| @@ -5,8 +5,17 @@ | ||
| 5 | 5 | |
| 6 | 6 | class MxChat_Utils { |
| 7 | 7 | |
| 8 | 8 | /** |
| 9 | + * True while a storage routine is re-writing an entry it is about to re-add | |
| 10 | + * (submit_chunked_content's clean-slate delete). delete_chunks_for_url() skips | |
| 11 | + * the Vector Store mirror-delete while set — an internal re-store is not an | |
| 12 | + * entry removal, and mirroring it would delete-then-reupload the entry's file | |
| 13 | + * on every chunked save (plan 15b5c6). | |
| 14 | + */ | |
| 15 | +private static $vectorstore_mirror_suspended = false; | |
| 16 | + | |
| 17 | +/** | |
| 9 | 18 | * Validate a client-supplied session id (plan-mxchat-20260731-d42bec). |
| 10 | 19 | * |
| 11 | 20 | * sanitize_text_field() — which every session_id read site used before this — |
| 12 | 21 | * preserves '/' and '..'. Harmless where the value is only an option or |
| @@ -38,8 +47,47 @@ | ||
| 38 | 47 | return preg_match('/\A[A-Za-z0-9_-]{1,128}\z/', $val) ? $val : ''; |
| 39 | 48 | } |
| 40 | 49 | |
| 41 | 50 | /** |
| 51 | + * Sanitize the lead-capture consent-checkbox label (plan b062c4). | |
| 52 | + * | |
| 53 | + * The label is owner-supplied and renders inside the widget's email form, so | |
| 54 | + * this is a security boundary, not a formatting nicety. One explicit | |
| 55 | + * allowlist, used at BOTH save time (options.php sanitize + autosave AJAX) | |
| 56 | + * and render time (widget form, admin surfaces) so the two can never drift: | |
| 57 | + * an anchor — the whole point is "I agree to the <a>Privacy Policy</a>" — | |
| 58 | + * plus inline emphasis. No block tags, no images, no style attributes. | |
| 59 | + * | |
| 60 | + * The stored consent record keeps this exact sanitized string as "the text | |
| 61 | + * the visitor saw", so it must be deterministic: same input, same output, | |
| 62 | + * whichever path ran it. | |
| 63 | + * | |
| 64 | + * @param mixed $raw Owner-entered label. | |
| 65 | + * @return string Sanitized label, capped at 1000 chars. | |
| 66 | + */ | |
| 67 | +public static function sanitize_consent_label($raw) { | |
| 68 | + if (!is_scalar($raw)) { | |
| 69 | + return ''; | |
| 70 | + } | |
| 71 | + | |
| 72 | + $allowed = array( | |
| 73 | + 'a' => array( | |
| 74 | + 'href' => true, | |
| 75 | + 'title' => true, | |
| 76 | + 'target' => true, | |
| 77 | + 'rel' => true, | |
| 78 | + ), | |
| 79 | + 'strong' => array(), | |
| 80 | + 'em' => array(), | |
| 81 | + 'br' => array(), | |
| 82 | + ); | |
| 83 | + | |
| 84 | + $label = wp_kses(trim((string) $raw), $allowed); | |
| 85 | + | |
| 86 | + return mb_substr($label, 0, 1000); | |
| 87 | +} | |
| 88 | + | |
| 89 | +/** | |
| 42 | 90 | * Per-request cache for get_session_history(). Mirrors get_option()'s |
| 43 | 91 | * request-scoped caching, which the mxchat_history_ option reads got for |
| 44 | 92 | * free before plan 839c4c moved history reads onto the transcripts table. |
| 45 | 93 | */ |
| @@ -137,8 +185,78 @@ | ||
| 137 | 185 | unset(self::$history_cache[(string) $session_id]); |
| 138 | 186 | } |
| 139 | 187 | |
| 140 | 188 | /** |
| 189 | + * Append one message to a session's transcript on behalf of an add-on | |
| 190 | + * (plan 4a664e). The Forms add-on's wizard steps travel over the add-on's | |
| 191 | + * own AJAX action, never through the chat request, so nothing wrote them | |
| 192 | + * to the transcript: Transcripts, the live-agent handoff card, Download | |
| 193 | + * Transcript and every transcript email showed "[WIZARD_COMPLETE]" and no | |
| 194 | + * answers. This is the write half of get_session_history() — same row | |
| 195 | + * shape as the chat writer (visitor identity, captured name/email and | |
| 196 | + * originating page from the session store, GMT timestamp), without the | |
| 197 | + * new-session side effects (ownership, notification email, delayed | |
| 198 | + * transcript) that only the visitor's first message should trigger. | |
| 199 | + * | |
| 200 | + * Public and static so add-ons can call it, guarded with method_exists | |
| 201 | + * against an older mxchat-basic. | |
| 202 | + * | |
| 203 | + * @param string $session_id Chat session id. | |
| 204 | + * @param string $role user|bot|system|agent. | |
| 205 | + * @param string $message Stored as-is: bot rows may carry HTML, user rows | |
| 206 | + * are rendered as text by every reader. | |
| 207 | + * @return int The new transcripts row id, 0 when nothing was written. | |
| 208 | + */ | |
| 209 | +public static function save_session_message($session_id, $role, $message) { | |
| 210 | + global $wpdb; | |
| 211 | + | |
| 212 | + $session_id = self::sanitize_session_id($session_id); | |
| 213 | + $role = is_string($role) ? strtolower(trim($role)) : ''; | |
| 214 | + $message = is_scalar($message) ? trim((string) $message) : ''; | |
| 215 | + | |
| 216 | + if ($session_id === '' || $message === '' | |
| 217 | + || !in_array($role, array('user', 'bot', 'system', 'agent'), true)) { | |
| 218 | + return 0; | |
| 219 | + } | |
| 220 | + | |
| 221 | + $table = $wpdb->prefix . 'mxchat_chat_transcripts'; | |
| 222 | + | |
| 223 | + $saved_email = MxChat_Session_Store::get($session_id, 'email'); | |
| 224 | + $saved_name = MxChat_Session_Store::get($session_id, 'name'); | |
| 225 | + | |
| 226 | + $row = array( | |
| 227 | + 'user_id' => is_user_logged_in() ? get_current_user_id() : 0, | |
| 228 | + 'user_identifier' => MxChat_User::mxchat_get_user_identifier(), | |
| 229 | + 'user_email' => $saved_email ?: MxChat_User::mxchat_get_user_email(), | |
| 230 | + 'user_name' => $saved_name ?: '', | |
| 231 | + 'session_id' => $session_id, | |
| 232 | + 'role' => $role, | |
| 233 | + 'message' => $message, | |
| 234 | + 'timestamp' => current_time('mysql', 1), | |
| 235 | + ); | |
| 236 | + | |
| 237 | + // Same page attribution as the rest of the session; the column pair was | |
| 238 | + // added by an upgrade routine, so tolerate its absence like the chat | |
| 239 | + // writer does. | |
