced = preg_replace_callback('/<title[^>]*>.*?<\/title>/is', function ($m) use ($title_tag) {
2329 + return $title_tag;
2330 + }, $html, 1, $count);
2331 + if ($count > 0) {
2332 + $html = $replaced;
2333 + } elseif (stripos($html, '<title') === false) {
2334 + $html = preg_replace_callback('/(<head[^>]*>)/i', function ($m) use ($title_tag) {
2335 + return $m[1] . "\n" . $title_tag;
2336 + }, $html, 1);
2337 + }
2338 + }
2339 +
2340 + # Description — replace the resolved page description into the SEO
2341 + # <meta name="description"> only. OG/Twitter description is social
2342 + # precedence and remains outside this setting's scope.
2343 + $resolved_description = Metasync_Seo_Precedence::resolve(
2344 + $post_id,
2345 + Metasync_Seo_Precedence::FIELD_DESCRIPTION
2346 + );
2347 + if (!empty($resolved_description['value'])) {
2348 + $esc = htmlspecialchars($resolved_description['value'], ENT_QUOTES, 'UTF-8');
2349 + $marker = Metasync_Seo_Precedence::is_otto_value(
2350 + $post_id,
2351 + Metasync_Seo_Precedence::FIELD_DESCRIPTION,
2352 + $resolved_description
2353 + )
2354 + ? 'otto'
2355 + : 'custom';
2356 + $html = $this->force_custom_meta($html, 'name', 'description', $esc, $marker);
2357 + }
2358 +
2359 + # Keywords — the same final pass as the description above, on the chain
2360 + # the resolver owns, so the global priority setting governs this tag
2361 + # too. OTTO's keywords tag arrives via header_html_insertion carrying
2362 + # a data-otto marker and is spliced in after wp_head, so it is always
2363 + # the later tag and always wins deduplicate_keywords_tags(); forcing
2364 + # the resolved winner here is what keeps the losing tier reachable at
2365 + # all — the single-printing-owner rule the description branch follows.
2366 + #
2367 + # The tier this forces is NOT guaranteed customer-typed, unlike the
2368 + # title and description keys above: the meta_keywords persistence flag
2369 + # writes OTTO's value straight to _metasync_focus_keyword, and neither
2370 + # that conditional write nor the clone cleaner (which strips only the
2371 + # _metasync_otto_* keys) ever clears it, so a stale flag copy can
2372 + # outrank a newer suggestion under custom-first. That key sharing is
2373 + # tracked separately — the fix is a key of its own for the persisted
2374 + # keyword, not a special case here.
2375 + $resolved_keyword = Metasync_Seo_Precedence::resolve(
2376 + $post_id,
2377 + Metasync_Seo_Precedence::FIELD_FOCUS_KEYWORD
2378 + );
2379 + if (!empty($resolved_keyword['value'])) {
2380 + $esc = htmlspecialchars($resolved_keyword['value'], ENT_QUOTES, 'UTF-8');
2381 + $marker = Metasync_Seo_Precedence::is_otto_value(
2382 + $post_id,
2383 + Metasync_Seo_Precedence::FIELD_FOCUS_KEYWORD,
2384 + $resolved_keyword
2385 + )
2386 + ? 'otto'
2387 + : 'custom';
2388 + $html = $this->force_custom_meta($html, 'name', 'keywords', $esc, $marker);
2389 + }
2390 +
2391 + return $html;
2392 + }
2393 +
1452 2394 # this function does the header replacements
1453 2395 function do_header_replacements($replacement_data){
1454 2396
1455 2397 # check that we have header replacements
@@ -1499,21 +2441,19 @@
1499 2441
1500 2442 # handle canonical links
1501 2443 if($data['type'] == 'link' && $data['rel'] === 'canonical'){
1502 2444
1503 - # Protect manually-set canonical from OTTO override
2445 + # Protect manually-set canonical from OTTO override.
2446 + # Validated: a legacy row corrupted to "Array" must not
2447 + # count as a manual canonical and block OTTO's correct value.
1504 2448 if (function_exists('is_singular') && is_singular()) {
1505 2449 $post_id = get_the_ID();
1506 2450 if ($post_id) {
1507 - $custom_canonical = get_post_meta($post_id, '_metasync_canonical_url', true);
1508 - if (empty($custom_canonical)) {
1509 - $custom_canonical = get_post_meta($post_id, 'meta_canonical', true);
1510 - // Handle legacy array values
1511 - if (is_array($custom_canonical)) {
1512 - $custom_canonical = reset($custom_canonical) ?: '';
1513 - }
2451 + $custom_canonical = Metasync_Canonical_Sanitizer::sanitize(get_post_meta($post_id, '_metasync_canonical_url', true));
2452 + if ($custom_canonical === '') {
2453 + $custom_canonical = Metasync_Canonical_Sanitizer::sanitize(get_post_meta($post_id, 'meta_canonical', true));
1514 2454 }
1515 - if (!empty($custom_canonical)) {
2455 + if ($custom_canonical !== '') {
1516 2456 # Manual canonical takes priority — skip OTTO override
1517 2457 continue;
1518 2458 }
1519 2459 }
@@ -1521,11 +2461,15 @@
1521 2461
1522 2462 # find the cannonical dom element
1523 2463 $link = $this->dom->find('link[rel="canonical"]', 0);
1524 2464
1525 - # set the link property if not empty
2465 + # set the link property if not empty — only to a validated URL
2466 + # never inject a corrupted/non-URL suggestion.
1526 2467 if(!empty($link->href)){
1527 - $link->href = $data['recommended_value'] ?? $link->href;
2468 + $otto_canonical = Metasync_Canonical_Sanitizer::sanitize($data['recommended_value'] ?? '');
2469 + if ($otto_canonical !== '') {
2470 + $link->href = $otto_canonical;
2471 + }
1528 2472 }
1529 2473
1530 2474 #
1531 2475 continue;
@@ -1848,8 +2792,221 @@
1848 2792 return $html;
1849 2793 }
1850 2794
1851 2795 /**
2796 + * Read one of the meta box's social title/description keys with the "Auto Draft"
2797 + * placeholder collapsed to ''.
2798 + *
2799 + * Delegates to Metasync_OpenGraph so the placeholder definition lives in one
2800 + * place. Falls back to a raw read when that class — or that method on it — is
2801 + * unavailable: this buffer filter runs on the front end, and a partially updated
2802 + * install can pair a newer otto/ file with an older includes/ one, where calling
2803 + * a method the loaded class doesn't define would fatal the page rather than
2804 + * degrade.
2805 + *
2806 + * @param int $post_id
2807 + * @param string $key
2808 + * @return string
2809 + */
2810 + private function social_meta($post_id, $key) {
2811 + // PHPStan resolves the literal class-string against the current source and so
2812 + // sees the method as always present; the check is deliberate runtime
2813 + // version-skew defence. Suppressed the same way otto_pixel.php does.
2814 + // @phpstan-ignore-next-line function.alreadyNarrowedType
2815 + if (method_exists('Metasync_OpenGraph', 'get_social_meta')) {
2816 + return Metasync_OpenGraph::get_social_meta($post_id, $key);
2817 + }
2818 + return (string) get_post_meta($post_id, $key, true);
2819 + }
2820 +
2821 + /**
2822 + * Apply the per-post "Social Media & Open Graph" meta box precedence over OTTO.
2823 + *
2824 + * Precedence per tag: an explicit meta box value set by the user wins over OTTO;
2825 + * a field left blank falls back to OTTO's value; if OTTO also has none, a sensible
2826 + * default fills the gap. Mirrors how the custom SEO sidebar values take priority
2827 + * over OTTO (see do_header_replacements()).
2828 + *
2829 + * Runs *after* deduplicate_og_twitter_tags(), and only within <head>. For a field the
2830 + * user explicitly set it removes any existing tag (OTTO's) and re-adds the user's
2831 + * value; for the rest it only adds a tag when absent — so OTTO stays authoritative
2832 + * where the user left a field blank and no duplicate is produced. OTTO never emits an
2833 + * og:image, so this also surfaces a user-set social image on an OTTO-managed page.
2834 + *
2835 + * Only applies to singular posts/pages: the meta box values are per-post, and on
2836 + * archives/home get_queried_object_id() can return a term or user id.
2837 + *
2838 + * @param string $html Full HTML document (post-dedup).
2839 + * @return string
2840 + */
2841 + private function apply_metabox_og_precedence($html) {
2842 + if (!function_exists('get_queried_object_id') || !function_exists('is_singular')) {
2843 + return $html;
2844 + }
2845 + # Per-post OG only applies to singular posts/pages. On archives/home the queried
2846 + # object id can be a term or user id, which must not be read as post meta.
2847 + if (!is_singular()) {
2848 + return $html;
2849 + }
2850 + $post_id = get_queried_object_id();
2851 + if (!$post_id) {
2852 + return $html;
2853 + }
2854 +
2855 + # Respect the meta box opt-out (only an explicit '0' disables it)
2856 + if (get_post_meta($post_id, '_metasync_og_enabled', true) === '0') {
2857 + return $html;
2858 + }
2859 +
2860 + # Explicit (user-typed) meta box values. A non-empty value means the user set
2861 + # this field themselves and it must take priority over OTTO.
