Move df safeguard to affected term(s) only

Signed-off-by: Michael Froh <[email protected]>

Author: msfroh

server/src/main/java/org/apache/lucene/queries/BlendedTermQuery.java

@@ -120,7 +120,7 @@ protected void blend(final TermStates[] contexts, int maxDoc, IndexReader reader
         }
         int max = 0;
         long minSumTTF = Long.MAX_VALUE;
-        int minDocCount = Integer.MAX_VALUE;
+        int[] docCounts = new int[contexts.length];
         for (int i = 0; i < contexts.length; i++) {
             TermStates ctx = contexts[i];
             int df = ctx.docFreq();
@@ -134,15 +134,12 @@ protected void blend(final TermStates[] contexts, int maxDoc, IndexReader reader
                 // we need to find out the minimum sumTTF to adjust the statistics
                 // otherwise the statistics don't match
                 minSumTTF = Math.min(minSumTTF, reader.getSumTotalTermFreq(terms[i].field()));
-                minDocCount = Math.min(minDocCount, reader.getDocCount(terms[i].field()));
+                docCounts[i] = reader.getDocCount(terms[i].field());
             }
         }
         if (maxDoc > minSumTTF) {
             maxDoc = (int) minSumTTF;
         }
-        if (maxDoc > minDocCount) {
-            maxDoc = minDocCount;
-        }
         if (max == 0) {
             return; // we are done that term doesn't exist at all
         }
@@ -180,7 +177,11 @@ protected int compare(int i, int j) {
             if (prev > current) {
                 actualDf++;
             }
-            contexts[i] = ctx = adjustDF(reader.getContext(), ctx, Math.min(maxDoc, actualDf));
+            // Per field, we want to guarantee that the adjusted df does not exceed the number of docs with the field.
+            // That is, in the IDF formula (log(1 + (N - n + 0.5) / (n + 0.5))), we need to make sure that n (the
+            // adjusted df) is never bigger than N (the number of docs with the field).
+            int fieldMaxDoc = Math.min(maxDoc, docCounts[i]);
+            contexts[i] = ctx = adjustDF(reader.getContext(), ctx, Math.min(fieldMaxDoc, actualDf));
             prev = current;
             sumTTF += ctx.totalTermFreq();
         }