| 240 | + $originating = MxChat_Session_Store::get($session_id, 'originating_page'); | |
| 241 | + if (is_array($originating) && !empty($originating['url']) | |
| 242 | + && $wpdb->get_var("SHOW COLUMNS FROM `$table` LIKE 'originating_page_url'")) { | |
| 243 | + $row['originating_page_url'] = $originating['url']; | |
| 244 | + $row['originating_page_title'] = isset($originating['title']) ? $originating['title'] : ''; | |
| 245 | + } | |
| 246 | + | |
| 247 | + $inserted = $wpdb->insert($table, $row); | |
| 248 | + if ($inserted === false) { | |
| 249 | + return 0; | |
| 250 | + } | |
| 251 | + | |
| 252 | + $row_id = (int) $wpdb->insert_id; | |
| 253 | + self::flush_session_history_cache($session_id); | |
| 254 | + | |
| 255 | + return $row_id; | |
| 256 | +} | |
| 257 | + | |
| 258 | +/** | |
| 141 | 259 | * Most recipients the Notification Email field will accept (plan 2f131a). |
| 142 | 260 | * A settings field is not a mailing list. |
| 143 | 261 | */ |
| 144 | 262 | const NOTIFICATION_EMAIL_MAX = 5; |
| @@ -417,9 +535,9 @@ | ||
| 417 | 535 | * @param string $bot_id The bot ID for multi-bot support |
| 418 | 536 | * @param string $content_type The type of content (post, page, pdf, url, manual, product, etc.) |
| 419 | 537 | * @return bool|WP_Error True on success, WP_Error on failure |
| 420 | 538 | */ |
| 421 | -public static function submit_content_to_db($content, $source_url, $api_key, $vector_id = null, $bot_id = 'default', $content_type = 'content') { | |
| 539 | +public static function submit_content_to_db($content, $source_url, $api_key, $vector_id = null, $bot_id = 'default', $content_type = 'content', $title = '') { | |
| 422 | 540 | global $wpdb; |
| 423 | 541 | $table_name = $wpdb->prefix . 'mxchat_system_prompt_content'; |
| 424 | 542 | |
| 425 | 543 | //error_log('[MXCHAT-DB] Starting database submission for URL: ' . $source_url . ' (Bot: ' . $bot_id . ', Type: ' . $content_type . ')'); |
| @@ -424,10 +542,18 @@ | ||
| 424 | 542 | |
| 425 | 543 | //error_log('[MXCHAT-DB] Starting database submission for URL: ' . $source_url . ' (Bot: ' . $bot_id . ', Type: ' . $content_type . ')'); |
| 426 | 544 | //error_log('[MXCHAT-DB] Content length: ' . strlen($content) . ' bytes'); |
| 427 | 545 | |
| 428 | - // Sanitize the source URL | |
| 429 | - $source_url = esc_url_raw($source_url); | |
| 546 | + // Sanitize the source URL. Internal identities (mxchat:// manual docs, | |
| 547 | + // upload:// file uploads) are NOT web URLs — esc_url_raw EMPTIES them | |
| 548 | + // because its protocol list is statically cached and effectively | |
| 549 | + // unfilterable (measured, plan 945406 / 0485e5) — sanitize those as text | |
| 550 | + // so upserts stay keyed to a stable identity across re-imports. | |
| 551 | + if (preg_match('#^(mxchat|upload)://#i', $source_url)) { | |
| 552 | + $source_url = sanitize_text_field($source_url); | |
| 553 | + } else { | |
| 554 | + $source_url = esc_url_raw($source_url); | |
| 555 | + } | |
| 430 | 556 | |
| 431 | 557 | // Sanitize content_type |
| 432 | 558 | $content_type = sanitize_key($content_type); |
| 433 | 559 | if (empty($content_type)) { |
| @@ -433,8 +559,14 @@ | ||
| 433 | 559 | if (empty($content_type)) { |
| 434 | 560 | $content_type = 'content'; // Fallback for backwards compatibility |
| 435 | 561 | } |
| 436 | 562 | |
| 563 | + // Page title captured by the importer (plan d9ee66), registered under the | |
| 564 | + // SANITIZED url so the storage writers below find it by the same key. | |
| 565 | + if (is_string($title) && trim($title) !== '') { | |
| 566 | + self::set_source_title($source_url, $title); | |
| 567 | + } | |
| 568 | + | |
| 437 | 569 | // Just ensure UTF-8 validity without aggressive escaping |
| 438 | 570 | $safe_content = wp_check_invalid_utf8($content); |
| 439 | 571 | // Remove only null bytes and other control characters, but preserve newlines (\n = \x0A) and carriage returns (\r = \x0D) |
| 440 | 572 | $safe_content = preg_replace('/[\x00-\x08\x0B\x0C\x0E-\x1F\x7F]/u', '', $safe_content); |
| @@ -442,9 +574,16 @@ | ||
| 442 | 574 | // Check if chunking should be applied |
| 443 | 575 | $chunker = MxChat_Chunker::from_settings(); |
| 444 | 576 | if ($chunker->should_chunk($safe_content)) { |
| 445 | 577 | //error_log('[MXCHAT-DB] Content exceeds chunk threshold, using chunked submission'); |
| 446 | - return self::submit_chunked_content($safe_content, $source_url, $api_key, $bot_id, $content_type, $chunker); | |
| 578 | + $chunk_result = self::submit_chunked_content($safe_content, $source_url, $api_key, $bot_id, $content_type, $chunker); | |
| 579 | + // Vector Store mirror gets the entry WHOLE — one file per KB entry, | |
| 580 | + // OpenAI chunks server-side; local chunking is our own embedding | |
| 581 | + // concern and never reaches the store (plan 15b5c6). | |
| 582 | + if ($chunk_result === true && class_exists('MxChat_Vectorstore_Manager')) { | |
| 583 | + MxChat_Vectorstore_Manager::sync_upsert_entry($source_url, $safe_content, $bot_id, $content_type); | |
| 584 | + } | |
| 585 | + return $chunk_result; | |
| 447 | 586 | } |
| 448 | 587 | |
| 449 | 588 | // UPDATED: Generate the embedding using bot-specific configuration |
| 450 | 589 | $embedding_vector = self::generate_embedding($content, $api_key, $bot_id); |
| @@ -462,18 +601,175 @@ | ||
| 462 | 601 | // UPDATED: Check if Pinecone is enabled for this specific bot |
| 463 | 602 | if (self::is_pinecone_enabled_for_bot($bot_id)) { |
| 464 | 603 | //error_log('[MXCHAT-DB] Pinecone is enabled for bot ' . $bot_id . ' - using Pinecone storage'); |
| 465 | 604 | // Store in Pinecone only |
| 466 | - return self::store_in_pinecone_only($embedding_vector, $content, $source_url, $vector_id, $bot_id, $content_type); | |
| 605 | + $result = self::store_in_pinecone_only($embedding_vector, $content, $source_url, $vector_id, $bot_id, $content_type); | |
| 606 | + // If this URL previously stored as CHUNKED content and the new content fits in a | |