2862 + #
2863 + # The four social title/description keys are read through
2864 + # Metasync_OpenGraph::get_social_meta(), which collapses the "Auto Draft"
2865 + # placeholder to ''. Without that, a row polluted by the meta box pre-fill on a
2866 + # brand-new post reads as a deliberate override here (it differs from the real
2867 + # title) and would beat OTTO's correct og:title.
2868 + $og_title_set = $this->social_meta($post_id, '_metasync_og_title');
2869 + $og_desc_set = $this->social_meta($post_id, '_metasync_og_description');
2870 + $og_image_set = get_post_meta($post_id, '_metasync_og_image', true);
2871 + $og_type_set = get_post_meta($post_id, '_metasync_og_type', true);
2872 + $tw_card_set = get_post_meta($post_id, '_metasync_twitter_card', true);
2873 + $tw_site_set = get_post_meta($post_id, '_metasync_twitter_site', true);
2874 + $tw_title_set = $this->social_meta($post_id, '_metasync_twitter_title');
2875 + $tw_desc_set = $this->social_meta($post_id, '_metasync_twitter_description');
2876 + $tw_image_set = get_post_meta($post_id, '_metasync_twitter_image', true);
2877 + $tw_image_alt_set = get_post_meta($post_id, '_metasync_twitter_image_alt', true);
2878 +
2879 + # Default OG values the meta box pre-fills (post title / generated excerpt /
2880 + # featured image). The meta box PERSISTS these defaults on save, so a non-empty
2881 + # _metasync_og_* value alone does not prove the user customized it — a field is a
2882 + # genuine override only when its stored value differs from this default. Reuse the
2883 + # emitter's own resolver so the defaults match exactly (falling back to empty
2884 + # defaults, i.e. treat nothing as customized, if the instance is unavailable).
2885 + $defaults = ['title' => '', 'description' => '', 'image' => ''];
2886 + if (class_exists('Metasync_OpenGraph') && Metasync_OpenGraph::get_instance()) {
2887 + $defaults = Metasync_OpenGraph::get_instance()->get_default_og_values($post_id);
2888 + }
2889 +
2890 + $title_is_custom = ($og_title_set !== '' && $og_title_set !== $defaults['title']);
2891 + $desc_is_custom = ($og_desc_set !== '' && $og_desc_set !== $defaults['description']);
2892 + $image_is_custom = ($og_image_set !== '' && $og_image_set !== $defaults['image']);
2893 +
2894 + # Resolved values (explicit value first, then the meta box default).
2895 + $og_title = $og_title_set !== '' ? $og_title_set : $defaults['title'];
2896 + $og_desc = $og_desc_set !== '' ? $og_desc_set : $defaults['description'];
2897 + $og_image = $og_image_set !== '' ? $og_image_set : $defaults['image'];
2898 + $og_type = $og_type_set ?: 'article';
2899 + $site_name = get_bloginfo('name');
2900 +
2901 + # Twitter inherits the OG value when its own field is blank; a user-customized
2902 + # og:image therefore also overrides OTTO's twitter:image (the meta box treats them
2903 + # as one). A twitter field is custom if its own value differs from the default OR
2904 + # the OG field it inherits was customized.
2905 + $tw_card = $tw_card_set ?: 'summary_large_image';
2906 + $tw_title = $tw_title_set !== '' ? $tw_title_set : $og_title;
2907 + $tw_desc = $tw_desc_set !== '' ? $tw_desc_set : $og_desc;
2908 + $tw_image = $tw_image_set !== '' ? $tw_image_set : $og_image;
2909 +
2910 + $tw_title_is_custom = ($tw_title_set !== '' && $tw_title_set !== $defaults['title']) || $title_is_custom;
2911 + $tw_desc_is_custom = ($tw_desc_set !== '' && $tw_desc_set !== $defaults['description']) || $desc_is_custom;
2912 + $tw_image_is_custom = ($tw_image_set !== '' && $tw_image_set !== $defaults['image']) || $image_is_custom;
2913 +
2914 + # twitter:site: per-post value, else the site-wide handle from Social Meta settings.
2915 + # twitter:site / twitter:image:alt are not auto-defaulted by the meta box, so a
2916 + # non-empty per-post value is itself the override signal.
2917 + $tw_site = $tw_site_set;
2918 + if ($tw_site === '' && class_exists('Metasync')) {
2919 + $social = Metasync::get_option('social_meta');
2920 + if (is_array($social) && !empty($social['twitter_username'])) {
2921 + $tw_site = '@' . ltrim($social['twitter_username'], '@');
2922 + }
2923 + }
2924 +
2925 + # og:type / twitter:card are dropdowns that always carry a value, so they override
2926 + # OTTO only when the user picked a non-default option. og:url is auto-populated with
2927 + # the permalink on save, so it is fill-only (never an override) to avoid clobbering
2928 + # OTTO's canonical URL with a stale/auto value. og:site_name is a site default.
2929 + $og_type_override = ($og_type_set !== '' && $og_type_set !== 'article');
2930 + $tw_card_override = ($tw_card_set !== '' && $tw_card_set !== 'summary_large_image');
2931 +
2932 + # [attr, property/name, value, is_url, is_override]
2933 + $tags = [
2934 + ['property', 'og:title', $og_title, false, $title_is_custom],
2935 + ['property', 'og:description', $og_desc, false, $desc_is_custom],
2936 + ['property', 'og:image', $og_image, true, $image_is_custom],
2937 + ['property', 'og:url', get_permalink($post_id), true, false],
2938 + ['property', 'og:type', $og_type, false, $og_type_override],
2939 + ['property', 'og:site_name', $site_name, false, false],
2940 + ['name', 'twitter:card', $tw_card, false, $tw_card_override],
2941 + ['name', 'twitter:site', $tw_site, false, $tw_site_set !== ''],
2942 + ['name', 'twitter:title', $tw_title, false, $tw_title_is_custom],
2943 + ['name', 'twitter:description', $tw_desc, false, $tw_desc_is_custom],
2944 + ['name', 'twitter:image', $tw_image, true, $tw_image_is_custom],
2945 + ['name', 'twitter:image:alt', $tw_image_alt_set, false, $tw_image_alt_set !== ''],
2946 + ];
2947 +
2948 + # Operate only within <head> so meta tags in the body are never touched.
2949 + if (!preg_match('/(<head\b[^>]*>)(.*?)(<\/head>)/is', $html, $hm)) {
2950 + return $html;
2951 + }
2952 + $head = $hm[2];
2953 +
2954 + # 1) Remove existing head tags for user-overridden properties so the user's value wins.
2955 + foreach ($tags as $t) {
2956 + list($attr, $key, $val, $is_url, $is_override) = $t;
2957 + if ($is_override && $val !== '' && $val !== null) {
2958 + $stripped = preg_replace(
2959 + '/<meta\s[^>]*' . $attr . '\s*=\s*(["\'])' . preg_quote($key, '/') . '\1[^>]*>\s*/i',
2960 + '',
2961 + $head
2962 + );
2963 + if ($stripped !== null) {
2964 + $head = $stripped;
2965 + }
2966 + }
2967 + }
2968 +
2969 + # 2) Build additions: overrides (removed above) + gap-fills for anything still absent.
2970 + #
2971 + # When a third-party SEO plugin is active the conflict handler may have suppressed
2972 + # that plugin's og:title / twitter:title so MetaSync can supply the replacement
2973 + # (Metasync_SEO_Sidebar::output_seo_meta_description). Gap-filling those two tags
2974 + # here would inject the meta-box post-title default over that replacement, so skip
2975 + # them in that case. User-customized overrides are unaffected — they were already
2976 + # applied above.
2977 + $seo_plugin_active = class_exists('Metasync_SEO_Conflict_Handler')
2978 + && Metasync_SEO_Conflict_Handler::get_instance()->has_active_seo_plugin();
2979 +
2980 + $additions = '';
2981 + foreach ($tags as $t) {
2982 + list($attr, $key, $val, $is_url, $is_override) = $t;
2983 + if ($val === '' || $val === null) {
2984 + continue;
2985 + }
2986 + if (!$is_override && $seo_plugin_active && ($key === 'og:title' || $key === 'twitter:title')) {
2987 + continue; # never gap-fill the post title over a third-party plugin's own title
2988 + }
2989 + if (!$is_override && preg_match('/<meta\s[^>]*' . $attr . '\s*=\s*(["\'])' . preg_quote($key, '/') . '\1/i', $head)) {
2990 + continue; # OTTO (or another source) already provides it and the user did not override
2991 + }
2992 + $content = $is_url ? esc_url($val) : esc_attr($val);
2993 + if ($content === '') {
2994 + continue;
2995 + }
2996 + $marker = $is_override ? 'override' : 'fill';
2997 + $additions .= '<meta ' . $attr . '="' . $key . '" content="' . $content . '" data-metasync-og="' . $marker . '">' . "\n";
2998 + }
2999 +
3000 + # Reassemble the head block (str_replace avoids backreference issues from $ in values).
