| @@ -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; |
| @@ -339,8 +457,100 @@ | ||
| 339 | 457 | return $options['embedding_model'] ?? ''; |
| 340 | 458 | } |
| 341 | 459 | |
| 342 | 460 | /** |
| 461 | + * Dimension of a STANDARD embedding model, from the dropdown and its two | |
| 462 | + * adjustable-size settings. Says nothing about a Custom Provider model — ask | |
| 463 | + * expected_embedding_dimension() for the model the site actually embeds with. | |
| 464 | + */ | |
| 465 | +public static function standard_embedding_dimension($options = null) { | |
| 466 | + if (!is_array($options)) { | |
| 467 | + $options = get_option('mxchat_options', array()); | |
| 468 | + $options = is_array($options) ? $options : array(); | |
| 469 | + } | |
| 470 | + $selected_model = $options['embedding_model'] ?? 'text-embedding-ada-002'; | |
| 471 | + $model_dimensions = array( | |
| 472 | + 'text-embedding-ada-002' => 1536, | |
| 473 | + 'text-embedding-3-small' => 1536, | |
| 474 | + 'text-embedding-3-large' => 3072, | |
| 475 | + 'voyage-2' => 1024, | |
| 476 | + 'voyage-large-2' => 1536, | |
| 477 | + 'voyage-3-large' => 2048, | |
| 478 | + 'gemini-embedding-001' => 1536, | |
| 479 | + ); | |
| 480 | + if (strpos($selected_model, 'voyage-3-large') === 0) { | |
| 481 | + return intval($options['voyage_output_dimension'] ?? 2048); | |
| 482 | + } | |
| 483 | + if (strpos($selected_model, 'gemini-embedding') === 0) { | |
| 484 | + return intval($options['gemini_output_dimension'] ?? 1536); | |
| 485 | + } | |
| 486 | + return $model_dimensions[$selected_model] ?? 1536; | |
| 487 | +} | |
| 488 | + | |
| 489 | +/** | |
| 490 | + * Dimension of the vectors the NEXT embed will produce — the one place that | |
| 491 | + * answers it (plan 418afd). With Custom Provider embeddings on, the standard | |
| 492 | + * dropdown is disabled and says nothing about the custom model, so the number | |
| 493 | + * is read from the model itself: one short probe embed, remembered per base | |
| 494 | + * URL + model. Three copies of the dropdown table used to answer instead, and | |
| 495 | + * a custom model of another size got a wrong-sized index and wrong-sized | |
| 496 | + * listing vectors. | |
| 497 | + * | |
| 498 | + * @param array|null $options mxchat_options (or a bot's merged options). | |
| 499 | + * @param bool $force Probe again even when a recent probe failed. | |
| 500 | + * @return int|WP_Error WP_Error only when a custom model's size could not be read. | |
| 501 | + */ | |
| 502 | +public static function expected_embedding_dimension($options = null, $force = false) { | |
| 503 | + if (!is_array($options)) { | |
| 504 | + $options = get_option('mxchat_options', array()); | |
| 505 | + $options = is_array($options) ? $options : array(); | |
| 506 | + } | |
| 507 | + if (isset($options['custom_provider_for_embeddings']) && $options['custom_provider_for_embeddings'] === 'on') { | |
| 508 | + return self::custom_embedding_dimension($options, $force); | |
| 509 | + } | |
| 510 | + return self::standard_embedding_dimension($options); | |
| 511 | +} | |
| 512 | + | |
| 513 | +/** | |
| 514 | + * A Custom Provider embedding model's dimension, by asking it once. The | |
| 515 | + * answer is stored against the base URL + model, so changing either asks | |
| 516 | + * again. A failure is remembered for five minutes so a dead endpoint does not | |
| 517 | + * hold up every admin page that prints the number; $force skips that memory. | |
| 518 | + * | |
| 519 | + * @return int|WP_Error | |
| 520 | + */ | |
| 521 | +public static function custom_embedding_dimension($options, $force = false) { | |
| 522 | + $model = self::resolve_custom_embedding_model($options); | |
| 523 | + $base_url = isset($options['custom_provider_base_url']) ? rtrim(trim((string) $options['custom_provider_base_url']), '/') : ''; | |
| 524 | + $key = md5(strtolower($base_url) . '|' . $model); | |
| 525 | + | |
| 526 | + $known = get_option('mxchat_custom_embedding_dimension', array()); | |
| 527 | + if (is_array($known) && ($known['key'] ?? '') === $key && (int) ($known['dimension'] ?? 0) > 0) { | |
| 528 | + return (int) $known['dimension']; | |
| 529 | + } | |
| 530 | + | |
| 531 | + $failed = get_transient('mxchat_custom_embedding_dimension_failed'); | |
| 532 | + if (!$force && is_array($failed) && ($failed['key'] ?? '') === $key) { | |
| 533 | + return new WP_Error('embedding_dimension_unknown', (string) ($failed['reason'] ?? '')); | |
| 534 | + } | |
| 535 | + | |
| 536 | + $vector = self::generate_embedding_custom('MxChat embedding dimension check', $options); | |
| 537 | + if (!is_array($vector) || count($vector) === 0) { | |
| 538 | + $reason = (is_string($vector) && $vector !== '') ? $vector : __('The custom provider returned no embedding.', 'mxchat'); | |
| 539 | + set_transient('mxchat_custom_embedding_dimension_failed', array('key' => $key, 'reason' => $reason), 5 * MINUTE_IN_SECONDS); | |
| 540 | + return new WP_Error('embedding_dimension_unknown', $reason); | |
| 541 | + } | |
| 542 | + | |
| 543 | + update_option('mxchat_custom_embedding_dimension', array( | |
| 544 | + 'key' => $key, | |
| 545 | + 'model' => 'custom:' . $model, | |
| 546 | + 'dimension' => count($vector), | |
| 547 | + ), false); | |
| 548 | + delete_transient('mxchat_custom_embedding_dimension_failed'); | |
| 549 | + return count($vector); | |
| 550 | +} | |
| 551 | + | |
| 552 | +/** | |
| 343 | 553 | * Extract the 11-character YouTube video ID from a URL, or '' if the URL is |
| 344 | 554 | * not a single-video YouTube link. Single source of truth for both the KB |
| 345 | 555 | * ingestion side and the chat render side — do not duplicate this parsing. |
| 346 | 556 | * Channel, playlist, and search URLs deliberately return '' (only a URL that |
| @@ -417,9 +627,9 @@ | ||
| 417 | 627 | * @param string $bot_id The bot ID for multi-bot support |
| 418 | 628 | * @param string $content_type The type of content (post, page, pdf, url, manual, product, etc.) |
| 419 | 629 | * @return bool|WP_Error True on success, WP_Error on failure |
| 420 | 630 | */ |
| 421 | -public static function submit_content_to_db($content, $source_url, $api_key, $vector_id = null, $bot_id = 'default', $content_type = 'content') { | |
| 631 | +public static function submit_content_to_db($content, $source_url, $api_key, $vector_id = null, $bot_id = 'default', $content_type = 'content', $title = '') { | |
| 422 | 632 | global $wpdb; |
| 423 | 633 | $table_name = $wpdb->prefix . 'mxchat_system_prompt_content'; |
| 424 | 634 | |
| 425 | 635 | //error_log('[MXCHAT-DB] Starting database submission for URL: ' . $source_url . ' (Bot: ' . $bot_id . ', Type: ' . $content_type . ')'); |
| @@ -424,10 +634,18 @@ | ||
| 424 | 634 | |
| 425 | 635 | //error_log('[MXCHAT-DB] Starting database submission for URL: ' . $source_url . ' (Bot: ' . $bot_id . ', Type: ' . $content_type . ')'); |
| 426 | 636 | //error_log('[MXCHAT-DB] Content length: ' . strlen($content) . ' bytes'); |
| 427 | 637 | |
| 428 | - // Sanitize the source URL | |
| 429 | - $source_url = esc_url_raw($source_url); | |
| 638 | + // Sanitize the source URL. Internal identities (mxchat:// manual docs, | |
