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 +889 -147 2.6.22 → 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
@@ -563,8 +568,52 @@
563 568 return $out === null ? $html : $out;
564 569 }
565 570
566 571 /**
572 + * Mark the <head> as OTTO-processed without altering the <head> tag itself.
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 : $out;
613 + }
614 +
615 + /**
567 616 * Apply OTTO's canonical via string replacement (reliable fallback).
568 617 *
569 618 * Why the DOM path fails: do_header_replacements() sets the canonical with a
570 619 * DOM attribute edit ($link->href = …). That edit works in isolation, but
@@ -736,9 +785,9 @@
736 785 $response = wp_remote_get($url_with_params, $args);
737 786
738 787 # Check for errors
739 788 if (is_wp_error($response)) {
740 - 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
741 790 return false;
742 791 }
743 792
744 793 # get the response body
@@ -748,9 +797,9 @@
748 797 $response_code = wp_remote_retrieve_response_code($response);
749 798
750 799 # if no change data skip
751 800 if (empty($body) || $response_code !== 200){
752 - 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: ' . $response_code); // phpcs:ignore WordPress.PHP.DevelopmentFunctions.error_log_error_log -- unreachable in production
753 802 return false;
754 803 }
755 804
756 805 # set the html file path
@@ -943,9 +992,11 @@
943 992
944 993 # Check for timeout or connection errors
945 994 if (is_wp_error($route_html)) {
946 995 $error_msg = $route_html->get_error_message();
947 - 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 + }
948 999 return false;
949 1000 }
950 1001
951 1002 # get body
@@ -953,12 +1004,51 @@
953 1004
954 1005 # Get the response code
955 1006 $response_code = wp_remote_retrieve_response_code($route_html);
956 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 + }
957 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 +
958 1046 # check not empty
959 1047 if(empty($html_body) || $response_code !== 200){
960 - 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 + }
961 1051 return false;
962 1052 }
963 1053
964 1054 # Skip DOM processing on oversized documents to avoid fatal OOM.
@@ -1122,9 +1212,9 @@
1122 1212
1123 1213 $result_html = preg_replace_callback(
1124 1214 '/(<' . $heading_type . '(?:\s[^>]*)?>)(.*?)(<\/' . $heading_type . '>)/is',
1125 1215 function ($m) use ($current_value, $recommended_value) {
1126 - $inner_text = trim(preg_replace('/\s+/', ' ', html_entity_decode(strip_tags($m[2]), ENT_QUOTES, 'UTF-8')));
1216 + $inner_text = trim(preg_replace('/\s+/', ' ', html_entity_decode(wp_strip_all_tags($m[2]), ENT_QUOTES, 'UTF-8')));
1127 1217 if ($inner_text !== $current_value) {
1128 1218 return $m[0];
1129 1219 }
1130 1220 # Keep the author's inline formatting; skip when it can't
@@ -1146,15 +1236,16 @@
1146 1236 $result_html = $this->otto_guard_html($this->apply_image_alt_text_via_string($result_html, $replacement_data), $result_html, 'apply_image_alt_text_via_string');
1147 1237
1148 1238 # Apply insertions — only if DOM insertion didn't already apply it
1149 1239 if (!empty($replacement_data['header_html_insertion'])) {
1150 - $header_html_check = trim($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);
1151 1247 if (strpos($result_html, $header_html_check) === false) {
1152 - $header_html_insertion = preg_replace(
1153 - '/<script(\s[^>]*)type\s*=\s*(["\'])application\/ld\+json\2/i',
1154 - '<script$1type=$2application/ld+json$2 data-otto="true"',
1155 - $replacement_data['header_html_insertion']
1156 - );
1157 1248 $safe_header = str_replace(array('\\', '$'), array('\\\\', '\\$'), $header_html_insertion);
1158 1249 $result_html = preg_replace('/(<\/head>)/i', $safe_header . "\n" . '$1', $result_html, 1);
1159 1250 }
1160 1251 }
@@ -1182,8 +1273,9 @@
1182 1273
1183 1274 # DEDUPLICATION: Remove duplicate <title>, meta description, OG, Twitter tags, canonical, and JSON-LD schema
1184 1275 $result_html = $this->otto_guard_html($this->deduplicate_title_tags($result_html), $result_html, 'deduplicate_title_tags');