| 607 | + // single vector, the upsert above only overwrote the base id — the old | |
| 608 | + // md5(url)_chunk_N vectors would keep serving the stale text. Sweep them. | |
| 609 | + // Only for a real source_url: chunk ids derive from it, and md5('') is shared | |
| 610 | + // by legacy URL-less entries so a blind sweep there could hit other entries. | |
| 611 | + if ($result === true && !empty($source_url)) { | |
| 612 | + self::cleanup_pinecone_chunk_stragglers($source_url, $bot_id); | |
| 613 | + } | |
| 614 | + if ($result === true && class_exists('MxChat_Vectorstore_Manager')) { | |
| 615 | + MxChat_Vectorstore_Manager::sync_upsert_entry($source_url, $safe_content, $bot_id, $content_type); | |
| 616 | + } | |
| 617 | + return $result; | |
| 467 | 618 | } else { |
| 468 | 619 | //error_log('[MXCHAT-DB] Pinecone not enabled for bot ' . $bot_id . ' - using WordPress storage'); |
| 469 | 620 | // Store in WordPress database only |
| 470 | 621 | $embedding_vector_serialized = maybe_serialize($embedding_vector); |
| 471 | - return self::store_in_wordpress_db($safe_content, $source_url, $embedding_vector_serialized, $table_name, $content_type); | |
| 622 | + $result = self::store_in_wordpress_db($safe_content, $source_url, $embedding_vector_serialized, $table_name, $content_type); | |
| 623 | + if ($result === true && class_exists('MxChat_Vectorstore_Manager')) { | |
| 624 | + MxChat_Vectorstore_Manager::sync_upsert_entry($source_url, $safe_content, $bot_id, $content_type); | |
| 625 | + } | |
| 626 | + return $result; | |
| 472 | 627 | } |
| 473 | 628 | } |
| 474 | 629 | |
| 630 | +/* ------------------------------------------------------------------------- | |
| 631 | + * Source titles (3.2.23, plan d9ee66) | |
| 632 | + * | |
| 633 | + * A knowledge entry imported from a URL or sitemap remembers the page's own | |
| 634 | + * title, so an add-on that names a source (AI Search's source cards) can show | |
| 635 | + * "About Us" instead of a guess from the slug ("Aboutus"). Stored in the | |
| 636 | + * `title` column of the WP knowledge table and as `title` metadata on Pinecone | |
| 637 | + * vectors / documents. The importer hands the title to submit_content_to_db(), | |
| 638 | + * which registers it under the source URL; the writers look it up by that key. | |
| 639 | + * ---------------------------------------------------------------------- */ | |
| 640 | + | |
| 641 | +/** @var array<string,string> source_url => title, for the current request */ | |
| 642 | +private static $source_titles = array(); | |
| 643 | + | |
| 644 | +public static function set_source_title($source_url, $title) { | |
| 645 | + $key = trim((string) $source_url); | |
| 646 | + $title = self::clean_title($title); | |
| 647 | + if ($key === '' || $title === '') { | |
| 648 | + return; | |
| 649 | + } | |
| 650 | + self::$source_titles[$key] = $title; | |
| 651 | +} | |
| 652 | + | |
| 653 | +public static function source_title_for($source_url) { | |
| 654 | + $key = trim((string) $source_url); | |
| 655 | + return ($key !== '' && isset(self::$source_titles[$key])) ? self::$source_titles[$key] : ''; | |
| 656 | +} | |
| 657 | + | |
| 658 | +/** Entities decoded, tags gone, whitespace collapsed, capped at the column width. */ | |
| 659 | +public static function clean_title($title) { | |
| 660 | + $title = html_entity_decode((string) $title, ENT_QUOTES | ENT_HTML5, 'UTF-8'); | |
| 661 | + $title = wp_strip_all_tags($title); | |
| 662 | + $title = trim((string) preg_replace('/\s+/u', ' ', $title)); | |
| 663 | + if (mb_strlen($title, 'UTF-8') > 255) { | |
| 664 | + $title = rtrim(mb_substr($title, 0, 255, 'UTF-8')); | |
| 665 | + } | |
| 666 | + return $title; | |
| 667 | +} | |
| 668 | + | |
| 475 | 669 | /** |
| 670 | + * The page's own title from fetched HTML: og:title, else <title> with a | |
| 671 | + * trailing "| Site Name" style suffix removed, else the first <h1>. Returns '' | |
| 672 | + * when nothing usable is there, so the caller keeps its own fallback. | |
| 673 | + */ | |
| 674 | +public static function extract_page_title($html) { | |
| 675 | + $html = (string) $html; | |
| 676 | + if ($html === '') { | |
| 677 | + return ''; | |
| 678 | + } | |
| 679 | + $head = substr($html, 0, 200000); | |
| 680 | + $site = ''; | |
| 681 | + if (preg_match('/<meta\s+[^>]*property=["\']og:site_name["\'][^>]*content=["\']([^"\']{1,120})["\']/i', $head, $m) | |
| 682 | + || preg_match('/<meta\s+[^>]*content=["\']([^"\']{1,120})["\'][^>]*property=["\']og:site_name["\']/i', $head, $m)) { | |
| 683 | + $site = self::clean_title($m[1]); | |
| 684 | + } | |
| 685 | + $candidate = ''; | |
| 686 | + if (preg_match('/<meta\s+[^>]*property=["\']og:title["\'][^>]*content=["\']([^"\']{1,500})["\']/i', $head, $m) | |
| 687 | + || preg_match('/<meta\s+[^>]*content=["\']([^"\']{1,500})["\'][^>]*property=["\']og:title["\']/i', $head, $m)) { | |
| 688 | + $candidate = self::clean_title($m[1]); | |
| 689 | + } | |
| 690 | + if ($candidate === '' && preg_match('/<title[^>]*>(.*?)<\/title>/is', $head, $m)) { | |
| 691 | + $candidate = self::clean_title($m[1]); | |
| 692 | + } | |
| 693 | + if ($candidate !== '') { | |
| 694 | + $candidate = self::strip_site_suffix($candidate, $site); | |
| 695 | + } | |
| 696 | + if ($candidate === '' && preg_match('/<h1[^>]*>(.*?)<\/h1>/is', $html, $m)) { | |
| 697 | + $candidate = self::clean_title($m[1]); | |
| 698 | + } | |
| 699 | + return $candidate; | |
| 700 | +} | |
| 701 | + | |
| 702 | +/** | |
| 703 | + * "About Us | Just Facts" -> "About Us". Only a TRAILING segment after a spaced | |
| 704 | + * separator goes, only when what is left is a real title (3+ characters) and the | |
| 705 | + * removed part is short enough to be a site name; a known og:site_name wins. | |
| 706 | + * Byte offsets throughout: every separator starts with an ASCII space, so a cut | |
| 707 | + * there never lands inside a multibyte character. | |
| 708 | + */ | |