| 639 | + // upload:// file uploads) are NOT web URLs — esc_url_raw EMPTIES them | |
| 640 | + // because its protocol list is statically cached and effectively | |
| 641 | + // unfilterable (measured, plan 945406 / 0485e5) — sanitize those as text | |
| 642 | + // so upserts stay keyed to a stable identity across re-imports. | |
| 643 | + if (preg_match('#^(mxchat|upload)://#i', $source_url)) { | |
| 644 | + $source_url = sanitize_text_field($source_url); | |
| 645 | + } else { | |
| 646 | + $source_url = esc_url_raw($source_url); | |
| 647 | + } | |
| 430 | 648 | |
| 431 | 649 | // Sanitize content_type |
| 432 | 650 | $content_type = sanitize_key($content_type); |
| 433 | 651 | if (empty($content_type)) { |
| @@ -433,8 +651,14 @@ | ||
| 433 | 651 | if (empty($content_type)) { |
| 434 | 652 | $content_type = 'content'; // Fallback for backwards compatibility |
| 435 | 653 | } |
| 436 | 654 | |
| 655 | + // Page title captured by the importer (plan d9ee66), registered under the | |
| 656 | + // SANITIZED url so the storage writers below find it by the same key. | |
| 657 | + if (is_string($title) && trim($title) !== '') { | |
| 658 | + self::set_source_title($source_url, $title); | |
| 659 | + } | |
| 660 | + | |
| 437 | 661 | // Just ensure UTF-8 validity without aggressive escaping |
| 438 | 662 | $safe_content = wp_check_invalid_utf8($content); |
| 439 | 663 | // Remove only null bytes and other control characters, but preserve newlines (\n = \x0A) and carriage returns (\r = \x0D) |
| 440 | 664 | $safe_content = preg_replace('/[\x00-\x08\x0B\x0C\x0E-\x1F\x7F]/u', '', $safe_content); |
| @@ -442,9 +666,16 @@ | ||
| 442 | 666 | // Check if chunking should be applied |
| 443 | 667 | $chunker = MxChat_Chunker::from_settings(); |
| 444 | 668 | if ($chunker->should_chunk($safe_content)) { |
| 445 | 669 | //error_log('[MXCHAT-DB] Content exceeds chunk threshold, using chunked submission'); |
| 446 | - return self::submit_chunked_content($safe_content, $source_url, $api_key, $bot_id, $content_type, $chunker); | |
| 670 | + $chunk_result = self::submit_chunked_content($safe_content, $source_url, $api_key, $bot_id, $content_type, $chunker); | |
| 671 | + // Vector Store mirror gets the entry WHOLE — one file per KB entry, | |
| 672 | + // OpenAI chunks server-side; local chunking is our own embedding | |
| 673 | + // concern and never reaches the store (plan 15b5c6). | |
| 674 | + if ($chunk_result === true && class_exists('MxChat_Vectorstore_Manager')) { | |
| 675 | + MxChat_Vectorstore_Manager::sync_upsert_entry($source_url, $safe_content, $bot_id, $content_type); | |
| 676 | + } | |
| 677 | + return $chunk_result; | |
| 447 | 678 | } |
| 448 | 679 | |
| 449 | 680 | // UPDATED: Generate the embedding using bot-specific configuration |
| 450 | 681 | $embedding_vector = self::generate_embedding($content, $api_key, $bot_id); |
| @@ -462,18 +693,175 @@ | ||
| 462 | 693 | // UPDATED: Check if Pinecone is enabled for this specific bot |
| 463 | 694 | if (self::is_pinecone_enabled_for_bot($bot_id)) { |
| 464 | 695 | //error_log('[MXCHAT-DB] Pinecone is enabled for bot ' . $bot_id . ' - using Pinecone storage'); |
| 465 | 696 | // Store in Pinecone only |
| 466 | - return self::store_in_pinecone_only($embedding_vector, $content, $source_url, $vector_id, $bot_id, $content_type); | |
| 697 | + $result = self::store_in_pinecone_only($embedding_vector, $content, $source_url, $vector_id, $bot_id, $content_type); | |
| 698 | + // If this URL previously stored as CHUNKED content and the new content fits in a | |
| 699 | + // single vector, the upsert above only overwrote the base id — the old | |
| 700 | + // md5(url)_chunk_N vectors would keep serving the stale text. Sweep them. | |
| 701 | + // Only for a real source_url: chunk ids derive from it, and md5('') is shared | |
| 702 | + // by legacy URL-less entries so a blind sweep there could hit other entries. | |
| 703 | + if ($result === true && !empty($source_url)) { | |
| 704 | + self::cleanup_pinecone_chunk_stragglers($source_url, $bot_id); | |
| 705 | + } | |
| 706 | + if ($result === true && class_exists('MxChat_Vectorstore_Manager')) { | |
| 707 | + MxChat_Vectorstore_Manager::sync_upsert_entry($source_url, $safe_content, $bot_id, $content_type); | |
| 708 | + } | |
| 709 | + return $result; | |
| 467 | 710 | } else { |
| 468 | 711 | //error_log('[MXCHAT-DB] Pinecone not enabled for bot ' . $bot_id . ' - using WordPress storage'); |
| 469 | 712 | // Store in WordPress database only |
| 470 | 713 | $embedding_vector_serialized = maybe_serialize($embedding_vector); |
| 471 | - return self::store_in_wordpress_db($safe_content, $source_url, $embedding_vector_serialized, $table_name, $content_type); | |
| 714 | + $result = self::store_in_wordpress_db($safe_content, $source_url, $embedding_vector_serialized, $table_name, $content_type); | |
| 715 | + if ($result === true && class_exists('MxChat_Vectorstore_Manager')) { | |
| 716 | + MxChat_Vectorstore_Manager::sync_upsert_entry($source_url, $safe_content, $bot_id, $content_type); | |
| 717 | + } | |
| 718 | + return $result; | |
| 472 | 719 | } |
| 473 | 720 | } |
| 474 | 721 | |
| 722 | +/* ------------------------------------------------------------------------- | |
| 723 | + * Source titles (3.2.23, plan d9ee66) | |
| 724 | + * | |
| 725 | + * A knowledge entry imported from a URL or sitemap remembers the page's own | |
| 726 | + * title, so an add-on that names a source (AI Search's source cards) can show | |
| 727 | + * "About Us" instead of a guess from the slug ("Aboutus"). Stored in the | |
| 728 | + * `title` column of the WP knowledge table and as `title` metadata on Pinecone | |
| 729 | + * vectors / documents. The importer hands the title to submit_content_to_db(), | |
| 730 | + * which registers it under the source URL; the writers look it up by that key. | |
| 731 | + * ---------------------------------------------------------------------- */ | |
| 732 | + | |
| 733 | +/** @var array<string,string> source_url => title, for the current request */ | |
| 734 | +private static $source_titles = array(); | |
| 735 | + | |
| 736 | +public static function set_source_title($source_url, $title) { | |
| 737 | + $key = trim((string) $source_url); | |
| 738 | + $title = self::clean_title($title); | |
| 739 | + if ($key === '' || $title === '') { | |
| 740 | + return; | |
| 741 | + } | |
| 742 | + self::$source_titles[$key] = $title; | |
| 743 | +} | |
| 744 | + | |
| 745 | +public static function source_title_for($source_url) { | |
| 746 | + $key = trim((string) $source_url); | |
| 747 | + return ($key !== '' && isset(self::$source_titles[$key])) ? self::$source_titles[$key] : ''; | |
| 748 | +} | |
| 749 | + | |
| 750 | +/** Entities decoded, tags gone, whitespace collapsed, capped at the column width. */ | |
| 751 | +public static function clean_title($title) { | |
| 752 | + $title = html_entity_decode((string) $title, ENT_QUOTES | ENT_HTML5, 'UTF-8'); | |
| 753 | + $title = wp_strip_all_tags($title); | |
| 754 | + $title = trim((string) preg_replace('/\s+/u', ' ', $title)); | |
| 755 | + if (mb_strlen($title, 'UTF-8') > 255) { | |