1185 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');
1186 1278 $result_html = $this->otto_guard_html($this->deduplicate_og_twitter_tags($result_html), $result_html, 'deduplicate_og_twitter_tags');
1187 1279 $result_html = $this->otto_guard_html($this->apply_metabox_og_precedence($result_html), $result_html, 'apply_metabox_og_precedence');
1188 1280 $result_html = $this->otto_guard_html($this->apply_custom_seo_precedence($result_html), $result_html, 'apply_custom_seo_precedence');
1189 1281 $result_html = $this->otto_guard_html($this->deduplicate_schema_tags($result_html), $result_html, 'deduplicate_schema_tags');
@@ -1188,11 +1280,11 @@
1188 1280 $result_html = $this->otto_guard_html($this->apply_custom_seo_precedence($result_html), $result_html, 'apply_custom_seo_precedence');
1189 1281 $result_html = $this->otto_guard_html($this->deduplicate_schema_tags($result_html), $result_html, 'deduplicate_schema_tags');
1190 1282 $result_html = $this->otto_guard_html($this->deduplicate_canonical_tags($result_html), $result_html, 'deduplicate_canonical_tags');
1191 1283
1192 - # Ensure metasync_optimized attribute on <head> (post-serialization so dom->clear() can't wipe it)
1193 - if (!$this->is_amp_page() && strpos($result_html, 'metasync_optimized') === false) {
1194 - $result_html = preg_replace('/<head(\s|>)/i', '<head metasync_optimized$1', $result_html, 1);
1284 + # Mark the <head> as processed (post-serialization so dom->clear() can't wipe it)
1285 + if (!$this->is_amp_page()) {
1286 + $result_html = $this->mark_head_optimized($result_html);
1195 1287 }
1196 1288
1197 1289 $result_html = $this->otto_guard_html($this->restore_case_sensitive_attributes($result_html), $result_html, 'restore_case_sensitive_attributes');
1198 1290
@@ -1221,8 +1313,12 @@
1221 1313 # undo double-encoded numeric/hex character references produced by the
1222 1314 # bundled simplehtmldom serializer from attributes like data-x-icon-s="&#xf3c5".
1223 1315 $result_html = $this->otto_guard_html($this->repair_double_encoded_entities($result_html), $result_html, 'repair_double_encoded_entities');
1224 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 +
1225 1321 # Remove any lazy-video facade <style> hoisted into the page
1226 1322 # (runs while srcdoc is still tokenized, so the iframe's own copy is safe).
1227 1323 $result_html = $this->strip_hoisted_facade_styles($result_html);
1228 1324
@@ -1240,8 +1336,111 @@
1240 1336 return $result_html;
1241 1337 }
1242 1338
1243 1339 /**
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.
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 + /**
1244 1443 * Repair double-encoded character references in serialized HTML.
1245 1444 *
1246 1445 * The bundled simplehtmldom serializer (HtmlNode::makeup(), reached via
1247 1446 * $this->dom->save() and $root->outertext) runs every attribute value through
@@ -1264,8 +1463,118 @@
1264 1463 }
1265 1464 return preg_replace('/&amp;(#x[0-9a-fA-F]+;?|#[0-9]+;?)/', '&$1', $html);
1266 1465 }
1267 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, $shielded);
1575 + }
1576 +
1268 1577 # do the footer html insertion
1269 1578 function do_footer_html_insertion($replacement_data){
1270 1579
1271 1580 # check that we have footer html
@@ -1792,10 +2101,10 @@
1792 2101 }
1793 2102
1794 2103 # If it's a protocol-relative URL (//example.com), check if it matches our domain
1795 2104 if (strpos($url, '//') === 0) {
1796 - $parsed_home = parse_url($home_url);
1797 - $parsed_link = parse_url($url);
2105 + $parsed_home = wp_parse_url($home_url);
2106 + $parsed_link = wp_parse_url($url);
1798 2107
1799 2108 if (isset($parsed_home['host']) && isset($parsed_link['host'])) {
1800 2109 if (strtolower($parsed_home['host']) === strtolower($parsed_link['host'])) {
1801 2110 return false;
@@ -1917,21 +2226,32 @@
1917 2226 return $post_id ? (int) $post_id : 0;
1918 2227 }
1919 2228
1920 2229 /**
1921 - * Replace the first <meta $attr="$val"> content with $content_escaped (marked
1922 - * data-metasync-seo="custom"); insert one after <head> if none is present.
