force sub columns to be ordered when the super column is deserialised (cassandra 0.7.4)

gistfile1.java

Index: src/java/org/apache/cassandra/db/SuperColumn.java
===================================================================
--- src/java/org/apache/cassandra/db/SuperColumn.java	(date 1300200460000)
+++ src/java/org/apache/cassandra/db/SuperColumn.java	(revision )
@@ -359,8 +359,11 @@
         /* read the number of columns */
         int size = dis.readInt();
         ColumnSerializer serializer = Column.serializer();
-        ColumnSortedMap preSortedMap = new ColumnSortedMap(comparator, serializer, dis, size);
-        SuperColumn superColumn = new SuperColumn(name, new ConcurrentSkipListMap<ByteBuffer,IColumn>(preSortedMap));
+        final ColumnSortedMap preSortedMap = new ColumnSortedMap(comparator, serializer, dis, size);
+        ConcurrentSkipListMap<ByteBuffer, IColumn> columns = new ConcurrentSkipListMap<ByteBuffer, IColumn>(preSortedMap.comparator()){{
+            putAll(preSortedMap);
+        }};
+        SuperColumn superColumn = new SuperColumn(name, columns);
         if (localDeleteTime != Integer.MIN_VALUE && localDeleteTime <= 0)
         {
             throw new IOException("Invalid localDeleteTime read: " + localDeleteTime);

This is for a specific case where the on disk order of the sub columns needed to change. Running nodetool scrub should re-write the sub columns in the new order. Do not use in production other than to fix this. Only works if all rows smaller than in memory compaction limit.