| 756 | + $title = rtrim(mb_substr($title, 0, 255, 'UTF-8')); | |
| 757 | + } | |
| 758 | + return $title; | |
| 759 | +} | |
| 760 | + | |
| 475 | 761 | /** |
| 762 | + * The page's own title from fetched HTML: og:title, else <title> with a | |
| 763 | + * trailing "| Site Name" style suffix removed, else the first <h1>. Returns '' | |
| 764 | + * when nothing usable is there, so the caller keeps its own fallback. | |
| 765 | + */ | |
| 766 | +public static function extract_page_title($html) { | |
| 767 | + $html = (string) $html; | |
| 768 | + if ($html === '') { | |
| 769 | + return ''; | |
| 770 | + } | |
| 771 | + $head = substr($html, 0, 200000); | |
| 772 | + $site = ''; | |
| 773 | + if (preg_match('/<meta\s+[^>]*property=["\']og:site_name["\'][^>]*content=["\']([^"\']{1,120})["\']/i', $head, $m) | |
| 774 | + || preg_match('/<meta\s+[^>]*content=["\']([^"\']{1,120})["\'][^>]*property=["\']og:site_name["\']/i', $head, $m)) { | |
| 775 | + $site = self::clean_title($m[1]); | |
| 776 | + } | |
| 777 | + $candidate = ''; | |
| 778 | + if (preg_match('/<meta\s+[^>]*property=["\']og:title["\'][^>]*content=["\']([^"\']{1,500})["\']/i', $head, $m) | |
| 779 | + || preg_match('/<meta\s+[^>]*content=["\']([^"\']{1,500})["\'][^>]*property=["\']og:title["\']/i', $head, $m)) { | |
| 780 | + $candidate = self::clean_title($m[1]); | |
| 781 | + } | |
| 782 | + if ($candidate === '' && preg_match('/<title[^>]*>(.*?)<\/title>/is', $head, $m)) { | |
| 783 | + $candidate = self::clean_title($m[1]); | |
| 784 | + } | |
| 785 | + if ($candidate !== '') { | |
| 786 | + $candidate = self::strip_site_suffix($candidate, $site); | |
| 787 | + } | |
| 788 | + if ($candidate === '' && preg_match('/<h1[^>]*>(.*?)<\/h1>/is', $html, $m)) { | |
| 789 | + $candidate = self::clean_title($m[1]); | |
| 790 | + } | |
| 791 | + return $candidate; | |
| 792 | +} | |
| 793 | + | |
| 794 | +/** | |
| 795 | + * "About Us | Just Facts" -> "About Us". Only a TRAILING segment after a spaced | |
| 796 | + * separator goes, only when what is left is a real title (3+ characters) and the | |
| 797 | + * removed part is short enough to be a site name; a known og:site_name wins. | |
| 798 | + * Byte offsets throughout: every separator starts with an ASCII space, so a cut | |
| 799 | + * there never lands inside a multibyte character. | |
| 800 | + */ | |
| 801 | +private static function strip_site_suffix($title, $site_name = '') { | |
| 802 | + $seps = array(' | ', ' – ', ' — ', ' - ', ' :: ', ' · ', ' » ', ' • '); | |
| 803 | + if ($site_name !== '') { | |
| 804 | + foreach ($seps as $sep) { | |
| 805 | + $tail = $sep . $site_name; | |
| 806 | + if (strlen($title) > strlen($tail) && substr($title, -strlen($tail)) === $tail) { | |
| 807 | + return trim(substr($title, 0, strlen($title) - strlen($tail))); | |
| 808 | + } | |
| 809 | + } | |
| 810 | + } | |
| 811 | + $best_pos = -1; | |
| 812 | + $best_sep = ''; | |
| 813 | + foreach ($seps as $sep) { | |
| 814 | + $pos = strrpos($title, $sep); | |
| 815 | + if ($pos !== false && $pos > $best_pos) { | |
| 816 | + $best_pos = $pos; | |
| 817 | + $best_sep = $sep; | |
| 818 | + } | |
| 819 | + } | |
| 820 | + if ($best_pos > 0) { | |
| 821 | + $left = trim(substr($title, 0, $best_pos)); | |
| 822 | + $right = trim(substr($title, $best_pos + strlen($best_sep))); | |
| 823 | + if (mb_strlen($left, 'UTF-8') >= 3 && $right !== '' && mb_strlen($right, 'UTF-8') <= 60) { | |
| 824 | + return $left; | |
| 825 | + } | |
| 826 | + } | |
| 827 | + return $title; | |
| 828 | +} | |
| 829 | + | |
| 830 | +/** | |
| 831 | + * Make sure the knowledge table has the `title` column (3.2.23, plan d9ee66). | |
| 832 | + * Activation adds it; this is the safety net for a copy updated in place, and | |
| 833 | + * the one place add-ons ask "may I read/write title?". A positive answer is | |
| 834 | + * remembered in an option, so after the first time this costs one option read. | |
| 835 | + */ | |
| 836 | +public static function ensure_kb_title_column() { | |
| 837 | + global $wpdb; | |
| 838 | + static $ready = null; | |
| 839 | + if ($ready !== null) { | |
| 840 | + return $ready; | |
| 841 | + } | |
| 842 | + if (get_option('mxchat_kb_title_column') === '1') { | |
| 843 | + $ready = true; | |
| 844 | + return true; | |
| 845 | + } | |
| 846 | + $table = $wpdb->prefix . 'mxchat_system_prompt_content'; | |
| 847 | + $cols = $wpdb->get_col("SHOW COLUMNS FROM {$table}", 0); | |
| 848 | + if (empty($cols)) { | |
| 849 | + $ready = false; | |
| 850 | + return false; | |
| 851 | + } | |
| 852 | + if (!in_array('title', $cols, true)) { | |
| 853 | + $wpdb->query("ALTER TABLE {$table} ADD COLUMN title VARCHAR(255) DEFAULT NULL AFTER source_url"); | |
| 854 | + $cols = $wpdb->get_col("SHOW COLUMNS FROM {$table}", 0); | |
| 855 | + } | |
| 856 | + $ready = in_array('title', $cols, true); | |
| 857 | + if ($ready) { | |
| 858 | + update_option('mxchat_kb_title_column', '1', false); | |
| 859 | + } | |
| 860 | + return $ready; | |
| 861 | +} | |
| 862 | + | |
| 863 | +/** | |
| 476 | 864 | * UPDATED: Check if Pinecone is enabled and properly configured for a specific bot |
| 477 | 865 | */ |
| 478 | 866 | private static function is_pinecone_enabled_for_bot($bot_id = 'default') { |
| 479 | 867 | // For default bot or when multi-bot is not active, use original method |
| @@ -585,11 +973,15 @@ | ||
| 585 | 973 | // Check if this is truly manual content (no URL at all) vs a real URL that filter_var rejects |
| 586 | 974 | // filter_var(FILTER_VALIDATE_URL) rejects valid URLs with encoded chars, non-ASCII, fragments, etc. |
| 587 | 975 | // Use a looser check: if it starts with http(s):// or has a scheme, it's a URL |
| 588 | 976 | $has_url_scheme = !empty($source_url) && preg_match('#^https?://#i', $source_url); |
| 977 | + // upload:// identities (admin document/PDF uploads, plan 0485e5) are stable | |
| 978 | + // and deduplicable — treat them like URLs so a re-upload UPDATES the row | |
| 979 | + // instead of minting a fresh manual identity (which would duplicate). | |
| 980 | + $has_stable_identity = $has_url_scheme || (!empty($source_url) && preg_match('#^upload://#i', $source_url)); | |
| 589 | 981 | // Treat legacy mxchat.ai source URLs as manual — old bug assigned the site URL to manual entries |
| 590 | 982 | $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; | |
| 983 | + $is_manual_content = empty($source_url) || $source_url === '' || !$has_stable_identity || $is_legacy_mxchat_url; | |
| 592 | 984 | |
| 593 | 985 | if ($is_manual_content) { |
| 594 | 986 | // Generate unique identifier for manual content to prevent overwrites |
| 595 | 987 | $source_url = 'mxchat://manual-content/' . time() . '-' . wp_generate_password(8, false); |
| @@ -610,8 +1002,13 @@ | ||
| 610 | 1002 | //error_log('[MXCHAT-DB] Manual content - will create new entry (no duplicate check)'); |
| 611 | 1003 | } |
| 612 | 1004 | // ===== END FIX ===== |
| 613 | 1005 | |
| 1006 | + // Page title captured at import (plan d9ee66): written only when the | |
| 1007 | + // importer supplied one, so a re-index without a title keeps the old value. | |
| 1008 | + $title_value = $is_manual_content ? '' : self::source_title_for($source_url); | |
