Add a comment about the recursion threshold of StringMSBRadixSorter.

lucene/core/src/java/org/apache/lucene/util/InPlaceMergeSorter.java

@@ -33,7 +33,7 @@ public abstract class InPlaceMergeSorter extends Sorter {
   }
 
   void mergeSort(int from, int to) {
-    if (to - from < THRESHOLD) {
+    if (to - from < INSERTION_SORT_THRESHOLD) {
       insertionSort(from, to);
     } else {
       final int mid = (from + to) >>> 1;

lucene/core/src/java/org/apache/lucene/util/IntroSorter.java

@@ -38,7 +38,7 @@ public abstract class IntroSorter extends Sorter {
   }
 
   void quicksort(int from, int to, int maxDepth) {
-    if (to - from < THRESHOLD) {
+    if (to - from < INSERTION_SORT_THRESHOLD) {
       insertionSort(from, to);
       return;
     } else if (--maxDepth < 0) {

lucene/core/src/java/org/apache/lucene/util/Sorter.java

@@ -23,7 +23,7 @@ import java.util.Comparator;
  * @lucene.internal */
 public abstract class Sorter {
 
-  static final int THRESHOLD = 20;
+  static final int INSERTION_SORT_THRESHOLD = 20;
 
   /** Sole constructor, used for inheritance. */
   protected Sorter() {}

lucene/core/src/java/org/apache/lucene/util/StringMSBRadixSorter.java

@@ -21,10 +21,13 @@ import java.util.Arrays;
 /** Radix sorter for variable-length strings. This class sorts based on the most
  *  significant byte first and falls back to {@link IntroSorter} when the size
  *  of the buckets to sort becomes small. It is <b>NOT</b> stable.
- *  Worst-case memory usage is about {@code 2.3 KB} */
+ *  Worst-case memory usage is about {@code 2.3 KB}. */
 abstract class StringMSBRadixSorter extends Sorter {
 
   // after that many levels of recursion we fall back to introsort anyway
+  // this is used as a protection against the fact that radix sort performs
+  // worse when there are long common prefixes (probably because of cache
+  // locality)
   private static final int LEVEL_THRESHOLD = 8;
   // size of histograms: 256 + 1 to indicate that the string is finished
   private static final int HISTOGRAM_SIZE = 257;