| 709 | +private static function strip_site_suffix($title, $site_name = '') { | |
| 710 | + $seps = array(' | ', ' – ', ' — ', ' - ', ' :: ', ' · ', ' » ', ' • '); | |
| 711 | + if ($site_name !== '') { | |
| 712 | + foreach ($seps as $sep) { | |
| 713 | + $tail = $sep . $site_name; | |
| 714 | + if (strlen($title) > strlen($tail) && substr($title, -strlen($tail)) === $tail) { | |
| 715 | + return trim(substr($title, 0, strlen($title) - strlen($tail))); | |
| 716 | + } | |
| 717 | + } | |
| 718 | + } | |
| 719 | + $best_pos = -1; | |
| 720 | + $best_sep = ''; | |
| 721 | + foreach ($seps as $sep) { | |
| 722 | + $pos = strrpos($title, $sep); | |
| 723 | + if ($pos !== false && $pos > $best_pos) { | |
| 724 | + $best_pos = $pos; | |
| 725 | + $best_sep = $sep; | |
| 726 | + } | |
| 727 | + } | |
| 728 | + if ($best_pos > 0) { | |
| 729 | + $left = trim(substr($title, 0, $best_pos)); | |
| 730 | + $right = trim(substr($title, $best_pos + strlen($best_sep))); | |
| 731 | + if (mb_strlen($left, 'UTF-8') >= 3 && $right !== '' && mb_strlen($right, 'UTF-8') <= 60) { | |
| 732 | + return $left; | |
| 733 | + } | |
| 734 | + } | |
| 735 | + return $title; | |
| 736 | +} | |
| 737 | + | |
| 738 | +/** | |
| 739 | + * Make sure the knowledge table has the `title` column (3.2.23, plan d9ee66). | |
| 740 | + * Activation adds it; this is the safety net for a copy updated in place, and | |
| 741 | + * the one place add-ons ask "may I read/write title?". A positive answer is | |
| 742 | + * remembered in an option, so after the first time this costs one option read. | |
| 743 | + */ | |
| 744 | +public static function ensure_kb_title_column() { | |
| 745 | + global $wpdb; | |
| 746 | + static $ready = null; | |
| 747 | + if ($ready !== null) { | |
| 748 | + return $ready; | |
| 749 | + } | |
| 750 | + if (get_option('mxchat_kb_title_column') === '1') { | |
| 751 | + $ready = true; | |
| 752 | + return true; | |
| 753 | + } | |
| 754 | + $table = $wpdb->prefix . 'mxchat_system_prompt_content'; | |
| 755 | + $cols = $wpdb->get_col("SHOW COLUMNS FROM {$table}", 0); | |
| 756 | + if (empty($cols)) { | |
| 757 | + $ready = false; | |
| 758 | + return false; | |
| 759 | + } | |
| 760 | + if (!in_array('title', $cols, true)) { | |
| 761 | + $wpdb->query("ALTER TABLE {$table} ADD COLUMN title VARCHAR(255) DEFAULT NULL AFTER source_url"); | |
| 762 | + $cols = $wpdb->get_col("SHOW COLUMNS FROM {$table}", 0); | |
| 763 | + } | |
| 764 | + $ready = in_array('title', $cols, true); | |
| 765 | + if ($ready) { | |
| 766 | + update_option('mxchat_kb_title_column', '1', false); | |
| 767 | + } | |
| 768 | + return $ready; | |
| 769 | +} | |
| 770 | + | |
| 771 | +/** | |
| 476 | 772 | * UPDATED: Check if Pinecone is enabled and properly configured for a specific bot |
| 477 | 773 | */ |
| 478 | 774 | private static function is_pinecone_enabled_for_bot($bot_id = 'default') { |
| 479 | 775 | // For default bot or when multi-bot is not active, use original method |
| @@ -585,11 +881,15 @@ | ||
| 585 | 881 | // Check if this is truly manual content (no URL at all) vs a real URL that filter_var rejects |
| 586 | 882 | // filter_var(FILTER_VALIDATE_URL) rejects valid URLs with encoded chars, non-ASCII, fragments, etc. |
| 587 | 883 | // Use a looser check: if it starts with http(s):// or has a scheme, it's a URL |
| 588 | 884 | $has_url_scheme = !empty($source_url) && preg_match('#^https?://#i', $source_url); |
| 885 | + // upload:// identities (admin document/PDF uploads, plan 0485e5) are stable | |
| 886 | + // and deduplicable — treat them like URLs so a re-upload UPDATES the row | |
| 887 | + // instead of minting a fresh manual identity (which would duplicate). | |
| 888 | + $has_stable_identity = $has_url_scheme || (!empty($source_url) && preg_match('#^upload://#i', $source_url)); | |
| 589 | 889 | // Treat legacy mxchat.ai source URLs as manual — old bug assigned the site URL to manual entries |
| 590 | 890 | $is_legacy_mxchat_url = $has_url_scheme && strpos($source_url, 'mxchat.ai') !== false; |
| 591 | - $is_manual_content = empty($source_url) || $source_url === '' || !$has_url_scheme || $is_legacy_mxchat_url; | |
| 891 | + $is_manual_content = empty($source_url) || $source_url === '' || !$has_stable_identity || $is_legacy_mxchat_url; | |
| 592 | 892 | |
| 593 | 893 | if ($is_manual_content) { |
| 594 | 894 | // Generate unique identifier for manual content to prevent overwrites |
| 595 | 895 | $source_url = 'mxchat://manual-content/' . time() . '-' . wp_generate_password(8, false); |
| @@ -610,8 +910,13 @@ | ||
| 610 | 910 | //error_log('[MXCHAT-DB] Manual content - will create new entry (no duplicate check)'); |
| 611 | 911 | } |
| 612 | 912 | // ===== END FIX ===== |
| 613 | 913 | |
| 914 | + // Page title captured at import (plan d9ee66): written only when the | |
| 915 | + // importer supplied one, so a re-index without a title keeps the old value. | |
| 916 | + $title_value = $is_manual_content ? '' : self::source_title_for($source_url); | |
| 917 | + $title_ready = ($title_value !== '') && self::ensure_kb_title_column(); | |
| 918 | + | |
| 614 | 919 | // Progressive fallback mechanism for problematic content |
| 615 | 920 | $attempt = 1; |
| 616 | 921 | $max_attempts = 3; |
| 617 | 922 | $current_content = $safe_content; |
| @@ -622,39 +927,41 @@ | ||
| 622 | 927 | if ($existing_id) { |
| 623 | 928 | //error_log('[MXCHAT-DB] Found existing entry (ID: ' . $existing_id . '). Updating... (Attempt ' . $attempt . ')'); |
| 624 | 929 | |
| 625 | 930 | // Update the existing row - UPDATED 2.5.6: Added content_type |
| 626 | - $result = $wpdb->update( | |
| 627 | - $table_name, | |
| 628 | - array( | |
| 629 | - 'url' => $source_url, | |
| 630 | - 'article_content' => $current_content, | |
| 631 | - 'embedding_vector' => $embedding_vector_serialized, | |
| 632 | - 'source_url' => $source_url, | |
| 633 | - 'content_type' => $content_type, | |
| 634 | - 'timestamp' => current_time('mysql'), | |
| 635 | - ), | |