| 1009 | + $title_ready = ($title_value !== '') && self::ensure_kb_title_column(); | |
| 1010 | + | |
| 614 | 1011 | // Progressive fallback mechanism for problematic content |
| 615 | 1012 | $attempt = 1; |
| 616 | 1013 | $max_attempts = 3; |
| 617 | 1014 | $current_content = $safe_content; |
| @@ -622,39 +1019,41 @@ | ||
| 622 | 1019 | if ($existing_id) { |
| 623 | 1020 | //error_log('[MXCHAT-DB] Found existing entry (ID: ' . $existing_id . '). Updating... (Attempt ' . $attempt . ')'); |
| 624 | 1021 | |
| 625 | 1022 | // 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') | |
| 1023 | + $row = array( | |
| 1024 | + 'url' => $source_url, | |
| 1025 | + 'article_content' => $current_content, | |
| 1026 | + 'embedding_vector' => $embedding_vector_serialized, | |
| 1027 | + 'source_url' => $source_url, | |
| 1028 | + 'content_type' => $content_type, | |
| 1029 | + 'timestamp' => current_time('mysql'), | |
| 639 | 1030 | ); |
| 1031 | + $formats = array('%s','%s','%s','%s','%s','%s'); | |
| 1032 | + if ($title_ready) { | |
| 1033 | + $row['title'] = $title_value; | |
| 1034 | + $formats[] = '%s'; | |
| 1035 | + } | |
| 1036 | + $result = $wpdb->update($table_name, $row, array('id' => $existing_id), $formats, array('%d')); | |
| 640 | 1037 | } else { |
| 641 | 1038 | //error_log('[MXCHAT-DB] No existing entry found. Inserting new row... (Attempt ' . $attempt . ')'); |
| 642 | 1039 | //error_log('[MXCHAT-DB] Content sample: ' . substr($current_content, 0, 1000)); |
| 643 | 1040 | |
| 644 | 1041 | // 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') | |
| 1042 | + $row = array( | |
| 1043 | + 'url' => $source_url, // Now unique for manual content | |
| 1044 | + 'article_content' => $current_content, | |
| 1045 | + 'embedding_vector' => $embedding_vector_serialized, | |
| 1046 | + 'source_url' => $source_url, // Now unique for manual content | |
| 1047 | + 'content_type' => $content_type, | |
| 1048 | + 'timestamp' => current_time('mysql'), | |
| 656 | 1049 | ); |
| 1050 | + $formats = array('%s','%s','%s','%s','%s','%s'); | |
| 1051 | + if ($title_ready) { | |
| 1052 | + $row['title'] = $title_value; | |
| 1053 | + $formats[] = '%s'; | |
| 1054 | + } | |
| 1055 | + $result = $wpdb->insert($table_name, $row, $formats); | |
| 657 | 1056 | } |
| 658 | 1057 | |
| 659 | 1058 | if ($result === false) { |
| 660 | 1059 | //error_log('[MXCHAT-DB] Database operation failed (Attempt ' . $attempt . ')'); |
| @@ -713,10 +1112,11 @@ | ||
| 713 | 1112 | // ===== UPDATED: Handle manual content with unique vector IDs ===== |
| 714 | 1113 | if ($vector_id) { |
| 715 | 1114 | // Use provided vector ID |
| 716 | 1115 | //error_log('[MXCHAT-PINECONE-MAIN] Using provided vector ID: ' . $vector_id); |
| 717 | - } elseif (!empty($url) && preg_match('#^https?://#i', $url)) { | |
| 718 | - // For URLs, use URL-based ID (existing behavior) | |
| 1116 | + } elseif (!empty($url) && preg_match('#^(https?|upload)://#i', $url)) { | |
| 1117 | + // For URLs — and stable upload:// identities (plan 0485e5) — use an | |
| 1118 | + // identity-derived ID so a re-import upserts the same vector. | |
| 719 | 1119 | $vector_id = md5($url); |
| 720 | 1120 | //error_log('[MXCHAT-PINECONE-MAIN] Generated vector ID from URL: ' . $vector_id); |
| 721 | 1121 | } else { |
| 722 | 1122 | // For manual content (empty/no URL scheme), generate unique ID |
| @@ -777,9 +1177,23 @@ | ||
| 777 | 1177 | 'last_updated' => time(), |
| 778 | 1178 | 'created_at' => time(), // Add creation timestamp |
| 779 | 1179 | 'bot_id' => $bot_id, // Add bot identification |
| 780 | 1180 | ); |
| 781 | - | |
| 1181 | + $page_title = self::source_title_for($url); | |
| 1182 | + if ($page_title !== '') { | |
| 1183 | + $metadata['title'] = $page_title; // page title captured at import (plan d9ee66) | |
| 1184 | + } | |
| 1185 | + | |
| 1186 | + // Document index (plan 362c31): same id, same metadata, through the | |
| 1187 | + // documents API. A site on the classic vector index never enters here. | |
| 1188 | + if (class_exists('MxChat_Pinecone_Documents') && MxChat_Pinecone_Documents::is_document_index($bot_id)) { | |
| 1189 | + return MxChat_Pinecone_Documents::upsert_one($vector_id, $content, $embedding_vector, $metadata, array( | |
| 1190 | + 'api_key' => $api_key, | |
| 1191 | + 'host' => $host, | |
| 1192 | + 'namespace' => $namespace, | |
| 1193 | + )); | |
| 1194 | + } | |
| 1195 | + | |
| 782 | 1196 | $vector_data = array( |
| 783 | 1197 | 'id' => $vector_id, |
| 784 | 1198 | 'values' => $embedding_vector, |
| 785 | 1199 | 'metadata' => $metadata |
| @@ -1247,10 +1661,26 @@ | ||
| 1247 | 1661 | |
| 1248 | 1662 | //error_log('[MXCHAT-CHUNK-DEBUG] Starting chunked submission for: ' . $source_url); |
| 1249 | 1663 | //error_log('[MXCHAT-CHUNK-DEBUG] Content length: ' . strlen($content) . ' chars'); |
| 1250 | 1664 | |
| 1251 | - // First, delete any existing chunks for this URL (clean slate) | |
| 1665 | + // URL-less (manual) content needs a minted identity BEFORE chunk ids are derived: | |
| 1666 | + // every chunk id is md5(source_url)_chunk_N, so with source_url = '' EVERY long | |
| 1667 | + // manual document shared the md5('') prefix — and the clean-slate delete below | |
| 1668 | + // wiped the PREVIOUS manual entry's chunks each time a new one was added. The | |
| 1669 | + // single-vector paths already mint (mxchat:// in WP, manual_* in Pinecone); this | |
| 1670 | + // was the one storage path that didn't. Keep an mxchat:// identity if the caller | |
| 1671 | + // already carries one. | |
| 1672 | + if (strpos($source_url, 'mxchat://') !== 0 && strpos($source_url, 'upload://') !== 0 && !preg_match('#^https?://#i', $source_url)) { | |
| 1673 | + $source_url = 'mxchat://manual-content/' . time() . '-' . wp_generate_password(8, false); | |
| 1674 | + } | |
| 1675 | + | |
| 1676 | + // First, delete any existing chunks for this URL (clean slate). | |
| 1677 | + // Mirror-suspended: this is a re-store, not an entry removal — the Vector | |
| 1678 | + // Store file is replaced (or kept, on hash match) by the caller's | |
| 1679 | + // sync_upsert_entry after storage succeeds. | |
| 1680 | + self::$vectorstore_mirror_suspended = true; | |
| 1252 | 1681 | $delete_result = self::delete_chunks_for_url($source_url, $bot_id); |
| 1682 | + self::$vectorstore_mirror_suspended = false; | |
| 1253 | 1683 | if (is_wp_error($delete_result)) { |
| 1254 | 1684 | //error_log('[MXCHAT-CHUNK-DEBUG] Warning: Failed to delete existing chunks: ' . $delete_result->get_error_message()); |
| 1255 | 1685 | // Continue anyway - we'll overwrite with upsert |
| 1256 | 1686 | } |
| @@ -1439,9 +1869,22 @@ | ||
| 1439 | 1869 | 'last_updated' => time(), |
| 1440 | 1870 | 'created_at' => time(), |
| 1441 | 1871 | 'bot_id' => $bot_id, |
| 1442 | 1872 | ); |
| 1873 | + $page_title = self::source_title_for($source_url); | |
| 1874 | + if ($page_title !== '') { | |
| 1875 | + $metadata['title'] = $page_title; // page title captured at import (plan d9ee66) | |
| 1876 | + } | |
| 1443 | 1877 | |
| 1878 | + // Document index (plan 362c31): same chunk id and metadata, documents API. | |