3001 + if ($head === $hm[2] && $additions === '') {
3002 + return $html; # nothing removed, nothing added
3003 + }
3004 + $new_head_block = $hm[1] . $head . $additions . $hm[3];
3005 + return str_replace($hm[0], $new_head_block, $html);
3006 + }
3007 +
3008 + /**
1852 3009 * Deduplicate meta tags by a specific attribute (property= or name=).
1853 3010 *
1854 3011 * When duplicates exist and one carries a data-otto marker, keep only
1855 3012 * the OTTO version. Otherwise keep the first occurrence.
@@ -1869,21 +3026,176 @@
1869 3026 }
1870 3027
1871 3028 $all_tags = $matches[0];
1872 3029
1873 - # Find the OTTO tag (has data-otto-pixel or data-otto attribute)
3030 + # Choose the keeper by precedence: custom sidebar → OTTO → first.
3031 + # A user-set SEO sidebar value (data-metasync-seo="custom") must win over
3032 + # OTTO, matching deduplicate_description_tags() and the sidebar's documented
3033 + # "custom always wins over OTTO" intent. Without this, the generic keeper
3034 + # (OTTO-or-first) drops the sidebar's og:/twitter: description in favor of
3035 + # OTTO's injected tag.
3036 + $custom_tag = null;
1874 3037 $otto_tag = null;
1875 3038 foreach ($all_tags as $tag) {
1876 - if (stripos($tag, 'data-otto') !== false) {
3039 + if ($custom_tag === null && stripos($tag, 'data-metasync-seo') !== false) {
3040 + $custom_tag = $tag;
3041 + }
3042 + if ($otto_tag === null && stripos($tag, 'data-otto') !== false) {
1877 3043 $otto_tag = $tag;
3044 + }
3045 + }
3046 +
3047 + # Determine the keeper: custom sidebar tag, else OTTO tag, else the first tag
3048 + $keeper = $custom_tag ?: ($otto_tag ?: $all_tags[0]);
3049 +
3050 + # Remove all occurrences, then re-insert the keeper at the first position
3051 + $first_replaced = false;
3052 + $html = preg_replace_callback($pattern, function ($m) use ($keeper, &$first_replaced) {
3053 + if (!$first_replaced) {
3054 + $first_replaced = true;
3055 + return $keeper;
3056 + }
3057 + return ''; # Remove subsequent duplicates
3058 + }, $html);
3059 +
3060 + return $html;
3061 + }
3062 +
3063 + /**
3064 + * Deduplicate <meta name="description"> tags after OTTO processing.
3065 + *
3066 + * Unlike title, og, twitter, and canonical tags, the plain name="description"
3067 + * tag had NO dedup pass, so a page could end up with 2-3 copies when several
3068 + * subsystems each emit one:
3069 + * - OTTO backend payload (header_html_insertion) → data-otto-pixel="dynamic-seo"
3070 + * (spliced in additively before </head>, never replacing an existing tag)
3071 + * - MetaSync's persisted-meta wp_head hook → data-metasync-otto="true"
3072 + * - MetaSync SEO sidebar custom value → data-metasync-seo="custom"
3073 + *
3074 + * This runs at the buffer level — after every source has written its tag —
3075 + * and keeps exactly ONE, by precedence (matching the SEO sidebar's documented
3076 + * "custom always wins over OTTO" intent):
3077 + * 1. custom sidebar (data-metasync-seo)
3078 + * 2. OTTO (data-otto-pixel OR data-metasync-otto OR data-otto)
3079 + * 3. first occurrence
3080 + *
3081 + * Note: the generic deduplicate_meta_by_attr() keeper-detection only matches
3082 + * substring "data-otto", which misses "data-metasync-otto"; this method checks
3083 + * both OTTO markers explicitly, so either OTTO source is recognized.
3084 + *
3085 + * @param string $html Full HTML document.
3086 + * @return string HTML with at most one <meta name="description">.
3087 + */
3088 + private function deduplicate_description_tags($html) {
3089 + if (!is_string($html) || $html === '') {
3090 + return $html;
3091 + }
3092 +
3093 + # Match <meta ... name="description" ...> in either attribute order.
3094 + # [^>]* is bounded to a single tag; name="twitter:description" is NOT matched
3095 + # because the opening quote must be immediately followed by "description".
3096 + $pattern = '/<meta\s[^>]*name\s*=\s*["\']description["\'][^>]*\/?>/i';
3097 +
3098 + if (preg_match_all($pattern, $html, $matches) <= 1) {
3099 + return $html; # 0 or 1 — nothing to deduplicate
3100 + }
3101 +
3102 + $all_tags = $matches[0];
3103 +
3104 + # Choose the keeper by precedence: custom sidebar → OTTO → first.
3105 + $custom_tag = null;
3106 + $otto_tag = null;
3107 + foreach ($all_tags as $tag) {
3108 + if ($custom_tag === null && stripos($tag, 'data-metasync-seo') !== false) {
3109 + $custom_tag = $tag;
3110 + }
3111 + if ($otto_tag === null && (
3112 + stripos($tag, 'data-otto-pixel') !== false ||
3113 + stripos($tag, 'data-metasync-otto') !== false ||
3114 + stripos($tag, 'data-otto') !== false
3115 + )) {
3116 + $otto_tag = $tag;
3117 + }
3118 + }
3119 + $keeper = $custom_tag ?: ($otto_tag ?: $all_tags[0]);
3120 +
3121 + # Remove all occurrences, re-inserting the keeper at the first position.
3122 + # Callback form avoids backreference injection when the description content
3123 + # contains $ followed by digits (e.g. "$50 off").
3124 + $first_replaced = false;
3125 + $html = preg_replace_callback($pattern, function ($m) use ($keeper, &$first_replaced) {
3126 + if (!$first_replaced) {
3127 + $first_replaced = true;
3128 + return $keeper;
3129 + }
3130 + return ''; # Remove subsequent duplicates
3131 + }, $html);
3132 +
3133 + return $html;
3134 + }
3135 +
3136 + /**
3137 + * Remove duplicate <meta name="keywords"> tags from HTML.
3138 + *
3139 + * OTTO ships its keywords tag inside header_html_insertion, and MetaSync's
3140 + * own SEO output now emits one too whenever the precedence chain resolves a
3141 + * keyword — so a server-side-rendered page can carry both.
3142 + *
3143 + * The keeper is OTTO's tag, or the first occurrence when no OTTO tag is
3144 + * present. Unlike deduplicate_description_tags() there is no
3145 + * customer-authored branch here: print_metatag() emits no data attribute, so
3146 + * a keywords tag never carries data-metasync-seo="custom" at this point in
3147 + * the pipeline. The customer tier is re-applied afterwards by
3148 + * apply_custom_seo_precedence(), which runs later and stamps the marker via
3149 + * force_custom_meta(). Collapsing to OTTO's tag here is therefore safe: it is
3150 + * not the final word on what the page renders.
3151 + *
3152 + * @param string $html Full HTML document.
3153 + * @return string HTML with at most one <meta name="keywords">.
3154 + */
3155 + private function deduplicate_keywords_tags($html) {
3156 + # No is_string() guard, unlike deduplicate_description_tags() beside it.
3157 + # That asymmetry is forced, not an oversight: @param string makes the
3158 + # check provably always-true at PHPStan level 5, and the description
3159 + # sibling's copy is one of a fixed number of occurrences the baseline
3160 + # pins by count, so adding another fails the build. Both methods are
3161 + # only ever called with the buffer string.
3162 + if ($html === '') {
3163 + return $html;
3164 + }
3165 +
3166 + # Match <meta ... name="keywords" ...> in either attribute order.
3167 + #
3168 + # The optional prefix group must end on whitespace or a quote so the
3169 + # attribute is `name`, not something merely ending in it: a bare
3170 + # [^>]*name= also matches data-name="keywords". Making the group
3171 + # optional keeps the common `<meta name="keywords" ...>` form matching,
3172 + # where <meta\s has already consumed the only separator.
3173 + #
3174 + # The opening quote must be immediately followed by "keywords", so
3175 + # name="news_keywords" is not matched.
3176 + $pattern = '/<meta\s(?:[^>]*[\s"\'])?name\s*=\s*["\']keywords["\'][^>]*\/?>/i';
3177 +
3178 + if (preg_match_all($pattern, $html, $matches) <= 1) {
3179 + return $html; # 0 or 1 — nothing to deduplicate
3180 + }
3181 +
3182 + $all_tags = $matches[0];
3183 +
3184 + $otto_tag = null;
3185 + foreach ($all_tags as $tag) {
3186 + if (stripos($tag, 'data-otto-pixel') !== false ||
3187 + stripos($tag, 'data-metasync-otto') !== false ||
3188 + stripos($tag, 'data-otto') !== false
3189 + ) {
3190 + $otto_tag = $tag;
1878 3191 break;
1879 3192 }
1880 3193 }
1881 -
1882 - # Determine the keeper: OTTO tag if present, otherwise the first tag
1883 3194 $keeper = $otto_tag ?: $all_tags[0];
1884 3195
1885 - # Remove all occurrences, then re-insert the keeper at the first position
3196 + # Callback form avoids backreference injection when the keyword list
3197 + # contains $ followed by digits.
