dongjoon-hyun commented on code in PR #56992:
URL: https://github.com/apache/spark/pull/56992#discussion_r3525560266


##########
sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/orc/types/ops/OrcTypeOps.scala:
##########
@@ -0,0 +1,169 @@
+/*
+ * 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.sql.execution.datasources.orc.types.ops
+
+import org.apache.hadoop.hive.ql.io.sarg.PredicateLeaf
+import org.apache.hadoop.io.WritableComparable
+import org.apache.orc.TypeDescription
+
+import org.apache.spark.sql.catalyst.expressions.SpecializedGetters
+import org.apache.spark.sql.types.{DataType, TimestampLTZNanosType, 
TimestampNTZNanosType, TimeType}
+
+/**
+ * Optional trait for ORC storage-format integration in the Types Framework.
+ *
+ * Implement this trait to enable ORC read/write support for a framework type. 
Each framework
+ * type that supports ORC provides a concrete implementation and registers it 
in the companion
+ * object's apply() method.
+ *
+ * The trait covers the ORC concerns that are homogeneous across the row-based 
code paths:
+ *   - Schema mapping: Spark DataType -> ORC schema string 
(OrcUtils.getOrcSchemaString) and the
+ *     ORC TypeDescription category (OrcUtils.orcTypeDescription). ORC has no 
native TIME /
+ *     nanosecond-timestamp category, so framework types map onto an existing 
physical ORC
+ *     category and round-trip the true Spark type via the 
CATALYST_TYPE_ATTRIBUTE_NAME attribute
+ *     (stamped uniformly by the caller, so the ops only chooses the category).
+ *   - Value write (serialize): Catalyst value -> ORC WritableComparable
+ *     (OrcSerializer.newConverter)
+ *   - Row-based read (deserialize): ORC WritableComparable -> Catalyst value
+ *     (OrcDeserializer.newWriter)
+ *   - Predicate pushdown: PredicateLeaf.Type mapping + literal casting 
(OrcFilters)
+ *
+ * DELIBERATELY NOT ON THE TRAIT:
+ *   - Vectorized read. ORC's vectorized path (OrcAtomicColumnVector, a Java 
class) dispatches via
+ *     boolean `instanceof` flags set in the constructor plus typed accessor 
methods
+ *     (getTimestampNTZNanos/getTimestampLTZNanos), NOT via an 
`Ops(dt).map(_.x)` closure. It has
+ *     no per-type extension seam, so it stays inline. This mirrors Parquet, 
whose vectorized read
+ *     is likewise not routed through ParquetTypeOps (it dispatches on the 
Spark type inline in
+ *     ParquetVectorUpdaterFactory.getUpdater).
+ *   - supportDataType. OrcFileFormat.supportDataType / 
OrcTable.supportsDataType already admit
+ *     every `AtomicType` via `case _: AtomicType => true`, so framework types 
are supported with
+ *     no per-type arm; no gate method is needed (unlike Parquet, whose 
default differs).
+ *
+ * DISPATCH PATTERN: framework FIRST at each row-path integration site. Each 
ORC infrastructure
+ * method wraps itself with:
+ * {{{
+ *   OrcTypeOps(dt).map(_.method(...)).getOrElse(methodDefault(dt, ...))
+ * }}}
+ * The original inline code is extracted to a *Default method unchanged. For a 
framework-managed
+ * type the ops handles it; for any other type OrcTypeOps(dt) is None and the 
*Default fallback
+ * executes the original path.
+ *
+ * DECOUPLING NOTE: makeDeserializer takes the Catalyst setter callbacks it 
needs
+ * ((Int, Long) => Unit and (Int, Any) => Unit) rather than OrcDeserializer's 
CatalystDataUpdater,
+ * because that updater is a sealed trait nested in OrcDeserializer and is not 
visible to this
+ * sub-package. The callbacks are the only two setter shapes the current 
framework types use.
+ *
+ * @see TimeTypeOrcOps for a reference implementation (primitive Long-backed 
type)
+ * @see TimestampLTZNanosOrcOps / TimestampNTZNanosOrcOps for 
OrcTimestamp-backed types
+ * @since 4.4.0
+ */
+private[orc] trait OrcTypeOps extends Serializable {
+
+  // ==================== Schema Mapping ====================
+
+  /**
+   * The ORC schema-string fragment for this type 
(OrcUtils.getOrcSchemaString). Examples:
+   * TimeType -> "bigint" (LongType.catalogString), TimestampLTZNanosType -> 
"timestamp with local
+   * time zone", TimestampNTZNanosType -> "timestamp".
+   */
+  def orcSchemaString: String
+
+  /**
+   * The ORC TypeDescription category for this type 
(OrcUtils.orcTypeDescription). The caller
+   * stamps CATALYST_TYPE_ATTRIBUTE_NAME = sparkType.typeName onto the 
returned descriptor, so the
+   * ops only chooses the physical category.
+   */
+  def orcCategory: TypeDescription.Category
+
+  // ==================== Value Write (serialize) ====================
+
+  /**
+   * Creates a converter that turns a Catalyst value at an ordinal into an ORC 
WritableComparable
+   * (OrcSerializer.newConverter).
+   *
+   * @param reuseObj whether the serializer may reuse a single mutable 
Writable across rows
+   *                 (OrcSerializer passes this through; the primitive 
Long-backed TimeType reuses,
+   *                 the OrcTimestamp-backed nanos types do not).
+   */
+  def makeSerializer(reuseObj: Boolean): (SpecializedGetters, Int) => 
WritableComparable[_]
+
+  // ==================== Row-Based Read (deserialize) ====================
+
+  /**
+   * Creates a writer that sets a decoded ORC value into the Catalyst row at 
an ordinal
+   * (OrcDeserializer.newWriter). The WritableComparable is the raw ORC value 
(LongWritable for
+   * LONG-category types, OrcTimestamp for the timestamp categories).
+   *
+   * @param setLong callback into the row's setLong (used by primitive 
Long-backed types)
+   * @param set     callback into the row's generic set (used by object-backed 
types, e.g. nanos)
+   */
+  def makeDeserializer(
+      setLong: (Int, Long) => Unit,
+      set: (Int, Any) => Unit): (Int, WritableComparable[_]) => Unit
+
+  // ==================== Predicate Pushdown ====================
+
+  /**
+   * The ORC PredicateLeaf.Type for this type 
(OrcFilters.getPredicateLeafType), or None to opt out
+   * of pushdown (the caller then treats the column as non-convertible). 
TimeType returns

Review Comment:
   This might be misleading because `OrcFilters.getPredicateLeafType` simply 
throws exceptions. If a new future type simply returns `None`, it means a 
failure in the planning, doesn't it?
   > the caller then treats the column as non-convertible
   
   
https://github.com/apache/spark/blob/6a0e671b7672982dd4138c368ea71660ca70e604/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/orc/OrcFilters.scala#L151



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to