| 636 | - array('id' => $existing_id), | |
| 637 | - array('%s','%s','%s','%s','%s','%s'), | |
| 638 | - array('%d') | |
| 931 | + $row = array( | |
| 932 | + 'url' => $source_url, | |
| 933 | + 'article_content' => $current_content, | |
| 934 | + 'embedding_vector' => $embedding_vector_serialized, | |
| 935 | + 'source_url' => $source_url, | |
| 936 | + 'content_type' => $content_type, | |
| 937 | + 'timestamp' => current_time('mysql'), | |
| 639 | 938 | ); |
| 939 | + $formats = array('%s','%s','%s','%s','%s','%s'); | |
| 940 | + if ($title_ready) { | |
| 941 | + $row['title'] = $title_value; | |
| 942 | + $formats[] = '%s'; | |
| 943 | + } | |
| 944 | + $result = $wpdb->update($table_name, $row, array('id' => $existing_id), $formats, array('%d')); | |
| 640 | 945 | } else { |
| 641 | 946 | //error_log('[MXCHAT-DB] No existing entry found. Inserting new row... (Attempt ' . $attempt . ')'); |
| 642 | 947 | //error_log('[MXCHAT-DB] Content sample: ' . substr($current_content, 0, 1000)); |
| 643 | 948 | |
| 644 | 949 | // Insert a new row - UPDATED 2.5.6: Added content_type |
| 645 | - $result = $wpdb->insert( | |
| 646 | - $table_name, | |
| 647 | - array( | |
| 648 | - 'url' => $source_url, // Now unique for manual content | |
| 649 | - 'article_content' => $current_content, | |
| 650 | - 'embedding_vector' => $embedding_vector_serialized, | |
| 651 | - 'source_url' => $source_url, // Now unique for manual content | |
| 652 | - 'content_type' => $content_type, | |
| 653 | - 'timestamp' => current_time('mysql'), | |
| 654 | - ), | |
| 655 | - array('%s','%s','%s','%s','%s','%s') | |
| 950 | + $row = array( | |
| 951 | + 'url' => $source_url, // Now unique for manual content | |
| 952 | + 'article_content' => $current_content, | |
| 953 | + 'embedding_vector' => $embedding_vector_serialized, | |
| 954 | + 'source_url' => $source_url, // Now unique for manual content | |
| 955 | + 'content_type' => $content_type, | |
| 956 | + 'timestamp' => current_time('mysql'), | |
| 656 | 957 | ); |
| 958 | + $formats = array('%s','%s','%s','%s','%s','%s'); | |
| 959 | + if ($title_ready) { | |
| 960 | + $row['title'] = $title_value; | |
| 961 | + $formats[] = '%s'; | |
| 962 | + } | |
| 963 | + $result = $wpdb->insert($table_name, $row, $formats); | |
| 657 | 964 | } |
| 658 | 965 | |
| 659 | 966 | if ($result === false) { |
| 660 | 967 | //error_log('[MXCHAT-DB] Database operation failed (Attempt ' . $attempt . ')'); |
| @@ -713,10 +1020,11 @@ | ||
| 713 | 1020 | // ===== UPDATED: Handle manual content with unique vector IDs ===== |
| 714 | 1021 | if ($vector_id) { |
| 715 | 1022 | // Use provided vector ID |
| 716 | 1023 | //error_log('[MXCHAT-PINECONE-MAIN] Using provided vector ID: ' . $vector_id); |
| 717 | - } elseif (!empty($url) && preg_match('#^https?://#i', $url)) { | |
| 718 | - // For URLs, use URL-based ID (existing behavior) | |
| 1024 | + } elseif (!empty($url) && preg_match('#^(https?|upload)://#i', $url)) { | |
| 1025 | + // For URLs — and stable upload:// identities (plan 0485e5) — use an | |
| 1026 | + // identity-derived ID so a re-import upserts the same vector. | |
| 719 | 1027 | $vector_id = md5($url); |
| 720 | 1028 | //error_log('[MXCHAT-PINECONE-MAIN] Generated vector ID from URL: ' . $vector_id); |
| 721 | 1029 | } else { |
| 722 | 1030 | // For manual content (empty/no URL scheme), generate unique ID |
| @@ -777,9 +1085,23 @@ | ||
| 777 | 1085 | 'last_updated' => time(), |
| 778 | 1086 | 'created_at' => time(), // Add creation timestamp |
| 779 | 1087 | 'bot_id' => $bot_id, // Add bot identification |
| 780 | 1088 | ); |
| 781 | - | |
| 1089 | + $page_title = self::source_title_for($url); | |
| 1090 | + if ($page_title !== '') { | |
| 1091 | + $metadata['title'] = $page_title; // page title captured at import (plan d9ee66) | |
| 1092 | + } | |
| 1093 | + | |
| 1094 | + // Document index (plan 362c31): same id, same metadata, through the | |
| 1095 | + // documents API. A site on the classic vector index never enters here. | |
| 1096 | + if (class_exists('MxChat_Pinecone_Documents') && MxChat_Pinecone_Documents::is_document_index($bot_id)) { | |
| 1097 | + return MxChat_Pinecone_Documents::upsert_one($vector_id, $content, $embedding_vector, $metadata, array( | |
| 1098 | + 'api_key' => $api_key, | |
| 1099 | + 'host' => $host, | |
| 1100 | + 'namespace' => $namespace, | |
| 1101 | + )); | |
| 1102 | + } | |
| 1103 | + | |
| 782 | 1104 | $vector_data = array( |
| 783 | 1105 | 'id' => $vector_id, |
| 784 | 1106 | 'values' => $embedding_vector, |
| 785 | 1107 | 'metadata' => $metadata |
| @@ -1247,10 +1569,26 @@ | ||
| 1247 | 1569 | |
| 1248 | 1570 | //error_log('[MXCHAT-CHUNK-DEBUG] Starting chunked submission for: ' . $source_url); |
| 1249 | 1571 | //error_log('[MXCHAT-CHUNK-DEBUG] Content length: ' . strlen($content) . ' chars'); |
| 1250 | 1572 | |
| 1251 | - // First, delete any existing chunks for this URL (clean slate) | |
| 1573 | + // URL-less (manual) content needs a minted identity BEFORE chunk ids are derived: | |
| 1574 | + // every chunk id is md5(source_url)_chunk_N, so with source_url = '' EVERY long | |
| 1575 | + // manual document shared the md5('') prefix — and the clean-slate delete below | |
| 1576 | + // wiped the PREVIOUS manual entry's chunks each time a new one was added. The | |
| 1577 | + // single-vector paths already mint (mxchat:// in WP, manual_* in Pinecone); this | |
| 1578 | + // was the one storage path that didn't. Keep an mxchat:// identity if the caller | |
| 1579 | + // already carries one. | |
| 1580 | + if (strpos($source_url, 'mxchat://') !== 0 && strpos($source_url, 'upload://') !== 0 && !preg_match('#^https?://#i', $source_url)) { | |
| 1581 | + $source_url = 'mxchat://manual-content/' . time() . '-' . wp_generate_password(8, false); | |
| 1582 | + } | |
| 1583 | + | |
| 1584 | + // First, delete any existing chunks for this URL (clean slate). | |
| 1585 | + // Mirror-suspended: this is a re-store, not an entry removal — the Vector | |