1886 3198 $first_replaced = false;
1887 3199 $html = preg_replace_callback($pattern, function ($m) use ($keeper, &$first_replaced) {
1888 3200 if (!$first_replaced) {
1889 3201 $first_replaced = true;
@@ -1937,96 +3249,301 @@
1937 3249 return $html;
1938 3250 }
1939 3251
1940 3252 /**
3253 + * Normalize a JSON-LD `@type` into a list of lowercased type tokens.
3254 + *
3255 + * JSON-LD allows `@type` to be a string OR an array of strings, and Rank Math
3256 + * routinely emits multi-typed nodes (e.g. ["WebPage", "CollectionPage"]).
3257 + * Collapsing such a node to its first token hides it from a plain-typed
3258 + * duplicate, so every token is returned and indexed separately.
3259 + *
3260 + * @param mixed $raw The raw `@type` value.
3261 + * @return string[] Unique lowercased tokens, possibly empty.
3262 + */
3263 + private function schema_type_tokens($raw) {
3264 + $tokens = [];
3265 +
3266 + foreach ((is_array($raw) ? $raw : [$raw]) as $type) {
3267 + if (is_array($type) || is_object($type) || is_bool($type) || $type === null) {
3268 + continue;
3269 + }
3270 + $type = strtolower(trim((string) $type));
3271 + if ($type === '') {
3272 + continue;
3273 + }
3274 + $tokens[$type] = true;
3275 + }
3276 +
3277 + return array_keys($tokens);
3278 + }
3279 +
3280 + /**
1941 3281 * Deduplicate JSON-LD schema blocks.
1942 3282 *
1943 - * When OTTO and a third-party SEO plugin both inject <script type="application/ld+json">
1944 - * blocks, keep OTTO's version for any @type that appears in both.
1945 - * Third-party blocks whose @type is not covered by OTTO are preserved.
3283 + * Two nodes collide when they share an `@id`, or when one claims an `@type`
3284 + * token the page already carries. Collisions resolve strictly in OTTO's
3285 + * favour:
1946 3286 *
3287 + * - an OTTO node (marked `data-otto`) displaces every colliding
3288 + * third-party node, whichever appeared first in the document;
3289 + * - two OTTO nodes describing the same entity keep the first;
3290 + * - two third-party nodes never evict each other, even when they share a
3291 + * type: a Yoast graph legitimately carries two distinct ImageObjects,
3292 + * and MetaSync's own breadcrumb / local-business blocks are third-party
3293 + * output this pass has no mandate to fold.
3294 + *
3295 + * When OTTO contributed no node at all the document comes back untouched —
3296 + * a third-party-only page is not ours to rewrite. Edits are offset splices
3297 + * into the original document, so a block that loses nothing keeps its exact
3298 + * bytes (attributes, `class`, `@context`, position — head, body or footer)
3299 + * and nothing is ever relocated.
3300 + *
3301 + * Blocks are left exactly where they are when their JSON does not parse,
3302 + * when they carry neither `@type` nor `@graph`, or when they sit inside an
3303 + * HTML comment or <noscript> — a schema node the site disabled by wrapping
3304 + * it in a comment must stay disabled, not be re-published by this pass.
3305 + *
1947 3306 * @param string $html Full HTML.
1948 3307 * @return string
1949 3308 */
1950 3309 private function deduplicate_schema_tags($html) {
1951 - // Find all JSON-LD script blocks
1952 - $pattern = '/<script(\s[^>]*)type\s*=\s*(["\'])application\/ld\+json\2[^>]*>\s*([\s\S]*?)<\/script>/i';
1953 - if (preg_match_all($pattern, $html, $matches, PREG_SET_ORDER) <= 1) {
3310 + // Find all JSON-LD script blocks. Group 1 is the FULL attribute string —
3311 + // capturing only the part before `type=` would hide a `data-otto` marker
3312 + // that sits after it, which is where every insertion path puts it. The
3313 + // type VALUE is prefix-matched so `application/ld+json; charset=UTF-8`
3314 + // is recognised too.
3315 + $pattern = '/<script(\s[^>]*type\s*=\s*(["\'])application\/ld\+json[^"\'>]*\2[^>]*)>\s*([\s\S]*?)<\/script>/i';
3316 + if (!preg_match_all($pattern, $html, $matches, PREG_SET_ORDER | PREG_OFFSET_CAPTURE)) {
1954 3317 return $html;
1955 3318 }
1956 3319
1957 - $otto_by_type = []; // @type => decoded JSON object
1958 - $third_by_type = []; // @type => decoded JSON object
1959 - $otto_graph = []; // entries from OTTO @graph blocks
1960 - $third_graph = []; // entries from third-party @graph blocks
3320 + // Byte ranges this pass must not reach into. A commented-out (or
3321 + // <noscript>-wrapped) schema block is deliberately disabled output.
3322 + $masked = [];
3323 + if (preg_match_all('/<!--[\s\S]*?-->|<noscript\b[^>]*>[\s\S]*?<\/noscript>/i', $html, $comments, PREG_SET_ORDER | PREG_OFFSET_CAPTURE)) {
3324 + foreach ($comments as $comment) {
3325 + $masked[] = [$comment[0][1], $comment[0][1] + strlen($comment[0][0])];
3326 + }
3327 + }
1961 3328
3329 + // `data-otto` must be a whole attribute: `data-otto="true"` or a bare
3330 + // `data-otto`. A substring test would also catch a hypothetical
3331 + // `data-otto-thing="x"` and misattribute that block to OTTO.
3332 + $is_otto_attr = '/(^|\s)data-otto(\s*=|\s|$)/i';
3333 +
3334 + // Blocks able to take part, and their entities in document order. An
3335 + // entity is one addressable node: a flat block's document, one
3336 + // `@graph` entry, or one node of a top-level array.
3337 + $blocks = [];
3338 + $entities = [];
3339 +
1962 3340 foreach ($matches as $m) {
1963 - $attrs = $m[1];
1964 - $json_str = $m[3];
3341 + $start = $m[0][1];
3342 + foreach ($masked as $range) {
3343 + if ($start >= $range[0] && $start < $range[1]) {
3344 + continue 2;
3345 + }
3346 + }
3347 +
3348 + $attrs = $m[1][0];
3349 + $json_str = $m[3][0];
1965 3350 $decoded = json_decode($json_str, true);
1966 3351 if (json_last_error() !== JSON_ERROR_NONE || !is_array($decoded)) {
1967 - continue; // skip unparseable blocks — leave them in place
3352 + continue; // unparseable — leave the block in place
1968 3353 }
1969 - $is_otto = stripos($attrs, 'data-otto') !== false;
1970 3354
3355 + $is_otto = preg_match($is_otto_attr, $attrs) === 1;
3356 + $end = $start + strlen($m[0][0]);
3357 +
3358 + // A trailing newline swallowed on full removal keeps the output tidy.
3359 + $end_ws = $end;
3360 + if (substr($html, $end, 2) === "\r\n") {
3361 + $end_ws = $end + 2;
3362 + } elseif (substr($html, $end, 1) === "\n") {
3363 + $end_ws = $end + 1;
3364 + }
3365 +
3366 + $register = function (array $node) use (&$entities, $is_otto) {
3367 + // Keyed like metasync_otto_structured_entity_key(): @id first,
3368 + // @type tokens for the type fold. Entries with neither stay in
3369 + // their block but cannot collide, so they keep entity === null.
3370 + $types = isset($node['@type']) ? $this->schema_type_tokens($node['@type']) : [];
3371 + $id = (!empty($node['@id']) && is_string($node['@id'])) ? '@id:' . strtolower(trim($node['@id'])) : '';
3372 + if ($types === [] && $id === '') {
3373 + return null;
3374 + }
3375 + $entities[] = ['otto' => $is_otto, 'types' => $types, 'id' => $id, 'alive' => true];
3376 + return count($entities) - 1;
3377 + };
3378 +
3379 + $block = ['start' => $start, 'end' => $end, 'end_ws' => $end_ws, 'attrs' => $attrs, 'decoded' => $decoded, 'items' => []];
3380 +
1971 3381 if (isset($decoded['@graph']) && is_array($decoded['@graph'])) {
1972 - foreach ($decoded['@graph'] as $entry) {
1973 - if (!isset($entry['@type'])) continue;
1974 - // JSON-LD allows @type to be a string OR an array of strings.
1975 - // Rank Math routinely emits multi-typed entries (e.g. ["Person", "Organization"]).
1976 - // Using an array as an offset throws a fatal on PHP 8+, so normalize to scalar.