| 1879 | + if (class_exists('MxChat_Pinecone_Documents') && MxChat_Pinecone_Documents::is_document_index($bot_id)) { | |
| 1880 | + return MxChat_Pinecone_Documents::upsert_one($vector_id, $chunk_text, $embedding_vector, $metadata, array( | |
| 1881 | + 'api_key' => $api_key, | |
| 1882 | + 'host' => $host, | |
| 1883 | + 'namespace' => $namespace, | |
| 1884 | + )); | |
| 1885 | + } | |
| 1886 | + | |
| 1444 | 1887 | $vector_data = array( |
| 1445 | 1888 | 'id' => $vector_id, |
| 1446 | 1889 | 'values' => $embedding_vector, |
| 1447 | 1890 | 'metadata' => $metadata |
| @@ -1484,20 +1927,25 @@ | ||
| 1484 | 1927 | global $wpdb; |
| 1485 | 1928 | |
| 1486 | 1929 | // For chunks, we always insert new rows (no duplicate checking) |
| 1487 | 1930 | // 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') | |
| 1931 | + $row = array( | |
| 1932 | + 'url' => $source_url, | |
| 1933 | + 'article_content' => $content_with_metadata, | |
| 1934 | + 'embedding_vector' => $embedding_vector_serialized, | |
| 1935 | + 'source_url' => $source_url, | |
| 1936 | + 'content_type' => $content_type, | |
| 1937 | + 'timestamp' => current_time('mysql') | |
| 1499 | 1938 | ); |
| 1939 | + $formats = array('%s', '%s', '%s', '%s', '%s', '%s'); | |
| 1940 | + // Every chunk row carries the page title too (plan d9ee66) - readers group by | |
| 1941 | + // source_url and may pick any chunk as the representative. | |
| 1942 | + $page_title = self::source_title_for($source_url); | |
| 1943 | + if ($page_title !== '' && self::ensure_kb_title_column()) { | |
| 1944 | + $row['title'] = $page_title; | |
| 1945 | + $formats[] = '%s'; | |
| 1946 | + } | |
| 1947 | + $result = $wpdb->insert($table_name, $row, $formats); | |
| 1500 | 1948 | |
| 1501 | 1949 | if ($result === false) { |
| 1502 | 1950 | return new WP_Error('database_failed', 'Failed to insert chunk: ' . $wpdb->last_error); |
| 1503 | 1951 | } |
| @@ -1505,8 +1953,142 @@ | ||
| 1505 | 1953 | return true; |
| 1506 | 1954 | } |
| 1507 | 1955 | |
| 1508 | 1956 | /** |
| 1957 | + * Option holding the page an uploaded file should be cited as (plan 025ea2). | |
| 1958 | + * Keyed "<bot_id>|upload://<filename>" => https URL. The entry itself keeps | |
| 1959 | + * its upload:// identity, so a re-upload of the same file still replaces the | |
| 1960 | + * same entry and never collides with the cited page's own knowledge entry. | |
| 1961 | + */ | |
| 1962 | +const UPLOAD_SOURCE_URLS_OPTION = 'mxchat_upload_source_urls'; | |
| 1963 | + | |
| 1964 | +/** | |
| 1965 | + * Whether a stored source is an uploaded file (Document Upload / PDF Upload). | |
| 1966 | + * | |
| 1967 | + * @param string $source_url Stored source identity. | |
| 1968 | + * @return bool | |
| 1969 | + */ | |
| 1970 | +public static function is_upload_source($source_url) { | |
| 1971 | + return is_string($source_url) && stripos($source_url, 'upload://') === 0; | |
| 1972 | +} | |
| 1973 | + | |
| 1974 | +/** | |
| 1975 | + * File name of an uploaded source, for a plain-text label. | |
| 1976 | + * | |
| 1977 | + * @param string $source_url Stored source identity (upload://<filename>). | |
| 1978 | + * @return string | |
| 1979 | + */ | |
| 1980 | +public static function upload_source_name($source_url) { | |
| 1981 | + return self::is_upload_source($source_url) ? substr(self::upload_source_base($source_url), strlen('upload://')) : ''; | |
| 1982 | +} | |
| 1983 | + | |
| 1984 | +/** | |
| 1985 | + * An uploaded source without its page fragment (plan dcc19d). A page of an | |
| 1986 | + * uploaded PDF is stored as upload://<filename>#page=<n>; the file it belongs | |
| 1987 | + * to, and the key its Source URL is kept under, is upload://<filename>. | |
| 1988 | + * | |
| 1989 | + * @param string $source_url Stored source identity. | |
| 1990 | + * @return string | |
| 1991 | + */ | |
| 1992 | +public static function upload_source_base($source_url) { | |
| 1993 | + return is_string($source_url) ? preg_replace('/#page=\d+$/', '', $source_url) : ''; | |
| 1994 | +} | |
| 1995 | + | |
| 1996 | +/** | |
| 1997 | + * Page number carried by an uploaded PDF page's identity, 0 when it has none. | |
| 1998 | + * | |
| 1999 | + * @param string $source_url Stored source identity. | |
| 2000 | + * @return int | |
| 2001 | + */ | |
| 2002 | +public static function upload_source_page($source_url) { | |
| 2003 | + return (self::is_upload_source($source_url) && preg_match('/#page=(\d+)$/', $source_url, $m)) ? (int) $m[1] : 0; | |
| 2004 | +} | |
| 2005 | + | |
| 2006 | +/** | |
| 2007 | + * Identity one page of a PDF is stored under: <pdf>#page=<n>. A web PDF goes | |
| 2008 | + * through esc_url() as before. An uploaded PDF (upload://<filename>) must not: | |
| 2009 | + * esc_url() empties that scheme, which stored every page of an uploaded PDF | |
| 2010 | + * as anonymous manual content, so the file could not be named, cited, given | |
| 2011 | + * a Source URL, or replaced by a re-upload (plan dcc19d). | |
| 2012 | + * | |
| 2013 | + * @param string $pdf_url The PDF's own source (URL or upload://<filename>). | |
| 2014 | + * @param int $page_number 1-based page number. | |
| 2015 | + * @return string | |
| 2016 | + */ | |
| 2017 | +public static function pdf_page_identity($pdf_url, $page_number) { | |
| 2018 | + $identity = $pdf_url . '#page=' . (int) $page_number; | |
| 2019 | + return self::is_upload_source($pdf_url) ? sanitize_text_field($identity) : esc_url($identity); | |
| 2020 | +} | |
| 2021 | + | |
| 2022 | +/** | |
| 2023 | + * Remember (or clear) the page an uploaded file should be cited as. | |
| 2024 | + * Only http(s) URLs are kept; anything else clears the entry. | |
| 2025 | + * | |
| 2026 | + * @param string $source_label upload://<filename> | |
| 2027 | + * @param string $bot_id Bot the upload belongs to. | |
| 2028 | + * @param string $url Page URL, or '' to clear. | |
| 2029 | + * @return string The URL stored, '' when cleared. | |
| 2030 | + */ | |
| 2031 | +public static function set_upload_source_url($source_label, $bot_id, $url) { | |
| 2032 | + if (!self::is_upload_source($source_label)) { | |
| 2033 | + return ''; | |
| 2034 | + } | |
| 2035 | + $bot_id = ($bot_id === '' || $bot_id === null) ? 'default' : (string) $bot_id; | |
| 2036 | + $url = is_string($url) ? esc_url_raw(trim($url), array('http', 'https')) : ''; | |
| 2037 | + if ($url !== '' && !preg_match('#^https?://#i', $url)) { | |
| 2038 | + $url = ''; | |
| 2039 | + } | |
| 2040 | + | |
| 2041 | + $map = get_option(self::UPLOAD_SOURCE_URLS_OPTION, array()); | |
| 2042 | + if (!is_array($map)) { | |
| 2043 | + $map = array(); | |
| 2044 | + } | |
| 2045 | + $key = $bot_id . '|' . $source_label; | |
| 2046 | + if ($url === '') { | |
| 2047 | + if (!isset($map[$key])) { | |
| 2048 | + return ''; | |
| 2049 | + } | |
| 2050 | + unset($map[$key]); | |
| 2051 | + } else { | |
| 2052 | + $map[$key] = $url; | |
| 2053 | + } | |
| 2054 | + if (empty($map)) { | |
| 2055 | + delete_option(self::UPLOAD_SOURCE_URLS_OPTION); | |
| 2056 | + } else { | |
| 2057 | + update_option(self::UPLOAD_SOURCE_URLS_OPTION, $map, false); | |
| 2058 | + } | |
| 2059 | + return $url; | |
| 2060 | +} | |
| 2061 | + | |
| 2062 | +/** | |
