fix(wiki-migration): pdftotext als primären PDF-Extraktor nutzen

pdfparser scheitert an "secured" PDFs (Kopier-Schutz ohne Passwort).
pdftotext aus poppler-utils umgeht diesen Schutz und extrahiert den
eingebetteten Text korrekt. Falls pdftotext nicht verfügbar ist oder
nichts liefert, bleibt pdfparser als Fallback erhalten.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

module/knowledgecenter_update/Script/migrate_old_wiki.php

@@ -143,28 +143,37 @@ function assemble_content(int $cid): array {
 }
 
 function get_article_files(int $cid): array {
-    // sitepart_id=5 verweist auf filegallery_header
+    // sitepart_id 5, 6, 10 sind filegallery-Typen; sitepart_id=1 ist textcontent_header (nicht joinen!)
-    $sql = "SELECT mcl.main_sitepart_header_id
+    $sql = "SELECT DISTINCT fl.filename, fl.extension, fl.description
             FROM main_collection_link mcl
-            WHERE mcl.main_collection_id = $cid AND mcl.main_sitepart_id = 5";
+            JOIN filegallery_line fl ON fl.header_id = mcl.main_sitepart_header_id
+            WHERE mcl.main_collection_id = $cid
+              AND mcl.main_sitepart_id IN (5, 6, 10)
+              AND mcl.main_sitepart_header_id > 0
+              AND fl.filename != ''
+            ORDER BY fl.sorting";
     $res   = mysqli_query($GLOBALS['mysql_con'], $sql);
     $files = [];
     while ($row = mysqli_fetch_assoc($res)) {
-        $hid = (int)$row['main_sitepart_header_id'];
+        $files[] = $row;
-        if ($hid === 0) continue;
-        $fres = mysqli_query($GLOBALS['mysql_con'],
-            "SELECT filename, extension, description FROM filegallery_line WHERE header_id = $hid ORDER BY sorting");
-        while ($frow = mysqli_fetch_assoc($fres)) {
-            if (trim($frow['filename']) !== '') {
-                $files[] = $frow;
-            }
-        }
     }
     return $files;
 }
 
 function extract_pdf_text(string $filePath): string {
     if (!file_exists($filePath) || filesize($filePath) === 0) return '';
+
+    // pdftotext ist robuster bei gesicherten PDFs (nur Kopier-Schutz, kein Passwort)
+    $pdftotext = trim(shell_exec('which pdftotext 2>/dev/null') ?? '');
+    if ($pdftotext !== '') {
+        $escaped = escapeshellarg($filePath);
+        $out     = shell_exec("$pdftotext -nopgbrk $escaped - 2>/dev/null");
+        if ($out !== null && strlen(trim($out)) > 10) {
+            return preg_replace('/\s+/', ' ', trim($out));
+        }
+    }
+
+    // Fallback: pdfparser
     try {
         $parser = new \Smalot\PdfParser\Parser();
         $pdf    = $parser->parseFile($filePath);
@@ -175,23 +184,31 @@ function extract_pdf_text(string $filePath): string {
     }
 }
 
-function generate_summary(string $htmlContent, array $files, string $userdataDir, string $apiKey, string $title): ?string {
+function generate_summary(string $htmlContent, array $files, string $userdataDir, string $apiKey, string $title, string $teaser = ''): ?string {
-    // Fließtext: HTML vollständig entfernen, keine Zeichenbegrenzung
-    $plainText = html_entity_decode(strip_tags($htmlContent), ENT_HTML5 | ENT_QUOTES, 'UTF-8');
-    $plainText = trim(preg_replace('/\s+/', ' ', $plainText));
-
     $parts = [];
-    if (strlen($plainText) > 30) {
+
-        $parts[] = $plainText;
+    // Teaser bevorzugen: bereits plain text, sehr kurz → spart Input-Tokens
+    if (strlen($teaser) > 30) {
+        $parts[] = $teaser;
     }
 