1977 - $type = is_array($entry['@type'])
1978 - ? (string) reset($entry['@type'])
1979 - : (string) $entry['@type'];
1980 - if ($type === '') continue;
1981 - if ($is_otto) {
1982 - $otto_graph[$type] = $entry;
1983 - } else {
1984 - $third_graph[$type] = $entry;
1985 - }
3382 + foreach ($decoded['@graph'] as $node) {
3383 + $block['items'][] = ['node' => $node, 'e' => is_array($node) ? $register($node) : null];
1986 3384 }
3385 + $block['shape'] = 'graph';
1987 3386 } elseif (isset($decoded['@type'])) {
1988 - $type = is_array($decoded['@type'])
1989 - ? (string) reset($decoded['@type'])
1990 - : (string) $decoded['@type'];
1991 - if ($type === '') continue;
1992 - if ($is_otto) {
1993 - $otto_by_type[$type] = $decoded;
3387 + $block['shape'] = 'flat';
3388 + $block['items'][] = ['node' => $decoded, 'e' => $register($decoded)];
3389 + } elseif (isset($decoded[0])) {
3390 + // A bare top-level array: [{"@type":"WebPage",...}, ...]. The
3391 + // persistence path already folds this shape; the render path
3392 + // used to skip it entirely, leaving its duplicates standing.
3393 + $block['shape'] = 'list';
3394 + foreach ($decoded as $node) {
3395 + $block['items'][] = ['node' => $node, 'e' => is_array($node) ? $register($node) : null];
3396 + }
3397 + } else {
3398 + continue; // neither @type nor @graph nor array — nothing to key on
3399 + }
3400 +
3401 + $blocks[] = $block;
3402 + }
3403 +
3404 + foreach ($entities as $entity) {
3405 + if ($entity['otto']) {
3406 + $otto_present = true;
3407 + break;
3408 + }
3409 + }
3410 + if (empty($otto_present)) {
3411 + return $html; // OTTO contributed no schema — a third-party page is not ours to rewrite
3412 + }
3413 +
3414 + // Fold. Type claims map a token to EVERY alive holder, so a later OTTO
3415 + // node can displace a whole set of third-party duplicates at once.
3416 + /** @var array<string, int> $by_id */
3417 + $by_id = []; // '@id:...' => entity index
3418 + /** @var array<string, int[]> $by_type */
3419 + $by_type = []; // type token => entity index[]
3420 +
3421 + $release = function ($i) use (&$by_id, &$by_type, &$entities) {
3422 + foreach ($entities[$i]['types'] as $token) {
3423 + $holders = $by_type[$token] ?? [];
3424 + $holders = array_values(array_diff($holders, [$i]));
3425 + if ($holders === []) {
3426 + unset($by_type[$token]);
1994 3427 } else {
1995 - $third_by_type[$type] = $decoded;
3428 + $by_type[$token] = $holders;
1996 3429 }
1997 3430 }
3431 + $id = $entities[$i]['id'];
3432 + if ($id !== '' && ($by_id[$id] ?? -1) === $i) {
3433 + unset($by_id[$id]);
3434 + }
3435 + };
3436 +
3437 + foreach ($entities as $i => $entity) {
3438 + $alive = true;
3439 +
3440 + // The same real-world entity twice, whatever @type each copy carries.
3441 + if ($entity['id'] !== '' && isset($by_id[$entity['id']])) {
3442 + $occupant = $by_id[$entity['id']];
3443 + if ($entity['otto'] && !$entities[$occupant]['otto']) {
3444 + $entities[$occupant]['alive'] = false;
3445 + $release($occupant);
3446 + unset($by_id[$entity['id']]); // re-claimed below if the entity survives
3447 + } else {
3448 + $alive = false; // first copy stands (OTTO's, or the earlier OTTO's)
3449 + }
3450 + }
3451 +
3452 + if ($alive) {
3453 + foreach ($entity['types'] as $token) {
3454 + $holders = $by_type[$token] ?? [];
3455 + if ($holders === []) {
3456 + continue;
3457 + }
3458 + $otto_holder = false;
3459 + foreach ($holders as $holder) {
3460 + if ($entities[$holder]['otto']) {
3461 + $otto_holder = true;
3462 + break;
3463 + }
3464 + }
3465 + if ($otto_holder) {
3466 + $alive = false; // an OTTO node already holds this token
3467 + break;
3468 + }
3469 + if ($entity['otto']) {
3470 + // OTTO displaces every third-party holder of the token and
3471 + // releases ALL their claims, not just this one — otherwise
3472 + // tokens the evicted node held alone would stay blocked and
3473 + // later, legitimate nodes of that type would lose.
3474 + foreach ($holders as $holder) {
3475 + $entities[$holder]['alive'] = false;
3476 + $release($holder);
3477 + }
3478 + }
3479 + // Third party meeting third-party holders: both stand.
3480 + }
3481 + }
3482 +
3483 + if ($alive) {
3484 + foreach ($entity['types'] as $token) {
3485 + $by_type[$token][] = $i;
3486 + }
3487 + if ($entity['id'] !== '') {
3488 + $by_id[$entity['id']] = $i;
3489 + }
3490 + } else {
3491 + $entities[$i]['alive'] = false;
3492 + }
1998 3493 }
1999 3494
2000 - // If OTTO provided no schema at all, nothing to deduplicate
2001 - if (empty($otto_by_type) && empty($otto_graph)) {
2002 - return $html;
3495 + $any_evicted = false;
3496 + foreach ($entities as $entity) {
3497 + if (!$entity['alive']) {
3498 + $any_evicted = true;
3499 + break;
3500 + }
2003 3501 }
3502 + if (!$any_evicted) {
3503 + return $html; // no collision — every original byte stays
3504 + }
2004 3505
2005 - // Remove all JSON-LD blocks from HTML
2006 - $html = preg_replace($pattern, '', $html);
3506 + // Rebuild only the blocks that lost something, back-to-front so the
3507 + // recorded offsets of earlier blocks stay valid.
3508 + $edits = [];
3509 + foreach ($blocks as $block) {
3510 + $kept = [];
3511 + $lost = 0;
3512 + foreach ($block['items'] as $item) {
3513 + if ($item['e'] === null || $entities[$item['e']]['alive']) {
3514 + $kept[] = $item['node'];
3515 + } else {
3516 + $lost++;
3517 + }
3518 + }
3519 + if ($lost === 0) {
3520 + continue;
3521 + }
3522 + if ($kept === []) {
3523 + $edits[] = [$block['start'], $block['end_ws'], ''];
3524 + continue;
3525 + }
2007 3526
2008 - // Re-insert flat (non-@graph) blocks: OTTO wins for matching @type
2009 - $kept = array_merge($otto_by_type, array_diff_key($third_by_type, $otto_by_type));
2010 - $rebuilt = '';
2011 - foreach ($kept as $decoded) {
2012 - $rebuilt .= '<script type="application/ld+json" data-otto="true">' .
2013 - wp_json_encode($decoded) . "</script>\n";
2014 - }
3527 + if ($block['shape'] === 'list') {
3528 + $payload = $kept;
3529 + } else { // graph — a flat block that lost its only entity is removed above
3530 + $payload = $block['decoded'];
3531 + $payload['@graph'] = $kept;
3532 + }
2015 3533
2016 - // Re-insert merged @graph block (if any entries exist)
2017 - $merged_graph = array_merge($third_graph, $otto_graph); // OTTO wins on duplicate @type
2018 - if (!empty($merged_graph)) {
2019 - $graph_obj = ['@context' => 'https://schema.org', '@graph' => array_values($merged_graph)];
2020 - $rebuilt .= '<script type="application/ld+json" data-otto="true">' .
2021 - wp_json_encode($graph_obj) . "</script>\n";
3534 + $json = metasync_safe_json_ld_encode($payload, JSON_UNESCAPED_SLASHES | JSON_UNESCAPED_UNICODE);
3535 + if (!is_string($json) || $json === '') {
3536 + continue; // never drop a block whose payload failed to re-encode
3537 + }
3538 +
3539 + // The original attribute string is reused verbatim so class, id and
3540 + // the data-otto marker itself survive the rebuild.
3541 + $edits[] = [$block['start'], $block['end'], '<script' . $block['attrs'] . '>' . $json . '</script>'];
2022 3542 }
2023 3543
2024 - // Re-inject before </head>
2025 - if (!empty($rebuilt)) {
2026 - $html = preg_replace_callback('/(<\/head>)/i', function ($m) use ($rebuilt) {
2027 - return $rebuilt . $m[1];
2028 - }, $html, 1);
3544 + foreach (array_reverse($edits) as $edit) {
3545 + $html = substr($html, 0, $edit[0]) . $edit[2] . substr($html, $edit[1]);
2029 3546 }
2030 3547
2031 3548 return $html;
2032 3549 }
@@ -2045,14 +3562,14 @@
2045 3562 return true;
2046 3563 }
2047 3564
2048 3565 # Check if amp=1 query parameter is present
2049 - if (isset($_GET['amp']) && $_GET['amp'] == '1') {
3566 + if (isset($_GET['amp']) && $_GET['amp'] == '1') { // phpcs:ignore WordPress.Security.NonceVerification.Recommended -- read-only AMP detection
2050 3567 return true;
2051 3568 }
2052 3569
2053 3570 # Check for other common AMP query parameters
2054 - if (isset($_GET['amp']) && !empty($_GET['amp'])) {
3571 + if (isset($_GET['amp']) && !empty($_GET['amp'])) { // phpcs:ignore WordPress.Security.NonceVerification.Recommended -- read-only AMP detection
2055 3572 return true;
2056 3573 }
2057 3574
2058 3575 return false;
@@ -2057,8 +3574,32 @@
2057 3574
2058 3575 return false;
2059 3576 }
2060 3577
3578 + # function to mark OTTO's own JSON-LD blocks so deduplicate_schema_tags() can
3579 + # tell them apart from a third-party SEO plugin's blocks. Every insertion path
3580 + # (DOM, string fallback, HTTP fallback) runs the fragment through this so the
3581 + # bytes they insert are identical and the "already applied" checks still match.