1923 - * Helper for apply_custom_seo_precedence().
2230 + * Replace the first <meta $attr="$val"> content with $content_escaped.
1924 2231 *
1925 2232 * @param string $html
1926 2233 * @param string $attr "name" or "property"
1927 2234 * @param string $val e.g. "description", "og:description"
1928 2235 * @param string $content_escaped Already HTML-escaped content value
2236 + * @param string $marker Provenance marker for the winning tier
1929 2237 * @return string
1930 2238 */
1931 - private function force_custom_meta($html, $attr, $val, $content_escaped) {
1932 - $tag = '<meta ' . $attr . '="' . $val . '" content="' . $content_escaped . '" data-metasync-seo="custom" />';
1933 - $pattern = '/<meta\s[^>]*' . preg_quote($attr, '/') . '\s*=\s*["\']' . preg_quote($val, '/') . '["\'][^>]*\/?>/i';
2239 + private function force_custom_meta($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';
1934 2254 $count = 0;
1935 2255 $new = preg_replace_callback($pattern, function ($m) use ($tag) {
1936 2256 return $tag;
1937 2257 }, $html, 1, $count);
@@ -1944,23 +2264,37 @@
1944 2264 }, $html, 1);
1945 2265 }
1946 2266
1947 2267 /**
1948 - * Enforce the SEO-sidebar custom title/description as the FINAL word over OTTO.
2268 + * Make sure the precedence resolver is loaded before calling it.
1949 2269 *
1950 - * The DOM-pass guards (do_header_replacements/handle_meta_element) already skip
1951 - * OTTO for these fields, but later steps re-introduce OTTO's value:
1952 - * - the manual string-replacement title pass re-applies OTTO's title;
1953 - * - OTTO's description arrives via header_html_insertion (data-otto) and the
1954 - * dedup keeps it whenever no data-metasync-seo tag is present — which is
1955 - * exactly the case when the post is synced to a third-party SEO plugin and
1956 - * the sidebar deferred emitting its own tag (output_seo_meta_description()).
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.
1957 2288 *
1958 - * Running LAST — after all dedup/precedence steps — this makes the user's
1959 - * explicit sidebar value authoritative regardless of how the tag was produced,
1960 - * mirroring apply_metabox_og_precedence() for the OG meta box. Singular views
1961 - * only; a value must be non-empty (explicitly set) to take precedence.
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.
1962 2293 *
2294 + * OG and Twitter descriptions remain outside this method deliberately;
2295 + * social precedence is a separate follow-up scope.
2296 + *
1963 2297 * @param string $html Full HTML document (post-dedup).
1964 2298 * @return string
1965 2299 */
1966 2300 private function apply_custom_seo_precedence($html) {
@@ -1971,12 +2305,26 @@
1971 2305 if (!$post_id) {
1972 2306 return $html;
1973 2307 }
1974 2308
1975 - # Title — force the custom sidebar title into the (single, post-dedup) <title>.
1976 - $custom_title = get_post_meta($post_id, '_metasync_seo_title', true);
1977 - if (!empty($custom_title) && is_string($custom_title)) {
1978 - $title_tag = '<title>' . htmlspecialchars($custom_title, ENT_QUOTES, 'UTF-8') . '</title>';
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>';
1979 2327 $count = 0;
1980 2328 $replaced = preg_replace_callback('/<title[^>]*>.*?<\/title>/is', function ($m) use ($title_tag) {
1981 2329 return $title_tag;
1982 2330 }, $html, 1, $count);
@@ -1988,19 +2336,59 @@
1988 2336 }, $html, 1);
1989 2337 }
1990 2338 }
1991 2339
1992 - # Description — force the custom sidebar description into the SEO
1993 - # <meta name="description"> only. og:/twitter: description are social tags
1994 - # owned by the OG meta box and apply_metabox_og_precedence() (which runs
1995 - # earlier); this pass must NOT touch them, or it would override a
1996 - # separately user-set og:description with the plain SEO description.