| 1586 | + // Store file is replaced (or kept, on hash match) by the caller's | |
| 1587 | + // sync_upsert_entry after storage succeeds. | |
| 1588 | + self::$vectorstore_mirror_suspended = true; | |
| 1252 | 1589 | $delete_result = self::delete_chunks_for_url($source_url, $bot_id); |
| 1590 | + self::$vectorstore_mirror_suspended = false; | |
| 1253 | 1591 | if (is_wp_error($delete_result)) { |
| 1254 | 1592 | //error_log('[MXCHAT-CHUNK-DEBUG] Warning: Failed to delete existing chunks: ' . $delete_result->get_error_message()); |
| 1255 | 1593 | // Continue anyway - we'll overwrite with upsert |
| 1256 | 1594 | } |
| @@ -1439,9 +1777,22 @@ | ||
| 1439 | 1777 | 'last_updated' => time(), |
| 1440 | 1778 | 'created_at' => time(), |
| 1441 | 1779 | 'bot_id' => $bot_id, |
| 1442 | 1780 | ); |
| 1781 | + $page_title = self::source_title_for($source_url); | |
| 1782 | + if ($page_title !== '') { | |
| 1783 | + $metadata['title'] = $page_title; // page title captured at import (plan d9ee66) | |
| 1784 | + } | |
| 1443 | 1785 | |
| 1786 | + // Document index (plan 362c31): same chunk id and metadata, documents API. | |
| 1787 | + if (class_exists('MxChat_Pinecone_Documents') && MxChat_Pinecone_Documents::is_document_index($bot_id)) { | |
| 1788 | + return MxChat_Pinecone_Documents::upsert_one($vector_id, $chunk_text, $embedding_vector, $metadata, array( | |
| 1789 | + 'api_key' => $api_key, | |
| 1790 | + 'host' => $host, | |
| 1791 | + 'namespace' => $namespace, | |
| 1792 | + )); | |
| 1793 | + } | |
| 1794 | + | |
| 1444 | 1795 | $vector_data = array( |
| 1445 | 1796 | 'id' => $vector_id, |
| 1446 | 1797 | 'values' => $embedding_vector, |
| 1447 | 1798 | 'metadata' => $metadata |
| @@ -1484,20 +1835,25 @@ | ||
| 1484 | 1835 | global $wpdb; |
| 1485 | 1836 | |
| 1486 | 1837 | // For chunks, we always insert new rows (no duplicate checking) |
| 1487 | 1838 | // The URL includes chunk info in the metadata, but source_url stays the same for grouping |
| 1488 | - $result = $wpdb->insert( | |
| 1489 | - $table_name, | |
| 1490 | - array( | |
| 1491 | - 'url' => $source_url, | |
| 1492 | - 'article_content' => $content_with_metadata, | |
| 1493 | - 'embedding_vector' => $embedding_vector_serialized, | |
| 1494 | - 'source_url' => $source_url, | |
| 1495 | - 'content_type' => $content_type, | |
| 1496 | - 'timestamp' => current_time('mysql') | |
| 1497 | - ), | |
| 1498 | - array('%s', '%s', '%s', '%s', '%s', '%s') | |
| 1839 | + $row = array( | |
| 1840 | + 'url' => $source_url, | |
| 1841 | + 'article_content' => $content_with_metadata, | |
| 1842 | + 'embedding_vector' => $embedding_vector_serialized, | |
| 1843 | + 'source_url' => $source_url, | |
| 1844 | + 'content_type' => $content_type, | |
| 1845 | + 'timestamp' => current_time('mysql') | |
| 1499 | 1846 | ); |
| 1847 | + $formats = array('%s', '%s', '%s', '%s', '%s', '%s'); | |
| 1848 | + // Every chunk row carries the page title too (plan d9ee66) - readers group by | |
| 1849 | + // source_url and may pick any chunk as the representative. | |
| 1850 | + $page_title = self::source_title_for($source_url); | |
| 1851 | + if ($page_title !== '' && self::ensure_kb_title_column()) { | |
| 1852 | + $row['title'] = $page_title; | |
| 1853 | + $formats[] = '%s'; | |
| 1854 | + } | |
| 1855 | + $result = $wpdb->insert($table_name, $row, $formats); | |
| 1500 | 1856 | |
| 1501 | 1857 | if ($result === false) { |
| 1502 | 1858 | return new WP_Error('database_failed', 'Failed to insert chunk: ' . $wpdb->last_error); |
| 1503 | 1859 | } |
| @@ -1514,8 +1870,14 @@ | ||
| 1514 | 1870 | */ |
| 1515 | 1871 | public static function delete_chunks_for_url($source_url, $bot_id = 'default') { |
| 1516 | 1872 | //error_log('[MXCHAT-CHUNK-DELETE] Deleting chunks for URL: ' . $source_url); |
| 1517 | 1873 | |
| 1874 | + // Entry removal — mirror it to the Vector Store (unless a storage routine | |
| 1875 | + // is mid-re-store, see $vectorstore_mirror_suspended). | |
| 1876 | + if (!self::$vectorstore_mirror_suspended && class_exists('MxChat_Vectorstore_Manager')) { | |
| 1877 | + MxChat_Vectorstore_Manager::sync_delete_entry($source_url, $bot_id); | |
| 1878 | + } | |
| 1879 | + | |
| 1518 | 1880 | if (self::is_pinecone_enabled_for_bot($bot_id)) { |
| 1519 | 1881 | return self::delete_pinecone_chunks_by_url($source_url, $bot_id); |
| 1520 | 1882 | } else { |
| 1521 | 1883 | return self::delete_wordpress_chunks_by_url($source_url); |
| @@ -1524,10 +1886,16 @@ | ||
| 1524 | 1886 | |
| 1525 | 1887 | /** |
| 1526 | 1888 | * Delete all chunks for a URL from Pinecone |
| 1527 | 1889 | */ |
| 1528 | -private static function delete_pinecone_chunks_by_url($source_url, $bot_id) { | |
| 1529 | - // Get Pinecone configuration | |
| 1890 | +/** | |
| 1891 | + * Delete leftover md5(url)_chunk_N vectors after a URL's content was re-stored | |
| 1892 | + * as a SINGLE vector (content shrank below the chunk threshold on edit/re-import). | |
| 1893 | + * Unlike delete_pinecone_chunks_by_url this leaves the base id alone — the caller | |
| 1894 | + * just upserted the new content there. Failure is logged, not fatal: the save | |
| 1895 | + * itself succeeded, and the next save retries the sweep. | |
| 1896 | + */ | |
| 1897 | +private static function cleanup_pinecone_chunk_stragglers($source_url, $bot_id) { | |
| 1530 | 1898 | if ($bot_id === 'default' || !class_exists('MxChat_Multi_Bot_Manager')) { |
| 1531 | 1899 | $pinecone_options = get_option('mxchat_pinecone_addon_options'); |
| 1532 | 1900 | $api_key = $pinecone_options['mxchat_pinecone_api_key'] ?? ''; |
| 1533 | 1901 | $host = $pinecone_options['mxchat_pinecone_host'] ?? ''; |
| @@ -1546,21 +1914,30 @@ | ||
| 1546 | 1914 | } |
| 1547 | 1915 | } |
| 1548 | 1916 | |
| 1549 | 1917 | if (empty($host) || empty($api_key)) { |
| 1550 | - return new WP_Error('pinecone_config', 'Pinecone is not properly configured'); | |
| 1918 | + return; | |
| 1551 | 1919 | } |
| 1552 | 1920 | |
| 1553 | - $base_vector_id = md5($source_url); | |