| 2063 | + * The page an uploaded file is cited as, '' when it has none. | |
| 2064 | + * Looks under the asking bot first, then under the default bot (on the | |
| 2065 | + * WordPress database store every bot reads the same entries). | |
| 2066 | + * | |
| 2067 | + * @param string $source_label upload://<filename> | |
| 2068 | + * @param string $bot_id Bot answering the turn. | |
| 2069 | + * @return string | |
| 2070 | + */ | |
| 2071 | +public static function get_upload_source_url($source_label, $bot_id = 'default') { | |
| 2072 | + if (!self::is_upload_source($source_label)) { | |
| 2073 | + return ''; | |
| 2074 | + } | |
| 2075 | + $map = get_option(self::UPLOAD_SOURCE_URLS_OPTION, array()); | |
| 2076 | + if (!is_array($map) || empty($map)) { | |
| 2077 | + return ''; | |
| 2078 | + } | |
| 2079 | + // A PDF page (upload://<file>#page=<n>) is cited as its file's page. | |
| 2080 | + $source_label = self::upload_source_base($source_label); | |
| 2081 | + $bot_id = ($bot_id === '' || $bot_id === null) ? 'default' : (string) $bot_id; | |
| 2082 | + foreach (array($bot_id . '|' . $source_label, 'default|' . $source_label) as $key) { | |
| 2083 | + if (!empty($map[$key]) && preg_match('#^https?://#i', $map[$key])) { | |
| 2084 | + return $map[$key]; | |
| 2085 | + } | |
| 2086 | + } | |
| 2087 | + return ''; | |
| 2088 | +} | |
| 2089 | + | |
| 2090 | +/** | |
| 1509 | 2091 | * Delete all chunks for a given URL |
| 1510 | 2092 | * |
| 1511 | 2093 | * @param string $source_url The source URL |
| 1512 | 2094 | * @param string $bot_id The bot ID |
| @@ -1514,20 +2096,144 @@ | ||
| 1514 | 2096 | */ |
| 1515 | 2097 | public static function delete_chunks_for_url($source_url, $bot_id = 'default') { |
| 1516 | 2098 | //error_log('[MXCHAT-CHUNK-DELETE] Deleting chunks for URL: ' . $source_url); |
| 1517 | 2099 | |
| 2100 | + // Entry removal — mirror it to the Vector Store (unless a storage routine | |
| 2101 | + // is mid-re-store, see $vectorstore_mirror_suspended). | |
| 2102 | + if (!self::$vectorstore_mirror_suspended && class_exists('MxChat_Vectorstore_Manager')) { | |
| 2103 | + MxChat_Vectorstore_Manager::sync_delete_entry($source_url, $bot_id); | |
| 2104 | + } | |
| 2105 | + | |
| 1518 | 2106 | if (self::is_pinecone_enabled_for_bot($bot_id)) { |
| 1519 | - return self::delete_pinecone_chunks_by_url($source_url, $bot_id); | |
| 2107 | + $result = self::delete_pinecone_chunks_by_url($source_url, $bot_id); | |
| 1520 | 2108 | } else { |
| 1521 | - return self::delete_wordpress_chunks_by_url($source_url); | |
| 2109 | + $result = self::delete_wordpress_chunks_by_url($source_url); | |
| 1522 | 2110 | } |
| 2111 | + | |
| 2112 | + // The entry is gone for good (not the clean-slate half of a re-store): | |
| 2113 | + // a media file's "already indexed, nothing changed" signature must go | |
| 2114 | + // with it, or auto-sync keeps skipping the file on every later save. | |
| 2115 | + if (!self::$vectorstore_mirror_suspended && !is_wp_error($result)) { | |
| 2116 | + self::forget_attachment_signature($source_url, $bot_id); | |
| 2117 | + } | |
| 2118 | + | |
| 2119 | + return $result; | |
| 1523 | 2120 | } |
| 1524 | 2121 | |
| 1525 | 2122 | /** |
| 2123 | + * Clear the media auto-sync signature (_mxchat_kb_attachment_sig) of the | |
| 2124 | + * attachment a just-deleted knowledge entry was indexed from. | |
| 2125 | + * | |
| 2126 | + * Call it wherever a WHOLE entry is removed (never for the clean-slate half | |
| 2127 | + * of a re-store, and never when sibling chunk rows survive). Posts and pages | |
| 2128 | + * carry no signature; for them this is two indexed lookups that find nothing. | |
| 2129 | + * | |
| 2130 | + * @param string $source_url The entry's source URL. | |
| 2131 | + * @param string $bot_id The bot whose entry was removed. | |
| 2132 | + */ | |
| 2133 | +public static function forget_attachment_signature($source_url, $bot_id = 'default') { | |
| 2134 | + global $wpdb; | |
| 2135 | + | |
| 2136 | + $source_url = (string) $source_url; | |
| 2137 | + if (!preg_match('#^https?://#i', $source_url) || !self::attachment_signature_follows_bot($bot_id)) { | |
| 2138 | + return; | |
| 2139 | + } | |
| 2140 | + | |
| 2141 | + $attachment_ids = array(); | |
| 2142 | + $by_url = attachment_url_to_postid($source_url); | |
| 2143 | + if ($by_url) { | |
| 2144 | + $attachment_ids[] = (int) $by_url; | |
| 2145 | + } | |
| 2146 | + | |
| 2147 | + // The signature stores the exact URL the indexer used, which can differ | |
| 2148 | + // from where the file lives now (replaced file, scaled variant). Bounded | |
| 2149 | + // by the number of signed attachments: the meta_key index does the work. | |
| 2150 | + $candidates = $wpdb->get_results($wpdb->prepare( | |
| 2151 | + "SELECT post_id, meta_value FROM {$wpdb->postmeta} | |
| 2152 | + WHERE meta_key = '_mxchat_kb_attachment_sig' AND meta_value LIKE %s | |
| 2153 | + LIMIT 20", | |
| 2154 | + '%' . $wpdb->esc_like($source_url) . '%' | |
| 2155 | + )); | |
| 2156 | + foreach ((array) $candidates as $candidate) { | |
| 2157 | + $signature = maybe_unserialize($candidate->meta_value); | |
| 2158 | + if (is_array($signature) && isset($signature['url']) && $signature['url'] === $source_url) { | |
| 2159 | + $attachment_ids[] = (int) $candidate->post_id; | |
| 2160 | + } | |
| 2161 | + } | |
| 2162 | + | |
| 2163 | + foreach (array_unique($attachment_ids) as $attachment_id) { | |
| 2164 | + delete_post_meta($attachment_id, '_mxchat_kb_attachment_sig'); | |
| 2165 | + } | |
| 2166 | +} | |
| 2167 | + | |
| 2168 | +/** | |
| 2169 | + * The same, for callers that hold a Pinecone vector id instead of a URL. | |
| 2170 | + * Only a base id (md5 of the source URL) names a whole entry; a single | |
| 2171 | + * chunk id leaves the rest of the entry in place, so it is ignored. | |
| 2172 | + */ | |
| 2173 | +public static function forget_attachment_signature_by_key($entry_key, $bot_id = 'default') { | |
| 2174 | + global $wpdb; | |
| 2175 | + | |
| 2176 | + $entry_key = (string) $entry_key; | |
| 2177 | + if (!preg_match('/^[a-f0-9]{32}$/', $entry_key) || !self::attachment_signature_follows_bot($bot_id)) { | |
| 2178 | + return; | |
| 2179 | + } | |
| 2180 | + | |
| 2181 | + $signed = $wpdb->get_results( | |
| 2182 | + "SELECT post_id, meta_value FROM {$wpdb->postmeta} | |
| 2183 | + WHERE meta_key = '_mxchat_kb_attachment_sig' LIMIT 5000" | |
| 2184 | + ); | |
| 2185 | + foreach ((array) $signed as $row) { | |
| 2186 | + $signature = maybe_unserialize($row->meta_value); | |
| 2187 | + if (is_array($signature) && isset($signature['url']) && md5((string) $signature['url']) === $entry_key) { | |
| 2188 | + delete_post_meta((int) $row->post_id, '_mxchat_kb_attachment_sig'); | |
| 2189 | + } | |
| 2190 | + } | |
| 2191 | +} | |
| 2192 | + | |
| 2193 | +/** | |
| 2194 | + * Every media entry of a bot was removed at once ("Delete all"). | |
| 2195 | + */ | |
| 2196 | +public static function forget_all_attachment_signatures($bot_id = 'default') { | |
| 2197 | + if ($bot_id !== 'default' && $bot_id !== 'testing' && !self::is_pinecone_enabled_for_bot($bot_id)) { | |
| 2198 | + // WordPress-database "Delete all" for another bot removes that bot's | |
