drill git commit: DRILL-1872: Assertion failure and empty map on order by with complex data

[pa...@apache.org]

Repository: drill
Updated Branches:
  refs/heads/master ca28b9cba -> 066b6dbc3


DRILL-1872: Assertion failure and empty map on order by with complex data


Project: http://git-wip-us.apache.org/repos/asf/drill/repo
Commit: http://git-wip-us.apache.org/repos/asf/drill/commit/066b6dbc
Tree: http://git-wip-us.apache.org/repos/asf/drill/tree/066b6dbc
Diff: http://git-wip-us.apache.org/repos/asf/drill/diff/066b6dbc

Branch: refs/heads/master
Commit: 066b6dbc3a711dac822dd73441b1faedc1bebdfd
Parents: ca28b9c
Author: Parth Chandra <pc...@maprtech.com>
Authored: Mon Feb 2 13:39:39 2015 -0800
Committer: Parth Chandra <pc...@maprtech.com>
Committed: Thu Feb 12 07:08:15 2015 -0800

----------------------------------------------------------------------
 .../apache/drill/exec/record/MaterializedField.java | 16 ++++++----------
 .../apache/drill/exec/vector/complex/MapVector.java |  9 +++++++++
 2 files changed, 15 insertions(+), 10 deletions(-)
----------------------------------------------------------------------


http://git-wip-us.apache.org/repos/asf/drill/blob/066b6dbc/exec/java-exec/src/main/java/org/apache/drill/exec/record/MaterializedField.java
----------------------------------------------------------------------
diff --git a/exec/java-exec/src/main/java/org/apache/drill/exec/record/MaterializedField.java b/exec/java-exec/src/main/java/org/apache/drill/exec/record/MaterializedField.java
index a9f3292..bcc226f 100644
--- a/exec/java-exec/src/main/java/org/apache/drill/exec/record/MaterializedField.java
+++ b/exec/java-exec/src/main/java/org/apache/drill/exec/record/MaterializedField.java
@@ -181,10 +181,10 @@ public class MaterializedField {
 
   @Override
   public int hashCode() {
-    final int prime = 31;
     int result = 1;
-    result = prime * result + ((children == null) ? 0 : children.hashCode());
-    result = prime * result + ((key == null) ? 0 : key.hashCode());
+    // DRILL-1872: Compute hashCode only on key. See also the comment
+    // in MapVector$MapTransferPair
+    result = ((key == null) ? 0 : key.hashCode());
     return result;
   }
 
@@ -200,13 +200,9 @@ public class MaterializedField {
       return false;
     }
     MaterializedField other = (MaterializedField) obj;
-    if (children == null) {
-      if (other.children != null) {
-        return false;
-      }
-    } else if (!children.equals(other.children)) {
-      return false;
-    }
+    // DRILL-1872: Compute equals only on key. See also the comment
+    // in MapVector$MapTransferPair
+
     if (key == null) {
       if (other.key != null) {
         return false;

http://git-wip-us.apache.org/repos/asf/drill/blob/066b6dbc/exec/java-exec/src/main/java/org/apache/drill/exec/vector/complex/MapVector.java
----------------------------------------------------------------------
diff --git a/exec/java-exec/src/main/java/org/apache/drill/exec/vector/complex/MapVector.java b/exec/java-exec/src/main/java/org/apache/drill/exec/vector/complex/MapVector.java
index fb19afa..c5dc5ba 100644
--- a/exec/java-exec/src/main/java/org/apache/drill/exec/vector/complex/MapVector.java
+++ b/exec/java-exec/src/main/java/org/apache/drill/exec/vector/complex/MapVector.java
@@ -155,6 +155,15 @@ public class MapVector extends AbstractMapVector {
         if (vector == null) {
           continue;
         }
+        //DRILL-1872: we add the child fields for the vector, looking up the field by name. For a map vector,
+        // the child fields may be nested fields of the top level child. For example if the structure
+        // of a child field is oa.oab.oabc then we add oa, then add oab to oa then oabc to oab.
+        // But the children member of a Materialized field is a HashSet. If the fields are added in the
+        // children HashSet, and the hashCode of the Materialized field includes the hash code of the
+        // children, the hashCode value of oa changes *after* the field has been added to the HashSet.
+        // (This is similar to what happens in ScanBatch where the children cannot be added till they are
+        // read). To take care of this, we ensure that the hashCode of the MaterializedField does not
+        // include the hashCode of the children but is based only on MaterializedField$key.
         ValueVector newVector = to.addOrGet(child, vector.getField().getType(), vector.getClass());
         if (allocate && to.size() != preSize) {
           newVector.allocateNew();