| 1554 | - $vectors_to_delete = array(); | |
| 1921 | + // Document index (plan 362c31): list the chunk ids and delete them through the documents API. | |
| 1922 | + if (class_exists('MxChat_Pinecone_Documents') && MxChat_Pinecone_Documents::is_document_index($bot_id)) { | |
| 1923 | + $docs_cfg = array('api_key' => $api_key, 'host' => $host, 'namespace' => $namespace); | |
| 1924 | + $doc_stragglers = MxChat_Pinecone_Documents::list_ids(md5($source_url) . '_chunk_', $docs_cfg); | |
| 1925 | + if (!empty($doc_stragglers)) { | |
| 1926 | + $doc_result = MxChat_Pinecone_Documents::delete_ids($doc_stragglers, $docs_cfg); | |
| 1927 | + if (is_wp_error($doc_result) && class_exists('MxChat_Admin') && method_exists('MxChat_Admin', 'mxchat_log_debug')) { | |
| 1928 | + MxChat_Admin::mxchat_log_debug('pinecone_error', 'Failed to sweep stale chunk documents after single-vector re-store', array('source_url' => $source_url, 'bot_id' => $bot_id, 'count' => count($doc_stragglers))); | |
| 1929 | + } | |
| 1930 | + } | |
| 1931 | + return; | |
| 1932 | + } | |
| 1555 | 1933 | |
| 1556 | - // Add the original single-vector ID (for non-chunked content) | |
| 1557 | - $vectors_to_delete[] = $base_vector_id; | |
| 1934 | + $stragglers = array(); | |
| 1558 | 1935 | |
| 1559 | - // Pinecone /vectors/list is a GET endpoint with query-string parameters; a POST here returns a | |
| 1560 | - // non-200 silently and we end up only deleting the base vector, leaving chunks orphaned. | |
| 1936 | + // Pinecone /vectors/list is a GET endpoint with query-string parameters (a POST | |
| 1937 | + // answers 200-with-an-empty-body, which reads as "no stragglers"). | |
| 1561 | 1938 | $query_params = array( |
| 1562 | - 'prefix' => $base_vector_id . '_chunk_', | |
| 1939 | + 'prefix' => md5($source_url) . '_chunk_', | |
| 1563 | 1940 | 'limit' => 100, |
| 1564 | 1941 | ); |
| 1565 | 1942 | if (!empty($namespace)) { |
| 1566 | 1943 | $query_params['namespace'] = $namespace; |
| @@ -1567,9 +1944,8 @@ | ||
| 1567 | 1944 | } |
| 1568 | 1945 | |
| 1569 | 1946 | $list_url = "https://{$host}/vectors/list?" . http_build_query($query_params); |
| 1570 | 1947 | |
| 1571 | - // Paginate in case a URL has more than 100 chunks. | |
| 1572 | 1948 | do { |
| 1573 | 1949 | $list_response = wp_remote_get($list_url, array( |
| 1574 | 1950 | 'headers' => array( |
| 1575 | 1951 | 'Api-Key' => $api_key, |
| @@ -1585,9 +1961,9 @@ | ||
| 1585 | 1961 | $list_data = json_decode(wp_remote_retrieve_body($list_response), true); |
| 1586 | 1962 | if (!empty($list_data['vectors'])) { |
| 1587 | 1963 | foreach ($list_data['vectors'] as $vector) { |
| 1588 | 1964 | if (isset($vector['id'])) { |
| 1589 | - $vectors_to_delete[] = $vector['id']; | |
| 1965 | + $stragglers[] = $vector['id']; | |
| 1590 | 1966 | } |
| 1591 | 1967 | } |
| 1592 | 1968 | } |
| 1593 | 1969 | |
| @@ -1598,8 +1974,188 @@ | ||
| 1598 | 1974 | |
| 1599 | 1975 | $query_params['paginationToken'] = $next_token; |
| 1600 | 1976 | $list_url = "https://{$host}/vectors/list?" . http_build_query($query_params); |
| 1601 | 1977 | } while (true); |
| 1978 | + | |
| 1979 | + if (empty($stragglers)) { | |
| 1980 | + return; | |
| 1981 | + } | |
| 1982 | + | |
| 1983 | + $delete_body = array('ids' => $stragglers); | |
| 1984 | + if (!empty($namespace)) { | |
| 1985 | + $delete_body['namespace'] = $namespace; | |
| 1986 | + } | |
| 1987 | + | |
| 1988 | + $delete_response = wp_remote_post("https://{$host}/vectors/delete", array( | |
| 1989 | + 'headers' => array( | |
| 1990 | + 'Api-Key' => $api_key, | |
| 1991 | + 'accept' => 'application/json', | |
| 1992 | + 'content-type' => 'application/json' | |
| 1993 | + ), | |
| 1994 | + 'body' => wp_json_encode($delete_body), | |
| 1995 | + 'timeout' => 30 | |
| 1996 | + )); | |
| 1997 | + | |
| 1998 | + if ((is_wp_error($delete_response) || wp_remote_retrieve_response_code($delete_response) !== 200) | |
| 1999 | + && class_exists('MxChat_Admin') && method_exists('MxChat_Admin', 'mxchat_log_debug')) { | |
| 2000 | + MxChat_Admin::mxchat_log_debug('pinecone_error', 'Failed to sweep stale chunk vectors after single-vector re-store', array('source_url' => $source_url, 'bot_id' => $bot_id, 'count' => count($stragglers))); | |
| 2001 | + } | |
| 2002 | +} | |
| 2003 | + | |
| 2004 | +/** | |
| 2005 | + * Resolve the Pinecone credentials for a bot: the bot's own config when the | |
| 2006 | + * multi-bot add-on supplies one, else the site-wide options. Extracted so the | |
| 2007 | + * delete path and the count path can never resolve different credentials. | |
| 2008 | + * | |
| 2009 | + * @param string $bot_id | |
| 2010 | + * @return array{api_key:string,host:string,namespace:string} | |
| 2011 | + */ | |
| 2012 | +private static function pinecone_config_for_bot($bot_id) { | |
| 2013 | + if ($bot_id === 'default' || !class_exists('MxChat_Multi_Bot_Manager')) { | |
| 2014 | + $bot_pinecone_config = array(); | |
| 2015 | + } else { | |
| 2016 | + $bot_pinecone_config = apply_filters('mxchat_get_bot_pinecone_config', array(), $bot_id); | |
| 2017 | + } | |
| 2018 | + | |
| 2019 | + if (empty($bot_pinecone_config)) { | |
| 2020 | + $pinecone_options = get_option('mxchat_pinecone_addon_options'); | |
| 2021 | + return array( | |
| 2022 | + 'api_key' => $pinecone_options['mxchat_pinecone_api_key'] ?? '', | |
| 2023 | + 'host' => $pinecone_options['mxchat_pinecone_host'] ?? '', | |
| 2024 | + 'namespace' => $pinecone_options['mxchat_pinecone_namespace'] ?? '', | |
| 2025 | + // 'vector' | 'document' (plan 362c31) — the delete / count paths branch on it. | |
| 2026 | + 'index_type' => (($pinecone_options['mxchat_pinecone_index_type'] ?? 'vector') === 'document') ? 'document' : 'vector', | |
| 2027 | + ); | |
| 2028 | + } | |
| 2029 | + | |
| 2030 | + return array( | |
| 2031 | + 'api_key' => $bot_pinecone_config['api_key'] ?? '', | |
| 2032 | + 'host' => $bot_pinecone_config['host'] ?? '', | |
| 2033 | + 'namespace' => $bot_pinecone_config['namespace'] ?? '', | |