3582 + private function stamp_otto_json_ld($html){
3583 +
3584 + $stamped = preg_replace_callback(
3585 + '/<script(\s[^>]*type\s*=\s*(["\'])application\/ld\+json[^"\'>]*\2[^>]*)>/i',
3586 + function ($m) {
3587 + # idempotent: a block already carrying a data-otto attribute —
3588 + # anywhere in the tag, not just before `type` — is left alone,
3589 + # never given a second one
3590 + if (preg_match('/(^|\s)data-otto(\s|=|$)/i', $m[1])) {
3591 + return $m[0];
3592 + }
3593 + return '<script' . $m[1] . ' data-otto="true">';
3594 + },
3595 + $html
3596 + );
3597 +
3598 + # preg_replace returns null on a PCRE failure — never hand back less than we got
3599 + return is_string($stamped) ? $stamped : $html;
3600 + }
3601 +
2061 3602 # this function insterts header html to the dom
2062 3603 function insert_header_html($data){
2063 3604
2064 3605 # check that we have the header html
@@ -2071,20 +3612,27 @@
2071 3612 $head = $this->dom->find('head', 0);
2072 3613
2073 3614 if ($head) {
2074 3615
2075 - # Check if this is an AMP page - if so, don't add metasync_optimized attribute
3616 + # Check if this is an AMP page - if so, skip the processed marker
2076 3617 $is_amp_page = $this->is_amp_page();
2077 3618
2078 - # Append the new HTML at the start of the <head> tag
2079 - # For AMP pages: use clean <head> tag without metasync_optimized attribute
2080 - # For non-AMP pages: add metasync_optimized attribute to <head> tag
2081 - if ($is_amp_page) {
2082 - $head->outertext = '<head>' .$data['header_html_insertion']. $head->innertext . '</head>';
2083 - } else {
2084 - $head->outertext = '<head metasync_optimized>' .$data['header_html_insertion']. $head->innertext . '</head>';
2085 - }
3619 + # Append the new HTML at the start of the <head> tag. The tag itself
3620 + # stays a literal '<head>' either way, so plugins that inject with a
3621 + # literal str_replace('<head>', ...) still match the real tag —
3622 + # see mark_head_optimized() for what breaks when it does not.
3623 + $marker = $is_amp_page ? '' : self::HEAD_OPTIMIZED_MARKER;
2086 3624
3625 + # Escape JSON-LD payload closers so a `</script>` inside a JSON
3626 + # string value cannot terminate the element early (XSS/page corruption).
3627 + # Then stamp the blocks as OTTO's, exactly as the string and HTTP
3628 + # fallbacks do, so the deduper can attribute them.
3629 + $escaped_header_insertion = $this->stamp_otto_json_ld(
3630 + metasync_escape_json_ld_blocks_in_html($data['header_html_insertion'])
3631 + );
3632 +
3633 + $head->outertext = '<head>' . $marker . $escaped_header_insertion . $head->innertext . '</head>';
3634 +
2087 3635 }
2088 3636
2089 3637 # save and reload DOM
2090 3638 $this->save_reload();
@@ -2201,8 +3749,35 @@
2201 3749
2202 3750 }
2203 3751
2204 3752 /**
3753 + * Explicitly free the SimpleHtmlDom node tree and reclaim memory.
3754 + *
3755 + * SimpleHtmlDom nodes hold circular parent/child references, so simply
3756 + * dropping the document does not free the tree until request shutdown.
3757 + * HtmlNode::clear() breaks those cycles; we then drop the document, force a
3758 + * GC pass, and re-instantiate a fresh empty parser so the instance stays
3759 + * reusable for any subsequent route on the same request.
3760 + */
3761 + function free_dom(){
3762 + if (isset($this->dom)) {
3763 + # Break the node tree's parent/child cycles so GC can reclaim it now.
3764 + if (isset($this->dom->root) && is_object($this->dom->root)
3765 + && method_exists($this->dom->root, 'clear')) {
3766 + $this->dom->root->clear();
3767 + }
3768 + unset($this->dom);
3769 + }
3770 +
3771 + # Fresh, empty parser — mirrors the constructor so the object is reusable.
3772 + $this->dom = new HtmlDocument(null, true, true, 'UTF-8', false);
3773 +
3774 + if (function_exists('gc_collect_cycles')) {
3775 + gc_collect_cycles();
3776 + }
3777 + }
3778 +
3779 + /**
2205 3780 * Process HTML directly without HTTP request (for buffer approach)
2206 3781 * This is the FAST path - eliminates the internal wp_remote_get call
2207 3782 *
2208 3783 * CRITICAL: This method is called from the output buffer callback.
@@ -2224,15 +3799,25 @@
2224 3799 if (stripos($html, '<html') === false && stripos($html, '<!DOCTYPE') === false) {
2225 3800 return false;
2226 3801 }
2227 3802
3803 + # Skip DOM processing on oversized documents to avoid fatal OOM.
3804 + # Returning false makes the buffer callback serve the original HTML
3805 + # unmodified — the page still renders, just without OTTO changes.
3806 + if (class_exists('Metasync_Otto_Render_Strategy')
3807 + && !Metasync_Otto_Render_Strategy::is_document_processable(strlen($html))
3808 + ) {
3809 + Metasync_Otto_Render_Strategy::log_oversized_skip('process_html_directly', strlen($html));
3810 + return false;
3811 + }
3812 +
2228 3813 # Remove XML declaration if present
2229 3814 $html = preg_replace('/<\?xml[^?]*\?>\s*/i', '', $html);
2230 3815
2231 - # WP-355 / WP-315: Save ALL original <style> blocks before DOM processing.
3816 + # / Save ALL original <style> blocks before DOM processing.
2232 3817 $original_style_blocks = $this->capture_style_blocks($html);
2233 3818
2234 - # WP-465: Shield entity-encoded srcdoc values (lazy YouTube/video facades)
3819 + # Shield entity-encoded srcdoc values (lazy YouTube/video facades)
2235 3820 # so SimpleHtmlDom can't decode them and hoist their global <style> into <head>.
2236 3821 $srcdoc_store = array();
2237 3822 $html = $this->protect_srcdoc_attributes($html, $srcdoc_store);
2238 3823
@@ -2238,11 +3823,15 @@
2238 3823
2239 3824 # Escape bare < in text content before DOM parsing
2240 3825 $html = $this->sanitize_text_less_than($html);
2241 3826
2242 - # WP-355: Fix malformed self-closing non-void tags (e.g. <ul/ class="...">)
3827 + # Fix malformed self-closing non-void tags (e.g. <ul/ class="...">)
2243 3828 $html = $this->fix_malformed_self_closing_tags($html);
2244 3829
3830 + # Quote unquoted attribute values (e.g. data-autoplay-viewport=80%) —
3831 + # SimpleHtmlDom swallows every following attribute into such a value.
3832 + $html = $this->quote_unquoted_attribute_values($html);
3833 +
2245 3834 # Load HTML into DOM
2246 3835 $this->dom->load($html, true, false);
2247 3836
2248 3837 # Force UTF-8 charset to preserve emojis and special characters
@@ -2424,15 +4013,16 @@
2424 4013 }
2425 4014
2426 4015 # Apply header HTML insertion (for schema, etc.) — only if DOM insertion didn't already apply it
2427 4016 if (!empty($replacement_data['header_html_insertion'])) {
2428 - $header_html_check = trim($replacement_data['header_html_insertion']);
4017 + # Escape JSON-LD payload closers and stamp OTTO's blocks first; the DOM
4018 + # path inserts the same bytes, so the already-applied check must compare
4019 + # against the stamped form or it would insert a second copy.