| 2199 | + // rows only; the default bot's entries are still there. | |
| 2200 | + return; | |
| 2201 | + } | |
| 2202 | + if (self::attachment_signature_follows_bot($bot_id)) { | |
| 2203 | + delete_post_meta_by_key('_mxchat_kb_attachment_sig'); | |
| 2204 | + } | |
| 2205 | +} | |
| 2206 | + | |
| 2207 | +/** | |
| 2208 | + * The signature vouches for the DEFAULT bot's entry (media auto-sync only | |
| 2209 | + * indexes for the default bot). Removing another bot's copy from a Pinecone | |
| 2210 | + * store of its own leaves that entry, and so the signature, in place. | |
| 2211 | + */ | |
| 2212 | +private static function attachment_signature_follows_bot($bot_id) { | |
| 2213 | + if ($bot_id === 'default' || $bot_id === 'testing' || !self::is_pinecone_enabled_for_bot($bot_id)) { | |
| 2214 | + return true; | |
| 2215 | + } | |
| 2216 | + if (!class_exists('MxChat_Pinecone_Documents') || !self::is_pinecone_enabled_for_bot('default')) { | |
| 2217 | + return false; | |
| 2218 | + } | |
| 2219 | + $theirs = MxChat_Pinecone_Documents::config($bot_id); | |
| 2220 | + $ours = MxChat_Pinecone_Documents::config('default'); | |
| 2221 | + return strtolower((string) $theirs['host']) === strtolower((string) $ours['host']) | |
| 2222 | + && (string) $theirs['namespace'] === (string) $ours['namespace']; | |
| 2223 | +} | |
| 2224 | + | |
| 2225 | +/** | |
| 1526 | 2226 | * Delete all chunks for a URL from Pinecone |
| 1527 | 2227 | */ |
| 1528 | -private static function delete_pinecone_chunks_by_url($source_url, $bot_id) { | |
| 1529 | - // Get Pinecone configuration | |
| 2228 | +/** | |
| 2229 | + * Delete leftover md5(url)_chunk_N vectors after a URL's content was re-stored | |
| 2230 | + * as a SINGLE vector (content shrank below the chunk threshold on edit/re-import). | |
| 2231 | + * Unlike delete_pinecone_chunks_by_url this leaves the base id alone — the caller | |
| 2232 | + * just upserted the new content there. Failure is logged, not fatal: the save | |
| 2233 | + * itself succeeded, and the next save retries the sweep. | |
| 2234 | + */ | |
| 2235 | +private static function cleanup_pinecone_chunk_stragglers($source_url, $bot_id) { | |
| 1530 | 2236 | if ($bot_id === 'default' || !class_exists('MxChat_Multi_Bot_Manager')) { |
| 1531 | 2237 | $pinecone_options = get_option('mxchat_pinecone_addon_options'); |
| 1532 | 2238 | $api_key = $pinecone_options['mxchat_pinecone_api_key'] ?? ''; |
| 1533 | 2239 | $host = $pinecone_options['mxchat_pinecone_host'] ?? ''; |
| @@ -1546,21 +2252,30 @@ | ||
| 1546 | 2252 | } |
| 1547 | 2253 | } |
| 1548 | 2254 | |
| 1549 | 2255 | if (empty($host) || empty($api_key)) { |
| 1550 | - return new WP_Error('pinecone_config', 'Pinecone is not properly configured'); | |
| 2256 | + return; | |
| 1551 | 2257 | } |
| 1552 | 2258 | |
| 1553 | - $base_vector_id = md5($source_url); | |
| 1554 | - $vectors_to_delete = array(); | |
| 2259 | + // Document index (plan 362c31): list the chunk ids and delete them through the documents API. | |
| 2260 | + if (class_exists('MxChat_Pinecone_Documents') && MxChat_Pinecone_Documents::is_document_index($bot_id)) { | |
| 2261 | + $docs_cfg = array('api_key' => $api_key, 'host' => $host, 'namespace' => $namespace); | |
| 2262 | + $doc_stragglers = MxChat_Pinecone_Documents::list_ids(md5($source_url) . '_chunk_', $docs_cfg); | |
| 2263 | + if (!empty($doc_stragglers)) { | |
| 2264 | + $doc_result = MxChat_Pinecone_Documents::delete_ids($doc_stragglers, $docs_cfg); | |
| 2265 | + if (is_wp_error($doc_result) && class_exists('MxChat_Admin') && method_exists('MxChat_Admin', 'mxchat_log_debug')) { | |
| 2266 | + 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))); | |
| 2267 | + } | |
| 2268 | + } | |
| 2269 | + return; | |
| 2270 | + } | |
| 1555 | 2271 | |
| 1556 | - // Add the original single-vector ID (for non-chunked content) | |
| 1557 | - $vectors_to_delete[] = $base_vector_id; | |
| 2272 | + $stragglers = array(); | |
| 1558 | 2273 | |
| 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. | |
| 2274 | + // Pinecone /vectors/list is a GET endpoint with query-string parameters (a POST | |
| 2275 | + // answers 200-with-an-empty-body, which reads as "no stragglers"). | |
| 1561 | 2276 | $query_params = array( |
| 1562 | - 'prefix' => $base_vector_id . '_chunk_', | |
| 2277 | + 'prefix' => md5($source_url) . '_chunk_', | |
| 1563 | 2278 | 'limit' => 100, |
| 1564 | 2279 | ); |
| 1565 | 2280 | if (!empty($namespace)) { |
| 1566 | 2281 | $query_params['namespace'] = $namespace; |
| @@ -1567,9 +2282,8 @@ | ||
| 1567 | 2282 | } |
| 1568 | 2283 | |
| 1569 | 2284 | $list_url = "https://{$host}/vectors/list?" . http_build_query($query_params); |
| 1570 | 2285 | |
| 1571 | - // Paginate in case a URL has more than 100 chunks. | |
| 1572 | 2286 | do { |
| 1573 | 2287 | $list_response = wp_remote_get($list_url, array( |
| 1574 | 2288 | 'headers' => array( |
| 1575 | 2289 | 'Api-Key' => $api_key, |
| @@ -1585,9 +2299,9 @@ | ||
| 1585 | 2299 | $list_data = json_decode(wp_remote_retrieve_body($list_response), true); |
| 1586 | 2300 | if (!empty($list_data['vectors'])) { |
| 1587 | 2301 | foreach ($list_data['vectors'] as $vector) { |
| 1588 | 2302 | if (isset($vector['id'])) { |
| 1589 | - $vectors_to_delete[] = $vector['id']; | |
| 2303 | + $stragglers[] = $vector['id']; | |
| 1590 | 2304 | } |
| 1591 | 2305 | } |
| 1592 | 2306 | } |
| 1593 | 2307 | |
| @@ -1598,8 +2312,188 @@ | ||
| 1598 | 2312 | |
| 1599 | 2313 | $query_params['paginationToken'] = $next_token; |
| 1600 | 2314 | $list_url = "https://{$host}/vectors/list?" . http_build_query($query_params); |
| 1601 | 2315 | } while (true); |
| 2316 | + | |
| 2317 | + if (empty($stragglers)) { | |
| 2318 | + return; | |
| 2319 | + } | |
| 2320 | + | |
| 2321 | + $delete_body = array('ids' => $stragglers); | |
| 2322 | + if (!empty($namespace)) { | |
| 2323 | + $delete_body['namespace'] = $namespace; | |
| 2324 | + } | |
| 2325 | + | |
| 2326 | + $delete_response = wp_remote_post("https://{$host}/vectors/delete", array( | |
| 2327 | + 'headers' => array( | |
| 2328 | + 'Api-Key' => $api_key, | |
| 2329 | + 'accept' => 'application/json', | |
| 2330 | + 'content-type' => 'application/json' | |
| 2331 | + ), | |
| 2332 | + 'body' => wp_json_encode($delete_body), | |
| 2333 | + 'timeout' => 30 | |
| 2334 | + )); | |
| 2335 | + | |
| 2336 | + if ((is_wp_error($delete_response) || wp_remote_retrieve_response_code($delete_response) !== 200) | |
| 2337 | + && class_exists('MxChat_Admin') && method_exists('MxChat_Admin', 'mxchat_log_debug')) { | |
| 2338 | + 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))); | |
| 2339 | + } | |
| 2340 | +} | |
| 2341 | + | |
| 2342 | +/** | |
| 2343 | + * Resolve the Pinecone credentials for a bot: the bot's own config when the | |
| 2344 | + * multi-bot add-on supplies one, else the site-wide options. Extracted so the | |
| 2345 | + * delete path and the count path can never resolve different credentials. | |
| 2346 | + * | |
| 2347 | + * @param string $bot_id | |
| 2348 | + * @return array{api_key:string,host:string,namespace:string} | |
| 2349 | + */ | |
| 2350 | +private static function pinecone_config_for_bot($bot_id) { | |