| 2034 | + 'index_type' => (($bot_pinecone_config['index_type'] ?? 'vector') === 'document') ? 'document' : 'vector', | |
| 2035 | + ); | |
| 2036 | +} | |
| 2037 | + | |
| 2038 | +/** | |
| 2039 | + * List every Pinecone vector id under a prefix, following pagination. | |
| 2040 | + * | |
| 2041 | + * Pinecone /vectors/list is a GET endpoint with query-string parameters; a POST here returns a | |
| 2042 | + * non-200 silently and we end up seeing only the base vector, leaving chunks orphaned. | |
| 2043 | + * | |
| 2044 | + * @param string $prefix Vector-id prefix to match. | |
| 2045 | + * @param array $cfg From pinecone_config_for_bot(). | |
| 2046 | + * @return array List of vector ids (empty when none match or the API is unreachable). | |
| 2047 | + */ | |
| 2048 | +private static function pinecone_list_ids_by_prefix($prefix, array $cfg) { | |
| 2049 | + // Document index (plan 362c31): the documents API lists by prefix too. | |
| 2050 | + if (($cfg['index_type'] ?? 'vector') === 'document' && class_exists('MxChat_Pinecone_Documents')) { | |
| 2051 | + return MxChat_Pinecone_Documents::list_ids($prefix, $cfg); | |
| 2052 | + } | |
| 2053 | + | |
| 2054 | + $ids = array(); | |
| 2055 | + | |
| 2056 | + $query_params = array( | |
| 2057 | + 'prefix' => $prefix, | |
| 2058 | + 'limit' => 100, | |
| 2059 | + ); | |
| 2060 | + if (!empty($cfg['namespace'])) { | |
| 2061 | + $query_params['namespace'] = $cfg['namespace']; | |
| 2062 | + } | |
| 2063 | + | |
| 2064 | + $list_url = "https://{$cfg['host']}/vectors/list?" . http_build_query($query_params); | |
| 2065 | + | |
| 2066 | + // Paginate in case a URL has more than 100 chunks. | |
| 2067 | + do { | |
| 2068 | + $list_response = wp_remote_get($list_url, array( | |
| 2069 | + 'headers' => array( | |
| 2070 | + 'Api-Key' => $cfg['api_key'], | |
| 2071 | + 'accept' => 'application/json', | |
| 2072 | + ), | |
| 2073 | + 'timeout' => 30, | |
| 2074 | + )); | |
| 2075 | + | |
| 2076 | + if (is_wp_error($list_response) || wp_remote_retrieve_response_code($list_response) !== 200) { | |
| 2077 | + break; | |
| 2078 | + } | |
| 2079 | + | |
| 2080 | + $list_data = json_decode(wp_remote_retrieve_body($list_response), true); | |
| 2081 | + if (!empty($list_data['vectors'])) { | |
| 2082 | + foreach ($list_data['vectors'] as $vector) { | |
| 2083 | + if (isset($vector['id'])) { | |
| 2084 | + $ids[] = $vector['id']; | |
| 2085 | + } | |
| 2086 | + } | |
| 2087 | + } | |
| 2088 | + | |
| 2089 | + $next_token = $list_data['pagination']['next'] ?? ''; | |
| 2090 | + if (empty($next_token)) { | |
| 2091 | + break; | |
| 2092 | + } | |
| 2093 | + | |
| 2094 | + $query_params['paginationToken'] = $next_token; | |
| 2095 | + $list_url = "https://{$cfg['host']}/vectors/list?" . http_build_query($query_params); | |
| 2096 | + } while (true); | |
| 2097 | + | |
| 2098 | + return $ids; | |
| 2099 | +} | |
| 2100 | + | |
| 2101 | +/** | |
| 2102 | + * How many stored pieces a URL currently has, in whichever backend holds it. | |
| 2103 | + * | |
| 2104 | + * Branches on the SAME condition delete_chunks_for_url() branches on, so the | |
| 2105 | + * number reported is the number that call removes. Used by the REST DELETE | |
| 2106 | + * route to answer "was anything actually there?" without a second delete path. | |
| 2107 | + * | |
| 2108 | + * @param string $source_url | |
| 2109 | + * @param string $bot_id | |
| 2110 | + * @return int | |
| 2111 | + */ | |
| 2112 | +public static function count_chunks_for_url($source_url, $bot_id = 'default') { | |
| 2113 | + if (self::is_pinecone_enabled_for_bot($bot_id)) { | |
| 2114 | + $cfg = self::pinecone_config_for_bot($bot_id); | |
| 2115 | + if (empty($cfg['host']) || empty($cfg['api_key'])) { | |
| 2116 | + return 0; | |
| 2117 | + } | |
| 2118 | + // The base id and every "<base>_chunk_N" share the base as a prefix, | |
| 2119 | + // so one prefix listing covers chunked and non-chunked content alike. | |
| 2120 | + return count(self::pinecone_list_ids_by_prefix(md5($source_url), $cfg)); | |
| 2121 | + } | |
| 2122 | + | |
| 2123 | + global $wpdb; | |
| 2124 | + $table_name = $wpdb->prefix . 'mxchat_system_prompt_content'; | |
| 2125 | + | |
| 2126 | + return (int) $wpdb->get_var($wpdb->prepare( | |
| 2127 | + "SELECT COUNT(*) FROM {$table_name} WHERE source_url = %s", | |
| 2128 | + $source_url | |
| 2129 | + )); | |
| 2130 | +} | |
| 2131 | + | |
| 2132 | +private static function delete_pinecone_chunks_by_url($source_url, $bot_id) { | |
| 2133 | + $cfg = self::pinecone_config_for_bot($bot_id); | |
| 2134 | + $api_key = $cfg['api_key']; | |
| 2135 | + $host = $cfg['host']; | |
| 2136 | + $namespace = $cfg['namespace']; | |
| 2137 | + | |
| 2138 | + if (empty($host) || empty($api_key)) { | |
| 2139 | + return new WP_Error('pinecone_config', 'Pinecone is not properly configured'); | |
| 2140 | + } | |
| 2141 | + | |
| 2142 | + // Document index (plan 362c31): base id + every <md5>_chunk_N, documents API. | |
| 2143 | + if (($cfg['index_type'] ?? 'vector') === 'document' && class_exists('MxChat_Pinecone_Documents')) { | |
| 2144 | + $doc_ids = array_merge(array(md5($source_url)), MxChat_Pinecone_Documents::list_ids(md5($source_url) . '_chunk_', $cfg)); | |
| 2145 | + return MxChat_Pinecone_Documents::delete_ids($doc_ids, $cfg); | |
| 2146 | + } | |
| 2147 | + | |
| 2148 | + $base_vector_id = md5($source_url); | |
| 2149 | + $vectors_to_delete = array(); | |
| 2150 | + | |
| 2151 | + // Add the original single-vector ID (for non-chunked content) | |
| 2152 | + $vectors_to_delete[] = $base_vector_id; | |
| 2153 | + | |
| 2154 | + $vectors_to_delete = array_merge( | |
| 2155 | + $vectors_to_delete, | |
| 2156 | + self::pinecone_list_ids_by_prefix($base_vector_id . '_chunk_', $cfg) | |
| 2157 | + ); | |
| 1602 | 2158 | |
| 1603 | 2159 | if (empty($vectors_to_delete)) { |
| 1604 | 2160 | //error_log('[MXCHAT-CHUNK-DELETE] No vectors found to delete'); |
| 1605 | 2161 | return true; |