4020 + $header_html_insertion = $this->stamp_otto_json_ld(
4021 + metasync_escape_json_ld_blocks_in_html($replacement_data['header_html_insertion'])
4022 + );
4023 + $header_html_check = trim($header_html_insertion);
2429 4024 if (strpos($result_html, $header_html_check) === false) {
2430 - $header_html_insertion = preg_replace(
2431 - '/<script(\s[^>]*)type\s*=\s*(["\'])application\/ld\+json\2/i',
2432 - '<script$1type=$2application/ld+json$2 data-otto="true"',
2433 - $replacement_data['header_html_insertion']
2434 - );
2435 4025 $header_html = str_replace(array('\\', '$'), array('\\\\', '\\$'), $header_html_insertion);
2436 4026 # Insert before </head>
2437 4027 $result_html = preg_replace('/(<\/head>)/i', $header_html . "\n" . '$1', $result_html, 1);
2438 4028 }
@@ -2487,9 +4077,9 @@
2487 4077 }
2488 4078
2489 4079 # CRITICAL FIX: Apply image alt text manually via string replacement
2490 4080 # DOM changes don't persist, must use string replacement
2491 - $result_html = $this->apply_image_alt_text_via_string($result_html, $replacement_data);
4081 + $result_html = $this->otto_guard_html($this->apply_image_alt_text_via_string($result_html, $replacement_data), $result_html, 'apply_image_alt_text_via_string');
2492 4082
2493 4083 # String-based heading fallback for body_substitutions
2494 4084 # DOM changes via SimpleHtmlDom don't persist on Divi/page-builder sites
2495 4085 if (!empty($replacement_data['body_substitutions']['headings']) && is_array($replacement_data['body_substitutions']['headings'])) {
@@ -2504,13 +4094,19 @@
2504 4094
2505 4095 $result_html = preg_replace_callback(
2506 4096 '/(<' . $heading_type . '(?:\s[^>]*)?>)(.*?)(<\/' . $heading_type . '>)/is',
2507 4097 function ($m) use ($current_value, $recommended_value) {
2508 - $inner_text = trim(preg_replace('/\s+/', ' ', html_entity_decode(strip_tags($m[2]), ENT_QUOTES, 'UTF-8')));
2509 - if ($inner_text === $current_value) {
2510 - return $m[1] . $recommended_value . $m[3];
4098 + $inner_text = trim(preg_replace('/\s+/', ' ', html_entity_decode(wp_strip_all_tags($m[2]), ENT_QUOTES, 'UTF-8')));
4099 + if ($inner_text !== $current_value) {
4100 + return $m[0];
2511 4101 }
2512 - return $m[0];
4102 + # Keep the author's inline formatting; skip when it can't
4103 + # be preserved rather than flattening the heading.
4104 + $new_inner = self::build_heading_inner_html($m[2], $recommended_value);
4105 + if ($new_inner === null) {
4106 + return $m[0];
4107 + }
4108 + return $m[1] . $new_inner . $m[3];
2513 4109 },
2514 4110 $result_html,
2515 4111 -1
2516 4112 );
@@ -2552,38 +4148,46 @@
2552 4148 $result_html
2553 4149 );
2554 4150 }
2555 4151
2556 - # DEDUPLICATION: Remove duplicate <title>, OG, Twitter tags, canonical, and JSON-LD schema
2557 - $result_html = $this->deduplicate_title_tags($result_html);
2558 - $result_html = $this->deduplicate_og_twitter_tags($result_html);
2559 - $result_html = $this->deduplicate_schema_tags($result_html);
2560 - $result_html = $this->deduplicate_canonical_tags($result_html);
4152 + # Apply OTTO's canonical via string replacement (DOM edit is clobbered by the insert_header_html head-freeze; see apply_canonical_via_string).
4153 + $result_html = $this->apply_canonical_via_string($result_html, $replacement_data);
2561 4154
4155 + # DEDUPLICATION: Remove duplicate <title>, meta description, OG, Twitter tags, canonical, and JSON-LD schema
4156 + $result_html = $this->otto_guard_html($this->deduplicate_title_tags($result_html), $result_html, 'deduplicate_title_tags');
4157 + $result_html = $this->otto_guard_html($this->deduplicate_description_tags($result_html), $result_html, 'deduplicate_description_tags');
4158 + $result_html = $this->otto_guard_html($this->deduplicate_keywords_tags($result_html), $result_html, 'deduplicate_keywords_tags');
4159 + $result_html = $this->otto_guard_html($this->deduplicate_og_twitter_tags($result_html), $result_html, 'deduplicate_og_twitter_tags');
4160 + $result_html = $this->otto_guard_html($this->apply_metabox_og_precedence($result_html), $result_html, 'apply_metabox_og_precedence');
4161 + $result_html = $this->otto_guard_html($this->apply_custom_seo_precedence($result_html), $result_html, 'apply_custom_seo_precedence');
4162 + $result_html = $this->otto_guard_html($this->deduplicate_schema_tags($result_html), $result_html, 'deduplicate_schema_tags');
4163 + $result_html = $this->otto_guard_html($this->deduplicate_canonical_tags($result_html), $result_html, 'deduplicate_canonical_tags');
4164 +
2562 4165 # MEMORY OPTIMIZED: Free all large objects and arrays before returning
2563 4166 # This ensures memory is released immediately, especially important for high-traffic sites
2564 4167 unset($final_meta_matches, $matches);
2565 4168
2566 - # Clear SimpleHtmlDom internal cache to free memory
2567 - # Note: We don't unset $this->dom as the object may be reused
2568 - if ($this->dom && method_exists($this->dom, 'clear')) {
2569 - $this->dom->clear();
2570 - }
4169 + # Explicitly free the SimpleHtmlDom node tree now that the
4170 + # result has been serialized to a string. The bundled HtmlDocument has
4171 + # no clear() method, so the previous method_exists() guard was a no-op
4172 + # and the (circular-referenced) node tree lingered until object
4173 + # destruction. free_dom() breaks those references and reclaims memory.
4174 + $this->free_dom();
2571 4175
2572 4176 # Clear element cache array
2573 4177 $this->cached_elements = [];
2574 4178
2575 - # Ensure metasync_optimized attribute on <head> (post-serialization so dom->clear() can't wipe it)
2576 - if (!$this->is_amp_page() && strpos($result_html, 'metasync_optimized') === false) {
2577 - $result_html = preg_replace('/<head(\s|>)/i', '<head metasync_optimized$1', $result_html, 1);
4179 + # Mark the <head> as processed (post-serialization so dom->clear() can't wipe it)
4180 + if (!$this->is_amp_page()) {
4181 + $result_html = $this->mark_head_optimized($result_html);
2578 4182 }
2579 4183
2580 - $result_html = $this->restore_case_sensitive_attributes($result_html);
4184 + $result_html = $this->otto_guard_html($this->restore_case_sensitive_attributes($result_html), $result_html, 'restore_case_sensitive_attributes');
2581 4185
2582 - # WP-355 / WP-315: Re-inject any <style> blocks lost during processing.
2583 - $result_html = $this->restore_lost_style_blocks($result_html, $original_style_blocks);
4186 + # / Re-inject any <style> blocks lost during processing.
4187 + $result_html = $this->otto_guard_html($this->restore_lost_style_blocks($result_html, $original_style_blocks), $result_html, 'restore_lost_style_blocks');
2584 4188
2585 - # WP-480: Do NOT renumber Divi classes on the buffer/in-place path.
4189 + # Do NOT renumber Divi classes on the buffer/in-place path.
2586 4190 # fix_divi_class_renumbering() exists to undo the index offset that
2587 4191 # Divi's shortcode framework introduces ONLY during the internal
2588 4192 # wp_remote_get fetch (handle_route_html), where the Theme Builder
2589 4193 # header is counted twice. process_html_directly() processes the
@@ -2593,23 +4197,42 @@
2593 4197 # colliding page modules with header/global modules of the same
2594 4198 # number and collapsing layouts (Divi hero/slider) on Divi sites.
2595 4199 # The renumber stays only on the HTTP-fetch path (handle_route_html).
2596 4200
2597 - # WP-315: Clean OTTO internal fetch params from HTML output
2598 - $result_html = $this->clean_otto_fetch_params($result_html);
4201 + # Clean OTTO internal fetch params from HTML output
4202 + $result_html = $this->otto_guard_html($this->clean_otto_fetch_params($result_html), $result_html, 'clean_otto_fetch_params');
2599 4203
2600 - # WP-465: Remove any lazy-video facade <style> hoisted into the page
4204 + # Remove any lazy-video facade <style> hoisted into the page
2601 4205 # (runs while srcdoc is still tokenized, so the iframe's own copy is safe).
4206 + $result_html = $this->otto_guard_html($this->strip_hoisted_facade_styles($result_html), $result_html, 'strip_hoisted_facade_styles');
4207 +
4208 + # Restore the original encoded srcdoc values (undo tokenization).
4209 + $result_html = $this->otto_guard_html($this->restore_srcdoc_attributes($result_html, $srcdoc_store), $result_html, 'restore_srcdoc_attributes');
4210 +
4211 + # Keep <meta charset> within the first 1024 bytes so external CSS
4212 + # (e.g. checkmark content:"✓") doesn't mojibake on cached responses.
4213 + $result_html = $this->otto_guard_html($this->ensure_early_charset_meta($result_html), $result_html, 'ensure_early_charset_meta');
4214 +
4215 + # undo double-encoded numeric/hex character references produced by the
4216 + # bundled simplehtmldom serializer from attributes like data-x-icon-s="&#xf3c5".
4217 + $result_html = $this->otto_guard_html($this->repair_double_encoded_entities($result_html), $result_html, 'repair_double_encoded_entities');
4218 +
4219 + # restore literal '&' query-string separators the serializer escaped to
4220 + # '&amp;', which silently truncates multi-family Google Fonts URLs.