1997 - $custom_desc = get_post_meta($post_id, '_metasync_seo_desc', true);
1998 - if (!empty($custom_desc) && is_string($custom_desc)) {
1999 - $esc = htmlspecialchars($custom_desc, ENT_QUOTES, 'UTF-8');
2000 - $html = $this->force_custom_meta($html, 'name', 'description', $esc);
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);
2001 2357 }
2002 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 +
2003 2391 return $html;
2004 2392 }
2005 2393
2006 2394 # this function does the header replacements
@@ -2404,8 +2792,34 @@
2404 2792 return $html;
2405 2793 }
2406 2794
2407 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 + /**
2408 2822 * Apply the per-post "Social Media & Open Graph" meta box precedence over OTTO.
2409 2823 *
2410 2824 * Precedence per tag: an explicit meta box value set by the user wins over OTTO;
2411 2825 * a field left blank falls back to OTTO's value; if OTTO also has none, a sensible
@@ -2444,16 +2858,22 @@
2444 2858 }
2445 2859
2446 2860 # Explicit (user-typed) meta box values. A non-empty value means the user set
2447 2861 # this field themselves and it must take priority over OTTO.
2448 - $og_title_set = get_post_meta($post_id, '_metasync_og_title', true);
2449 - $og_desc_set = get_post_meta($post_id, '_metasync_og_description', true);
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');
2450 2870 $og_image_set = get_post_meta($post_id, '_metasync_og_image', true);
2451 2871 $og_type_set = get_post_meta($post_id, '_metasync_og_type', true);
2452 2872 $tw_card_set = get_post_meta($post_id, '_metasync_twitter_card', true);
2453 2873 $tw_site_set = get_post_meta($post_id, '_metasync_twitter_site', true);
2454 - $tw_title_set = get_post_meta($post_id, '_metasync_twitter_title', true);
2455 - $tw_desc_set = get_post_meta($post_id, '_metasync_twitter_description', 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');
2456 2876 $tw_image_set = get_post_meta($post_id, '_metasync_twitter_image', true);
2457 2877 $tw_image_alt_set = get_post_meta($post_id, '_metasync_twitter_image_alt', true);
2458 2878
2459 2879 # Default OG values the meta box pre-fills (post title / generated excerpt /
@@ -2713,8 +3133,82 @@
2713 3133 return $html;
2714 3134 }
2715 3135
2716 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;
3191 + break;
3192 + }
3193 + }
3194 + $keeper = $otto_tag ?: $all_tags[0];
3195 +
3196 + # Callback form avoids backreference injection when the keyword list
3197 + # contains $ followed by digits.
3198 + $first_replaced = false;
3199 + $html = preg_replace_callback($pattern, function ($m) use ($keeper, &$first_replaced) {
3200 + if (!$first_replaced) {
3201 + $first_replaced = true;
3202 + return $keeper;
3203 + }
3204 + return ''; # Remove subsequent duplicates
3205 + }, $html);
3206 +
3207 + return $html;
3208 + }
3209 +
3210 + /**
2717 3211 * Remove duplicate <link rel="canonical"> tags from HTML.
2718 3212 *
2719 3213 * When OTTO injects a canonical via header_html_insertion and MetaSync's
2720 3214 * SEO output (or WordPress core) has already emitted one, keep only the
@@ -2755,96 +3249,301 @@
2755 3249 return $html;
2756 3250 }
2757 3251
2758 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 + /**
2759 3281 * Deduplicate JSON-LD schema blocks.
2760 3282 *
2761 - * When OTTO and a third-party SEO plugin both inject <script type="application/ld+json">
2762 - * blocks, keep OTTO's version for any @type that appears in both.
2763 - * 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:
2764 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 + *
2765 3306 * @param string $html Full HTML.
