diff --git a/lucene/CHANGES.txt b/lucene/CHANGES.txt
index 0dced289cf3..a2fd2234fc9 100644
--- a/lucene/CHANGES.txt
+++ b/lucene/CHANGES.txt
@@ -200,6 +200,10 @@ Bug Fixes
 * LUCENE-7956: Fixed potential stack overflow error in ICUNormalizer2CharFilter.
   (Adrien Grand)
 
+* LUCENE-7963: Remove useless getAttribute() in DefaultIndexingChain that
+  causes performance drop, introduced by LUCENE-7626.  (Daniel Mitterdorfer
+  via Uwe Schindler)
+
 Improvements
 
 * LUCENE-7489: Better storage of sparse doc-values fields with the default
diff --git a/lucene/core/src/java/org/apache/lucene/index/DefaultIndexingChain.java b/lucene/core/src/java/org/apache/lucene/index/DefaultIndexingChain.java
index f2c3de11a4c..fd241057eca 100644
--- a/lucene/core/src/java/org/apache/lucene/index/DefaultIndexingChain.java
+++ b/lucene/core/src/java/org/apache/lucene/index/DefaultIndexingChain.java
@@ -27,7 +27,6 @@ import java.util.Map;
 import java.util.Set;
 
 import org.apache.lucene.analysis.TokenStream;
-import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;
 import org.apache.lucene.codecs.DocValuesConsumer;
 import org.apache.lucene.codecs.DocValuesFormat;
 import org.apache.lucene.codecs.NormsConsumer;
@@ -733,7 +732,6 @@ final class DefaultIndexingChain extends DocConsumer {
         stream.reset();
         invertState.setAttributeSource(stream);
         termsHashPerField.start(field, first);
-        CharTermAttribute termAtt = tokenStream.getAttribute(CharTermAttribute.class);
 
         while (stream.incrementToken()) {