-    // PDFs: Text extrahieren, auf 1500 Zeichen begrenzen (bei sehr langen PDFs)
+    // Fließtext: HTML entfernen, auf 1500 Zeichen cappen
-    foreach ($files as $file) {
+    $plainText = html_entity_decode(strip_tags($htmlContent), ENT_HTML5 | ENT_QUOTES, 'UTF-8');
-        if (strtolower($file['extension']) !== 'pdf') continue;
+    $plainText = trim(preg_replace('/\s+/', ' ', $plainText));
-        $pdfText = extract_pdf_text($userdataDir . '/' . $file['filename']);
+    if (strlen($plainText) > 30) {
-        if (strlen($pdfText) > 50) {
+        $parts[] = mb_substr($plainText, 0, 1500);
-            $label   = $file['description'] ?: basename($file['filename']);
+    }
-            $parts[] = 'Anhang "' . $label . '": ' . mb_substr($pdfText, 0, 1500);
+
+    // PDFs nur wenn sonst kein Inhalt vorhanden, auf 800 Zeichen begrenzen
+    if (empty($parts)) {
+        foreach ($files as $file) {
+            if (strtolower($file['extension']) !== 'pdf') continue;
+            $pdfText = extract_pdf_text($userdataDir . '/' . $file['filename']);
+            if (strlen($pdfText) > 50) {
+                $label   = $file['description'] ?: basename($file['filename']);
+                $parts[] = 'Anhang "' . $label . '": ' . mb_substr($pdfText, 0, 800);
+                break; // ein PDF reicht
+            }
         }
     }
 
@@ -199,19 +216,17 @@ function generate_summary(string $htmlContent, array $files, string $userdataDir
         return null;
     }
 
-    // Kompakter Prompt: Titel + vollständiger Inhalt
+    $prompt = "Fasse in 2-3 deutschen Sätzen zusammen. Nur Zusammenfassung.\nTitel: $title\n\n"
-    $prompt = "Fasse diesen AWO-Wiki-Beitrag in 2-3 deutschen Sätzen zusammen. Nur Zusammenfassung, kein Intro.\n"
-        . "Titel: $title\n\n"
         . implode("\n\n", $parts);
 
-    logmsg("  PROMPT (" . strlen($prompt) . " Zeichen):\n" . $prompt . "\n--- PROMPT END ---", 'PROMPT');
+    logmsg("  PROMPT (" . strlen($prompt) . " Zeichen)", 'PROMPT');
 
     $payload = json_encode([
         'contents'         => [['parts' => [['text' => $prompt]]]],
         'generationConfig' => [
-            'maxOutputTokens' => 300,
+            'maxOutputTokens' => 150,
             'temperature'     => 0.2,
-            'thinkingConfig'  => ['thinkingBudget' => 0],  // kein Thinking = alle Token für Antwort
+            'thinkingConfig'  => ['thinkingBudget' => 0],
         ],
     ]);
 
@@ -279,7 +294,7 @@ if ($opts['summaries-only']) {
     if (!$apiKey) { fwrite(STDERR, "--summaries-only benötigt --api-key\n"); exit(2); }
 
     $sumSql = "SELECT p.id as post_id, p.legacy_collection_id, pv.id as version_id, pv.title,
-                      pv.content
+                      pv.teaser, pv.content
                FROM knowledgecenter_posts p
                JOIN knowledgecenter_post_versions pv ON pv.post_id = p.id AND pv.version_number = 1
                WHERE (pv.summary IS NULL OR pv.summary = '')
@@ -301,8 +316,9 @@ if ($opts['summaries-only']) {
         $versionId = (int)$sumRow['version_id'];
         $title     = $sumRow['title'];
 
-        $files = get_article_files($collId);
+        $files   = get_article_files($collId);
-        $summary = generate_summary($sumRow['content'], $files, $userdataDir, $apiKey, $title) ?? '';
+        $teaser  = $sumRow['teaser'] ?? '';
+        $summary = generate_summary($sumRow['content'], $files, $userdataDir, $apiKey, $title, $teaser) ?? '';
 
         if ($summary === '') {
             logmsg("  [FAIL] Post $postId \"$title\": leere Summary", 'WARN');
@@ -314,7 +330,7 @@ if ($opts['summaries-only']) {
             logmsg("[OK  ] Post $postId \"$title\": " . mb_substr($summary, 0, 80), 'OK');
             $sumStats['ok']++;
         }
-        usleep(200000);
+        sleep(4); // 15 RPM Free-Tier: mind. 4 Sek. zwischen Requests
     }
 
     logmsg(sprintf("Summaries fertig. ok=%d  fail=%d", $sumStats['ok'], $sumStats['fail']), 'HEAD');
@@ -380,7 +396,7 @@ while ($row = mysqli_fetch_assoc($result)) {
         $generated = generate_summary($fullContent, $files, $userdataDir, $apiKey, $title);
         $summary   = $generated ?? '';
         logmsg("  Summary: " . (strlen($summary) > 0 ? '"' . mb_substr($summary, 0, 100) . '"' : '(leer)'));
-        usleep(200000); // 200ms Pause für API rate limiting
+        sleep(4); // 15 RPM Free-Tier: mind. 4 Sek. zwischen Requests
     }
 
     // HTML-Entities dekodieren für saubere UTF-8 Speicherung