2766 3307 * @return string
2767 3308 */
2768 3309 private function deduplicate_schema_tags($html) {
2769 - // Find all JSON-LD script blocks
2770 - $pattern = '/<script(\s[^>]*)type\s*=\s*(["\'])application\/ld\+json\2[^>]*>\s*([\s\S]*?)<\/script>/i';
2771 - if (preg_match_all($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)) {
2772 3317 return $html;
2773 3318 }
2774 3319
2775 - $otto_by_type = []; // @type => decoded JSON object
2776 - $third_by_type = []; // @type => decoded JSON object
2777 - $otto_graph = []; // entries from OTTO @graph blocks
2778 - $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 + }
2779 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 + $entities = [];
3339 +
2780 3340 foreach ($matches as $m) {
2781 - $attrs = $m[1];
2782 - $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];
2783 3350 $decoded = json_decode($json_str, true);
2784 3351 if (json_last_error() !== JSON_ERROR_NONE || !is_array($decoded)) {
2785 - continue; // skip unparseable blocks — leave them in place
3352 + continue; // unparseable — leave the block in place
2786 3353 }
2787 - $is_otto = stripos($attrs, 'data-otto') !== false;
2788 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' => $decoded, 'items' => []];
3380 +
2789 3381 if (isset($decoded['@graph']) && is_array($decoded['@graph'])) {
2790 - foreach ($decoded['@graph'] as $entry) {
2791 - if (!isset($entry['@type'])) continue;
2792 - // JSON-LD allows @type to be a string OR an array of strings.
2793 - // Rank Math routinely emits multi-typed entries (e.g. ["Person", "Organization"]).
2794 - // Using an array as an offset throws a fatal on PHP 8+, so normalize to scalar.
2795 - $type = is_array($entry['@type'])
2796 - ? (string) reset($entry['@type'])
2797 - : (string) $entry['@type'];
2798 - if ($type === '') continue;
2799 - if ($is_otto) {
2800 - $otto_graph[$type] = $entry;
2801 - } else {
2802 - $third_graph[$type] = $entry;
2803 - }
3382 + foreach ($decoded['@graph'] as $node) {
3383 + $block['items'][] = ['node' => $node, 'e' => is_array($node) ? $register($node) : null];
2804 3384 }
3385 + $block['shape'] = 'graph';
2805 3386 } elseif (isset($decoded['@type'])) {
2806 - $type = is_array($decoded['@type'])
2807 - ? (string) reset($decoded['@type'])
2808 - : (string) $decoded['@type'];
2809 - if ($type === '') continue;
2810 - if ($is_otto) {
2811 - $otto_by_type[$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]);
2812 3427 } else {
2813 - $third_by_type[$type] = $decoded;
3428 + $by_type[$token] = $holders;
2814 3429 }
2815 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 + }
2816 3493 }
2817 3494
2818 - // If OTTO provided no schema at all, nothing to deduplicate
2819 - if (empty($otto_by_type) && empty($otto_graph)) {
2820 - return $html;
3495 + $any_evicted = false;
3496 + foreach ($entities as $entity) {
3497 + if (!$entity['alive']) {
3498 + $any_evicted = true;
3499 + break;
3500 + }
2821 3501 }
3502 + if (!$any_evicted) {
3503 + return $html; // no collision — every original byte stays
3504 + }
2822 3505
2823 - // Remove all JSON-LD blocks from HTML
2824 - $html = preg_replace($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 + }
2825 3526
2826 - // Re-insert flat (non-@graph) blocks: OTTO wins for matching @type
2827 - $kept = array_merge($otto_by_type, array_diff_key($third_by_type, $otto_by_type));
2828 - $rebuilt = '';
2829 - foreach ($kept as $decoded) {
2830 - $rebuilt .= '<script type="application/ld+json" data-otto="true">' .
2831 - wp_json_encode($decoded) . "</script>\n";
2832 - }
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 + }
2833 3533
2834 - // Re-insert merged @graph block (if any entries exist)
2835 - $merged_graph = array_merge($third_graph, $otto_graph); // OTTO wins on duplicate @type
2836 - if (!empty($merged_graph)) {
2837 - $graph_obj = ['@context' => 'https://schema.org', '@graph' => array_values($merged_graph)];
2838 - $rebuilt .= '<script type="application/ld+json" data-otto="true">' .
