Force replace some HTML tags before sending messages to the search index

Stripping the HTML can cause certain words to be accidentally joined when there is no symbol between them that is recognized by the tokenizer. Inserting a whitespace at tag positions that are known to be prone is a stop-gap solution until we find a more stable replacement strategy.

See #4652 and WoltLab/com.woltlab.wcf.elasticSearch#14

diff --git a/wcfsetup/install/files/lib/system/search/SearchIndexManager.class.php b/wcfsetup/install/files/lib/system/search/SearchIndexManager.class.php
index 4c51d7afa1f950302ae053af8cd65fb2db31f5e7..a9f887422a84f3ecfebe42836e02e77c8e71b491 100644 (file)
--- a/wcfsetup/install/files/lib/system/search/SearchIndexManager.class.php
+++ b/wcfsetup/install/files/lib/system/search/SearchIndexManager.class.php
@@ -122,6 +122,24 @@ class SearchIndexManager extends SingletonFactory implements ISearchIndexManager
         $languageID = null,
         $metaData = ''
     ) {
+        // Force replace certain tags with a whitespace to prevent words from adjacent
+        // lines to be glued together.
+        $message = \str_replace([
+            '<br>',
+            '</h1>',
+            '</h2>',
+            '</h3>',
+            '</h4>',
+            '</h5>',
+            '</h6>',
+            '</kbd>',
+            '</li>',
+            '</p>',
+            '</pre>',
+            '</td>',
+            '</woltlab-metacode>',
+        ], ' ', $message);
+
         // strip html; remove whitespace from beginning and end of the message
         $message = StringUtil::trim(StringUtil::stripHTML($message));