4221 + $result_html = $this->otto_guard_html($this->repair_query_string_ampersands($result_html), $result_html, 'repair_query_string_ampersands');
4222 +
4223 + # Remove any lazy-video facade <style> hoisted into the page
4224 + # (runs while srcdoc is still tokenized, so the iframe's own copy is safe).
2602 4225 $result_html = $this->strip_hoisted_facade_styles($result_html);
2603 4226
2604 - # WP-465: Restore the original encoded srcdoc values (undo tokenization).
4227 + # Restore the original encoded srcdoc values (undo tokenization).
2605 4228 $result_html = $this->restore_srcdoc_attributes($result_html, $srcdoc_store);
2606 4229
2607 - # WP-465: Keep <meta charset> within the first 1024 bytes so external CSS
4230 + # Keep <meta charset> within the first 1024 bytes so external CSS
2608 4231 # (e.g. checkmark content:"✓") doesn't mojibake on cached responses.
2609 4232 $result_html = $this->ensure_early_charset_meta($result_html);
2610 4233
2611 - # WP-471: undo double-encoded numeric/hex character references produced by the
4234 + # undo double-encoded numeric/hex character references produced by the
2612 4235 # bundled simplehtmldom serializer from attributes like data-x-icon-s="&#xf3c5".
2613 4236 $result_html = $this->repair_double_encoded_entities($result_html);
2614 4237
2615 4238 return $result_html;
@@ -2614,12 +4237,26 @@
2614 4237
2615 4238 return $result_html;
2616 4239
2617 4240 } catch (Exception $e) {
2618 - error_log('MetaSync ' . Metasync::get_whitelabel_otto_name() . ': Exception in process_html_directly - ' . $e->getMessage());
4241 + if (function_exists('metasync_otto_report_render_failure')) {
4242 + metasync_otto_report_render_failure(
4243 + 'MODIFICATION_EXCEPTION',
4244 + 'OTTO rewrite threw: ' . $e->getMessage(),
4245 + ['exception' => get_class($e)],
4246 + 'error'
4247 + );
4248 + }
2619 4249 return false;
2620 4250 } catch (Error $e) {
2621 - error_log('MetaSync ' . Metasync::get_whitelabel_otto_name() . ': Error in process_html_directly - ' . $e->getMessage());
4251 + if (function_exists('metasync_otto_report_render_failure')) {
4252 + metasync_otto_report_render_failure(
4253 + 'MODIFICATION_ERROR',
4254 + 'OTTO rewrite errored: ' . $e->getMessage(),
4255 + ['error' => get_class($e)],
4256 + 'error'
4257 + );
4258 + }
2622 4259 return false;
2623 4260 }
2624 4261 }
2625 4262
@@ -2668,29 +4305,21 @@
2668 4305 }
2669 4306 }
2670 4307 }
2671 4308
2672 - if ($remove_title && !$has_custom_title) {
2673 - # Remove Open Graph and Twitter title tags (keep main <title>)
2674 - $title_selectors = [
2675 - 'meta[property=og:title]',
2676 - 'meta[name=twitter:title]',
2677 - ];
4309 + # Preserve third-party og:title/twitter:title tags here. They used to be
4310 + # stripped whenever $remove_title was set, but that also removes them
4311 + # when no custom title exists to replace them, leaving pages with no
4312 + # og:title/twitter:title at all. OTTO's own title/OG dedup and
4313 + # precedence passes already reconcile duplicates, so this branch no
4314 + # longer needs to delete those tags.
2678 4315
2679 - foreach ($title_selectors as $selector) {
2680 - $tags = $this->dom->find($selector);
2681 - foreach ($tags as $tag) {
2682 - $tag->outertext = ''; # Remove the tag
2683 - }
2684 - }
2685 - }
2686 -
2687 4316 # Save changes
2688 4317 $this->save_reload();
2689 4318 }
2690 4319
2691 4320 /**
2692 - * Clean OTTO internal fetch query parameters from rendered HTML (WP-315).
4321 + * Clean OTTO internal fetch query parameters from rendered HTML.
2693 4322 *
2694 4323 * @param string $html The rendered HTML
2695 4324 * @return string HTML with OTTO fetch params removed
2696 4325 */
@@ -2704,9 +4333,9 @@
2704 4333 '/[?&]amp;otto_block_desc=1/',
2705 4334 ];
2706 4335 $html = preg_replace($patterns, '', $html);
2707 4336
2708 - # WP-398: Protect <script>/<style> bodies before the URL cleanup below.
4337 + # Protect <script>/<style> bodies before the URL cleanup below.
2709 4338 # The href/src/action matcher can otherwise match JS that looks like an
2710 4339 # attribute (e.g. location.href="/x??y", el.src="a??b") and collapse the
2711 4340 # ?? inside it. Swap each block for a placeholder comment, run the cleanup,
2712 4341 # then restore byte-for-byte.
@@ -2720,17 +4349,25 @@
2720 4349 },
2721 4350 $html
2722 4351 );
2723 4352
2724 - # WP-398: Normalize leftover query-string separators (?? / ?& / ?#) created
2725 - # by the param removal above — but ONLY inside URL attributes (href/src/action).
2726 - # The previous implementation ran str_replace('??','?') over the ENTIRE document,
2727 - # which corrupted inline JavaScript nullish operators (?? / ??=) and any literal
2728 - # "??" in page text — breaking Slider Revolution and other modern inline scripts.
4353 + # Normalize leftover query-string separators (?& / ?#) created by the param
4354 + # removal above — but ONLY inside URL attributes (href/src/action).
4355 + #
4356 + # IMPORTANT: never collapse "??". It is valid, load-bearing syntax for
4357 + # file-concatenation endpoints — Jetpack Boost (/_jb_static/??<hash>) and
4358 + # WordPress core (load-styles.php??... / load-scripts.php??...). Collapsing
4359 + # "??" -> "?" points those at a different URL that returns HTTP 400 / empty,
4360 + # so every affected CSS/JS bundle fails to load and the page renders broken
4361 + # (Kadence + Jetpack Boost "defer non-essential CSS" was the reported case).
4362 + # Removing an OTTO fetch param never produces a "??" anyway, so there is
4363 + # nothing to normalize there.
4364 + # (A previous implementation ran str_replace('??','?') over the ENTIRE
4365 + # document, which additionally corrupted inline JavaScript nullish operators.)
2729 4366 $html = preg_replace_callback(
2730 4367 '/\b(href|src|action)(\s*=\s*)(["\'])(.*?)\3/is',
2731 4368 function ($m) {
2732 - $url = str_replace(['??', '?&', '?#'], ['?', '?', '#'], $m[4]);
4369 + $url = str_replace(['?&', '?#'], ['?', '#'], $m[4]);
2733 4370 return $m[1] . $m[2] . $m[3] . $url . $m[3];
2734 4371 },
2735 4372 $html
2736 4373 );
@@ -2743,9 +4380,9 @@
2743 4380 return $html;
2744 4381 }
2745 4382
2746 4383 /**
2747 - * Fix Divi 5 shortcode framework class renumbering (WP-315)
4384 + * Fix Divi 5 shortcode framework class renumbering
2748 4385 *
2749 4386 * When Divi 5's shortcode framework loads during the internal HTTP fetch,
2750 4387 * it uses shared counters across all template parts (header + page + footer),
2751 4388 * causing page content element classes to be offset (et_pb_section_0 becomes
@@ -2785,9 +4422,9 @@
2785 4422 # IMPORTANT: 'blog' and 'portfolio' MUST be included — OTTO's HTTP render
2786 4423 # shifts their numbering (et_pb_blog_0 → et_pb_blog_1) because the internal
2787 4424 # wp_remote_get includes the Theme Builder header template in Divi's counter.
2788 4425 # If the blog module class doesn't match between page 1 (OTTO-processed) and
2789 - # page 2 (AJAX, no OTTO), Divi's pagination JS can't find the container → empty results (WP-315).
4426 + # page 2 (AJAX, no OTTO), Divi's pagination JS can't find the container → empty results.
2790 4427 $types = [
2791 4428 'section', 'row', 'column', 'text', 'blurb', 'toggle',
2792 4429 'button', 'image', 'group_carousel',
2793 4430 'blog', 'portfolio', 'filterable_portfolio', 'shop',
@@ -2802,9 +4439,9 @@
2802 4439 # The width fraction is NOT an instance counter; rewriting it (e.g.
2803 4440 # et_pb_column_4_4 → et_pb_column_0_4) strips the column's width rule from
2804 4441 # Divi's static stylesheet (.et_pb_column_4_4{width:100%}) and collapses the
2805 4442 # layout. Bare instance indices (et_pb_column_16, followed by space/quote)
2806 - # still match and renumber as intended (WP-470 / WP-315).
4443 + # still match and renumber as intended.
2807 4444 $instance_lookahead = '(?!_(?:tb_|\d))';
2808 4445
2809 4446 foreach ($types as $type) {
2810 4447 # Find the FIRST numbered instance in page content (not _tb_ or width-fraction)
@@ -2900,5 +4537,5 @@
2900 4537 return $html;
2901 4538 }
2902 4539
2903 4540
2904 -}
4541 +}