2839 - wp_json_encode($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'] . '>' . $json . '</script>'];
2840 3542 }
2841 3543
2842 - // Re-inject before </head>
2843 - if (!empty($rebuilt)) {
2844 - $html = preg_replace_callback('/(<\/head>)/i', function ($m) use ($rebuilt) {
2845 - return $rebuilt . $m[1];
2846 - }, $html, 1);
3544 + foreach (array_reverse($edits) as $edit) {
3545 + $html = substr($html, 0, $edit[0]) . $edit[2] . substr($html, $edit[1]);
2847 3546 }
2848 3547
2849 3548 return $html;
2850 3549 }
@@ -2863,14 +3562,14 @@
2863 3562 return true;
2864 3563 }
2865 3564
2866 3565 # Check if amp=1 query parameter is present
2867 - if (isset($_GET['amp']) && $_GET['amp'] == '1') {
3566 + if (isset($_GET['amp']) && $_GET['amp'] == '1') { // phpcs:ignore WordPress.Security.NonceVerification.Recommended -- read-only AMP detection
2868 3567 return true;
2869 3568 }
2870 3569
2871 3570 # Check for other common AMP query parameters
2872 - if (isset($_GET['amp']) && !empty($_GET['amp'])) {
3571 + if (isset($_GET['amp']) && !empty($_GET['amp'])) { // phpcs:ignore WordPress.Security.NonceVerification.Recommended -- read-only AMP detection
2873 3572 return true;
2874 3573 }
2875 3574
2876 3575 return false;
@@ -2875,8 +3574,32 @@
2875 3574
2876 3575 return false;
2877 3576 }
2878 3577
3578 + # function to mark OTTO's own JSON-LD blocks so deduplicate_schema_tags() can
3579 + # tell them apart from a third-party SEO plugin's blocks. Every insertion path
3580 + # (DOM, string fallback, HTTP fallback) runs the fragment through this so the
3581 + # bytes they insert are identical and the "already applied" checks still match.
3582 + private function stamp_otto_json_ld($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 : $html;
3600 + }
3601 +
2879 3602 # this function insterts header html to the dom
2880 3603 function insert_header_html($data){
2881 3604
2882 3605 # check that we have the header html
@@ -2889,20 +3612,27 @@
2889 3612 $head = $this->dom->find('head', 0);
2890 3613
2891 3614 if ($head) {
2892 3615
2893 - # 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
2894 3617 $is_amp_page = $this->is_amp_page();
2895 3618
2896 - # Append the new HTML at the start of the <head> tag
2897 - # For AMP pages: use clean <head> tag without metasync_optimized attribute
2898 - # For non-AMP pages: add metasync_optimized attribute to <head> tag
2899 - if ($is_amp_page) {
2900 - $head->outertext = '<head>' .$data['header_html_insertion']. $head->innertext . '</head>';
2901 - } else {
2902 - $head->outertext = '<head metasync_optimized>' .$data['header_html_insertion']. $head->innertext . '</head>';
2903 - }
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;
2904 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 +
2905 3635 }
2906 3636
2907 3637 # save and reload DOM
2908 3638 $this->save_reload();
@@ -3283,15 +4013,16 @@
3283 4013 }
3284 4014
3285 4015 # Apply header HTML insertion (for schema, etc.) — only if DOM insertion didn't already apply it
3286 4016 if (!empty($replacement_data['header_html_insertion'])) {
3287 - $header_html_check = trim($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);
3288 4024 if (strpos($result_html, $header_html_check) === false) {
3289 - $header_html_insertion = preg_replace(
3290 - '/<script(\s[^>]*)type\s*=\s*(["\'])application\/ld\+json\2/i',
3291 - '<script$1type=$2application/ld+json$2 data-otto="true"',
3292 - $replacement_data['header_html_insertion']
3293 - );
3294 4025 $header_html = str_replace(array('\\', '$'), array('\\\\', '\\$'), $header_html_insertion);
3295 4026 # Insert before </head>
3296 4027 $result_html = preg_replace('/(<\/head>)/i', $header_html . "\n" . '$1', $result_html, 1);
3297 4028 }
@@ -3363,9 +4094,9 @@
3363 4094
3364 4095 $result_html = preg_replace_callback(
3365 4096 '/(<' . $heading_type . '(?:\s[^>]*)?>)(.*?)(<\/' . $heading_type . '>)/is',
3366 4097 function ($m) use ($current_value, $recommended_value) {
