Github user mccheah commented on a diff in the pull request:
https://github.com/apache/spark/pull/4972#discussion_r26277496
--- Diff: core/src/main/scala/org/apache/spark/util/CastedArray.scala ---
@@ -0,0 +1,126 @@
+/*
+* Licensed to the Apache Software Foundation (ASF) under one or more
+* contributor license agreements. See the NOTICE file distributed with
+* this work for additional information regarding copyright ownership.
+* The ASF licenses this file to You under the Apache License, Version 2.0
+* (the "License"); you may not use this file except in compliance with
+* the License. You may obtain a copy of the License at
+*
+* http://www.apache.org/licenses/LICENSE-2.0
+*
+* Unless required by applicable law or agreed to in writing, software
+* distributed under the License is distributed on an "AS IS" BASIS,
+* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+* See the License for the specific language governing permissions and
+* limitations under the License.
+*/
+
+package org.apache.spark.util
+
+/**
+ * Provides a wrapper around an object that is known to be an array, but
the specific
+ * type for the array is unknown.
+ *
+ * Normally, in situations when such an array is to be accessed
reflectively, one would use
+ * {@link java.lang.reflect.Array} using getLength() and get() methods.
However, it turns
+ * out that such methods are ill-performant.
+ *
+ * It turns out it is better to just use instanceOf and lots of casting
over calling through
+ * to the native C implementation. There is some discussion and a sample
code snippet in
+ * <a href=https://bugs.openjdk.java.net/browse/JDK-8051447>an open JDK
ticket</a>. In this
+ * class, that approach is implemented in an alternative way: creating a
wrapper object to
+ * wrap the array allows the cast to be done once, so the overhead of
casting multiple times
+ * is also avoided. It turns out we invoke the get() method to get the
value of the array
+ * numerous times, so doing the cast just once is worth the cost of
constructing the wrapper
+ * object for larger arrays.
+ */
+sealed trait CastedArray extends Any {
+ def get(i: Int): AnyRef
+ def getLength(): Int
+}
+
+object CastedArray {
+ def castAndWrap(arr: AnyRef): CastedArray = {
+ if (arr.isInstanceOf[Array[Boolean]]) {
+ return new BooleanCastedArray(arr.asInstanceOf[Array[Boolean]])
+ } else if (arr.isInstanceOf[Array[Byte]]) {
+ return new ByteCastedArray(arr.asInstanceOf[Array[Byte]])
+ } else if (arr.isInstanceOf[Array[Char]]) {
+ return new CharCastedArray(arr.asInstanceOf[Array[Char]])
+ } else if (arr.isInstanceOf[Array[Double]]) {
+ return new DoubleCastedArray(arr.asInstanceOf[Array[Double]])
+ } else if (arr.isInstanceOf[Array[Float]]) {
+ return new FloatCastedArray(arr.asInstanceOf[Array[Float]])
+ } else if (arr.isInstanceOf[Array[Int]]) {
+ return new IntCastedArray(arr.asInstanceOf[Array[Int]])
+ } else if (arr.isInstanceOf[Array[Long]]) {
+ return new LongCastedArray(arr.asInstanceOf[Array[Long]])
+ } else if (arr.isInstanceOf[Array[Object]]) {
+ return new ObjectCastedArray(arr.asInstanceOf[Array[Object]])
+ } else if (arr.isInstanceOf[Array[Short]]) {
+ return new ShortCastedArray(arr.asInstanceOf[Array[Short]])
+ } else {
+ throw createBadArrayException(arr)
+ }
+ }
+
+ // Boxing is not ideal, but we want to return AnyRef here. An
alternative implementation
+ // that used Java wouldn't force explicitly boxing... but returning
Object there would
+ // make the boxing happen implicitly anyways. In practice this tends to
be okay
+ // in terms of performance.
+ private class BooleanCastedArray(val arr: Array[Boolean]) extends AnyVal
with CastedArray {
+ override def get(i: Int): AnyRef = Boolean.box(arr(i))
--- End diff --
Part of the reason why boxing is okay is because SizeEstimator doesn't
actually use these on primitive arrays. If it gets a primitive array, size
estimation just multiplies the size of the array by the size of the primitive
type. So we don't ever invoke get() in those situations.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