| 2351 | + if ($bot_id === 'default' || !class_exists('MxChat_Multi_Bot_Manager')) { | |
| 2352 | + $bot_pinecone_config = array(); | |
| 2353 | + } else { | |
| 2354 | + $bot_pinecone_config = apply_filters('mxchat_get_bot_pinecone_config', array(), $bot_id); | |
| 2355 | + } | |
| 2356 | + | |
| 2357 | + if (empty($bot_pinecone_config)) { | |
| 2358 | + $pinecone_options = get_option('mxchat_pinecone_addon_options'); | |
| 2359 | + return array( | |
| 2360 | + 'api_key' => $pinecone_options['mxchat_pinecone_api_key'] ?? '', | |
| 2361 | + 'host' => $pinecone_options['mxchat_pinecone_host'] ?? '', | |
| 2362 | + 'namespace' => $pinecone_options['mxchat_pinecone_namespace'] ?? '', | |
| 2363 | + // 'vector' | 'document' (plan 362c31) — the delete / count paths branch on it. | |
| 2364 | + 'index_type' => (($pinecone_options['mxchat_pinecone_index_type'] ?? 'vector') === 'document') ? 'document' : 'vector', | |
| 2365 | + ); | |
| 2366 | + } | |
| 2367 | + | |
| 2368 | + return array( | |
| 2369 | + 'api_key' => $bot_pinecone_config['api_key'] ?? '', | |
| 2370 | + 'host' => $bot_pinecone_config['host'] ?? '', | |
| 2371 | + 'namespace' => $bot_pinecone_config['namespace'] ?? '', | |
| 2372 | + 'index_type' => (($bot_pinecone_config['index_type'] ?? 'vector') === 'document') ? 'document' : 'vector', | |
| 2373 | + ); | |
| 2374 | +} | |
| 2375 | + | |
| 2376 | +/** | |
| 2377 | + * List every Pinecone vector id under a prefix, following pagination. | |
| 2378 | + * | |
| 2379 | + * Pinecone /vectors/list is a GET endpoint with query-string parameters; a POST here returns a | |
| 2380 | + * non-200 silently and we end up seeing only the base vector, leaving chunks orphaned. | |
| 2381 | + * | |
| 2382 | + * @param string $prefix Vector-id prefix to match. | |
| 2383 | + * @param array $cfg From pinecone_config_for_bot(). | |
| 2384 | + * @return array List of vector ids (empty when none match or the API is unreachable). | |
| 2385 | + */ | |
| 2386 | +private static function pinecone_list_ids_by_prefix($prefix, array $cfg) { | |
| 2387 | + // Document index (plan 362c31): the documents API lists by prefix too. | |
| 2388 | + if (($cfg['index_type'] ?? 'vector') === 'document' && class_exists('MxChat_Pinecone_Documents')) { | |
| 2389 | + return MxChat_Pinecone_Documents::list_ids($prefix, $cfg); | |
| 2390 | + } | |
| 2391 | + | |
| 2392 | + $ids = array(); | |
| 2393 | + | |
| 2394 | + $query_params = array( | |
| 2395 | + 'prefix' => $prefix, | |
| 2396 | + 'limit' => 100, | |
| 2397 | + ); | |
| 2398 | + if (!empty($cfg['namespace'])) { | |
| 2399 | + $query_params['namespace'] = $cfg['namespace']; | |
| 2400 | + } | |
| 2401 | + | |
| 2402 | + $list_url = "https://{$cfg['host']}/vectors/list?" . http_build_query($query_params); | |
| 2403 | + | |
| 2404 | + // Paginate in case a URL has more than 100 chunks. | |
| 2405 | + do { | |
| 2406 | + $list_response = wp_remote_get($list_url, array( | |
| 2407 | + 'headers' => array( | |
| 2408 | + 'Api-Key' => $cfg['api_key'], | |
| 2409 | + 'accept' => 'application/json', | |
| 2410 | + ), | |
| 2411 | + 'timeout' => 30, | |
| 2412 | + )); | |
| 2413 | + | |
| 2414 | + if (is_wp_error($list_response) || wp_remote_retrieve_response_code($list_response) !== 200) { | |
| 2415 | + break; | |
| 2416 | + } | |
| 2417 | + | |
| 2418 | + $list_data = json_decode(wp_remote_retrieve_body($list_response), true); | |
| 2419 | + if (!empty($list_data['vectors'])) { | |
| 2420 | + foreach ($list_data['vectors'] as $vector) { | |
| 2421 | + if (isset($vector['id'])) { | |
| 2422 | + $ids[] = $vector['id']; | |
| 2423 | + } | |
| 2424 | + } | |
| 2425 | + } | |
| 2426 | + | |
| 2427 | + $next_token = $list_data['pagination']['next'] ?? ''; | |
| 2428 | + if (empty($next_token)) { | |
| 2429 | + break; | |
| 2430 | + } | |
| 2431 | + | |
| 2432 | + $query_params['paginationToken'] = $next_token; | |
| 2433 | + $list_url = "https://{$cfg['host']}/vectors/list?" . http_build_query($query_params); | |
| 2434 | + } while (true); | |
| 2435 | + | |
| 2436 | + return $ids; | |
| 2437 | +} | |
| 2438 | + | |
| 2439 | +/** | |
| 2440 | + * How many stored pieces a URL currently has, in whichever backend holds it. | |
| 2441 | + * | |
| 2442 | + * Branches on the SAME condition delete_chunks_for_url() branches on, so the | |
| 2443 | + * number reported is the number that call removes. Used by the REST DELETE | |
| 2444 | + * route to answer "was anything actually there?" without a second delete path. | |
| 2445 | + * | |
| 2446 | + * @param string $source_url | |
| 2447 | + * @param string $bot_id | |
| 2448 | + * @return int | |
| 2449 | + */ | |
| 2450 | +public static function count_chunks_for_url($source_url, $bot_id = 'default') { | |
| 2451 | + if (self::is_pinecone_enabled_for_bot($bot_id)) { | |
| 2452 | + $cfg = self::pinecone_config_for_bot($bot_id); | |
| 2453 | + if (empty($cfg['host']) || empty($cfg['api_key'])) { | |
| 2454 | + return 0; | |
| 2455 | + } | |
| 2456 | + // The base id and every "<base>_chunk_N" share the base as a prefix, | |
| 2457 | + // so one prefix listing covers chunked and non-chunked content alike. | |
| 2458 | + return count(self::pinecone_list_ids_by_prefix(md5($source_url), $cfg)); | |
| 2459 | + } | |
| 2460 | + | |
| 2461 | + global $wpdb; | |
| 2462 | + $table_name = $wpdb->prefix . 'mxchat_system_prompt_content'; | |
| 2463 | + | |
| 2464 | + return (int) $wpdb->get_var($wpdb->prepare( | |
| 2465 | + "SELECT COUNT(*) FROM {$table_name} WHERE source_url = %s", | |
| 2466 | + $source_url | |
| 2467 | + )); | |
| 2468 | +} | |
| 2469 | + | |
| 2470 | +private static function delete_pinecone_chunks_by_url($source_url, $bot_id) { | |
| 2471 | + $cfg = self::pinecone_config_for_bot($bot_id); | |
| 2472 | + $api_key = $cfg['api_key']; | |
| 2473 | + $host = $cfg['host']; | |
| 2474 | + $namespace = $cfg['namespace']; | |
| 2475 | + | |
| 2476 | + if (empty($host) || empty($api_key)) { | |
| 2477 | + return new WP_Error('pinecone_config', 'Pinecone is not properly configured'); | |
| 2478 | + } | |
| 2479 | + | |
| 2480 | + // Document index (plan 362c31): base id + every <md5>_chunk_N, documents API. | |
| 2481 | + if (($cfg['index_type'] ?? 'vector') === 'document' && class_exists('MxChat_Pinecone_Documents')) { | |
| 2482 | + $doc_ids = array_merge(array(md5($source_url)), MxChat_Pinecone_Documents::list_ids(md5($source_url) . '_chunk_', $cfg)); | |
| 2483 | + return MxChat_Pinecone_Documents::delete_ids($doc_ids, $cfg); | |
| 2484 | + } | |
| 2485 | + | |
| 2486 | + $base_vector_id = md5($source_url); | |
| 2487 | + $vectors_to_delete = array(); | |
| 2488 | + | |
| 2489 | + // Add the original single-vector ID (for non-chunked content) | |
| 2490 | + $vectors_to_delete[] = $base_vector_id; | |
| 2491 | + | |
| 2492 | + $vectors_to_delete = array_merge( | |
| 2493 | + $vectors_to_delete, | |
| 2494 | + self::pinecone_list_ids_by_prefix($base_vector_id . '_chunk_', $cfg) | |
| 2495 | + ); | |
| 1602 | 2496 | |
| 1603 | 2497 | if (empty($vectors_to_delete)) { |
| 1604 | 2498 | //error_log('[MXCHAT-CHUNK-DELETE] No vectors found to delete'); |
| 1605 | 2499 | return true; |