3367 - $inner_text = trim(preg_replace('/\s+/', ' ', html_entity_decode(strip_tags($m[2]), ENT_QUOTES, 'UTF-8')));
4098 + $inner_text = trim(preg_replace('/\s+/', ' ', html_entity_decode(wp_strip_all_tags($m[2]), ENT_QUOTES, 'UTF-8')));
3368 4099 if ($inner_text !== $current_value) {
3369 4100 return $m[0];
3370 4101 }
3371 4102 # Keep the author's inline formatting; skip when it can't
@@ -3423,8 +4154,9 @@
3423 4154
3424 4155 # DEDUPLICATION: Remove duplicate <title>, meta description, OG, Twitter tags, canonical, and JSON-LD schema
3425 4156 $result_html = $this->otto_guard_html($this->deduplicate_title_tags($result_html), $result_html, 'deduplicate_title_tags');
3426 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');
3427 4159 $result_html = $this->otto_guard_html($this->deduplicate_og_twitter_tags($result_html), $result_html, 'deduplicate_og_twitter_tags');
3428 4160 $result_html = $this->otto_guard_html($this->apply_metabox_og_precedence($result_html), $result_html, 'apply_metabox_og_precedence');
3429 4161 $result_html = $this->otto_guard_html($this->apply_custom_seo_precedence($result_html), $result_html, 'apply_custom_seo_precedence');
3430 4162 $result_html = $this->otto_guard_html($this->deduplicate_schema_tags($result_html), $result_html, 'deduplicate_schema_tags');
@@ -3443,11 +4175,11 @@
3443 4175
3444 4176 # Clear element cache array
3445 4177 $this->cached_elements = [];
3446 4178
3447 - # Ensure metasync_optimized attribute on <head> (post-serialization so dom->clear() can't wipe it)
3448 - if (!$this->is_amp_page() && strpos($result_html, 'metasync_optimized') === false) {
3449 - $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);
3450 4182 }
3451 4183
3452 4184 $result_html = $this->otto_guard_html($this->restore_case_sensitive_attributes($result_html), $result_html, 'restore_case_sensitive_attributes');
3453 4185
@@ -3483,8 +4215,12 @@
3483 4215 # undo double-encoded numeric/hex character references produced by the
3484 4216 # bundled simplehtmldom serializer from attributes like data-x-icon-s="&#xf3c5".
3485 4217 $result_html = $this->otto_guard_html($this->repair_double_encoded_entities($result_html), $result_html, 'repair_double_encoded_entities');
3486 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 +
3487 4223 # Remove any lazy-video facade <style> hoisted into the page
3488 4224 # (runs while srcdoc is still tokenized, so the iframe's own copy is safe).
3489 4225 $result_html = $this->strip_hoisted_facade_styles($result_html);
3490 4226
@@ -3501,12 +4237,26 @@
3501 4237
3502 4238 return $result_html;
3503 4239
3504 4240 } catch (Exception $e) {
3505 - 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 + }
3506 4249 return false;
3507 4250 } catch (Error $e) {
3508 - 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 + }
3509 4259 return false;
3510 4260 }
3511 4261 }
3512 4262
@@ -3555,23 +4305,15 @@
3555 4305 }
3556 4306 }
3557 4307 }
3558 4308
3559 - if ($remove_title && !$has_custom_title) {
3560 - # Remove Open Graph and Twitter title tags (keep main <title>)
3561 - $title_selectors = [
3562 - 'meta[property=og:title]',
3563 - 'meta[name=twitter:title]',
3564 - ];
4309 + # Preserve third-party og:title/twitter:title tags here. They used to be
4310 + # stripped whenever $remove_title was set, but that also removes them
4311 + # when no custom title exists to replace them, leaving pages with no
4312 + # og:title/twitter:title at all. OTTO's own title/OG dedup and
4313 + # precedence passes already reconcile duplicates, so this branch no
4314 + # longer needs to delete those tags.
3565 4315
3566 - foreach ($title_selectors as $selector) {
3567 - $tags = $this->dom->find($selector);
3568 - foreach ($tags as $tag) {
3569 - $tag->outertext = ''; # Remove the tag
3570 - }
3571 - }
3572 - }
3573 -
3574 4316 # Save changes
3575 4317 $this->save_reload();
3576 4318 }
3577 4319
@@ -3795,5 +4537,5 @@
3795 4537 return $html;
3796 4538 }
3797 4539
3798 4540
3799 -}
4541 +}