Handles anonymous nodes explicitly in IdMapper

previously anonymous nodes, i.e. nodes w/o :ID would not be put into the
IdMapper and so the EncodingIdMapper would have those data indexes set to
-1. Sometimes those -1 could affect sorting or lookup and potentially
collide with other encoded values by mistake.

This commit explicitly sets these anonymous values to 0 and treats that
value specifically during collision detection. Also, Encoder isn't allowed
to return value 0.

community/import-tool/src/test/java/org/neo4j/tooling/ImportToolTest.java

@@ -74,6 +74,7 @@
 import static org.neo4j.helpers.Exceptions.withMessage;
 import static org.neo4j.helpers.collection.Iterables.filter;
 import static org.neo4j.helpers.collection.IteratorUtil.count;
+import static org.neo4j.helpers.collection.IteratorUtil.single;
 import static org.neo4j.helpers.collection.IteratorUtil.singleOrNull;
 import static org.neo4j.tooling.ImportTool.MULTI_FILE_DELIMITER;
 
@@ -559,6 +560,55 @@ public void shouldDisallowImportWithoutNodesInput() throws Exception
         }
     }
 
+    @Test
+    public void shouldBeAbleToImportAnonymousNodes() throws Exception
+    {
+        // GIVEN
+        List<String> nodeIds = asList( "1", "", "", "", "3", "", "", "", "", "", "5" );
+        Configuration config = Configuration.COMMAS;
+        List<RelationshipDataLine> relationshipData = asList( relationship( "1", "3", "KNOWS" ) );
+
+        // WHEN
+        importTool(
+                "--into",          dbRule.getStoreDir().getAbsolutePath(),
+                "--nodes",         nodeData( true, config, nodeIds, alwaysTrue() ).getAbsolutePath(),
+                "--relationships", relationshipData( true, config, relationshipData.iterator(),
+                                   alwaysTrue(), true ).getAbsolutePath() );
+
+        // THEN
+        GraphDatabaseService db = dbRule.getGraphDatabaseService();
+        try ( Transaction tx = db.beginTx() )
+        {
+            Iterable<Node> allNodes = GlobalGraphOperations.at( db ).getAllNodes();
+            int anonymousCount = 0;
+            for ( final String id : nodeIds )
+            {
+                if ( id.isEmpty() )
+                {
+                    anonymousCount++;
+                }
+                else
+                {
+                    assertNotNull( single( filter( nodeFilter( id ), allNodes.iterator() ) ) );
+                }
+            }
+            assertEquals( anonymousCount, count( filter( nodeFilter( "" ), allNodes.iterator() ) ) );
+            tx.success();
+        }
+    }
+
+    private Predicate<Node> nodeFilter( final String id )
+    {
+        return new Predicate<Node>()
+        {
+            @Override
+            public boolean accept( Node node )
+            {
+                return node.getProperty( "id", "" ).equals( id );
+            }
+        };
+    }
+
     protected void assertNodeHasLabels( Node node, String[] names )
     {
         for ( String name : names )

community/kernel/src/main/java/org/neo4j/unsafe/impl/batchimport/cache/idmapping/string/Encoder.java

@@ -20,7 +20,9 @@
 package org.neo4j.unsafe.impl.batchimport.cache.idmapping.string;
 
 /**
- * Encodes a {@link Object} into a long.
+ * Encodes a {@link Object} into a long. The implementation should be designed to return as few collisions
+ * as possible, i.e. different incoming values being encoded into the same long.
+ * The returned encoded value must not be {@code 0} since that is a reserved value for {@link EncodingIdMapper}.
  */
 public interface Encoder
 {