Avoid producing nan values when hashing, they do not compare well.

1 files changed, 6 insertions, 1 deletions

diff --git a/searchlib/src/main/java/com/yahoo/searchlib/rankingexpression/evaluation/StringValue.java b/searchlib/src/main/java/com/yahoo/searchlib/rankingexpression/evaluation/StringValue.java
index be5b4242042..6325d8d0334 100644
--- a/searchlib/src/main/java/com/yahoo/searchlib/rankingexpression/evaluation/StringValue.java
+++ b/searchlib/src/main/java/com/yahoo/searchlib/rankingexpression/evaluation/StringValue.java
@@ -43,7 +43,12 @@ public class StringValue extends Value {
     public double asDouble() {
         XXHash64 hasher = XXHashFactory.fastestInstance().hash64();
         byte[] data = value.getBytes(StandardCharsets.UTF_8);
-        return Double.longBitsToDouble(hasher.hash(data, 0, data.length, 0));
+        long h = hasher.hash(data, 0, data.length, 0);
+        if ((h & 0x7ff0000000000000L) == 0x7ff0000000000000L) {
+            // Avoid nan
+            h = h & 0xffefffffffffffffL;
+        }
+        return Double.longBitsToDouble(h);
     }
 
     @Override