[GitHub] [arrow] houqp commented on a change in pull request #10856: [RFC] Arrow Compute Serialized Intermediate Representation draft for discussion

[GitBox <gi...@apache.org>]

houqp commented on a change in pull request #10856:
URL: https://github.com/apache/arrow/pull/10856#discussion_r681486225



##########
File path: format/ComputeIR.fbs
##########
@@ -0,0 +1,521 @@
+/// 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.
+
+/// Arrow Compute IR (Intermediate Representation)
+///
+/// The purpose of these data structures is to provide a language- and compute
+/// engine-agnostic representation of common analytical operations on Arrow
+/// data. This may include so-called "logical query plans" generated by SQL
+/// systems, but it can be used to serialize different types of expression or
+/// query fragments for various purposes. For example, a system could use this
+/// to serialize array expressions for transmitting filters/predicates.
+///
+/// The three main types of data objects dealt with in this IR are:
+///
+/// * Table: a data source having an Arrow schema, resolvable algebraically to
+///   a collection of Arrow record batches
+/// * Array: logically, a field in a Table
+/// * Scalar: a single value, which is broadcastable to Array as needed
+///
+/// This IR specifically does not provide for query planning or physical
+/// execution details. It also aims to be as comprehensive as possible in
+/// capturing compute operations expressible in different query engines or data
+/// frame libraries. Engines are not expected to implement everything here.
+///
+/// One of the most common areas of divergence in query engines are the names
+/// and semantics of functions that operation on scalar or array
+/// inputs. Efforts to standardize function names and their expected semantics
+/// will happen outside of the serialized IR format defined here.
+
+// We use the IPC Schema types to represent data typesa
+include "Schema.fbs";
+
+namespace org.apache.arrow.flatbuf.computeir;
+
+/// ----------------------------------------------------------------------
+/// Data serialization for literal (constant / scalar) values. This assumes
+/// that the consumer has basic knowledge of the Arrow format and data types
+/// such that the binary scalar data that is encoded here can be unpacked into
+/// an appropriate literal value object. For example, if the Type for a Literal
+/// is FloatingPoint with Precision::DOUBLE, then we would expect to have a
+/// PrimitiveLiteralData with an 8-byte value.
+
+/// Serialized data which, given a data type, can be unpacked into a scalar
+/// value data structure.
+///
+/// NB(wesm): This is simpler from a Flatbuffers perspective than having a
+/// separate data type for each Arrow type. Alternative proposals welcome.
+union LiteralData {
+  NullLiteralData,
+  PrimitiveLiteralData,
+  ListLiteralData,
+  StructLiteralData,
+  UnionLiteralData
+}
+
+/// Placeholder for any null value, whether with Null type or a different
+/// non-Null type.
+table NullLiteralData {}
+
+/// For all data types represented as fixed-size-binary value (numeric and
+/// binary/string types included). Boolean values are to be represented as a
+/// single byte with value 1 (true) or 0 (false).
+table PrimitiveLiteralData {
+  data:[ubyte] (required);
+}
+
+/// For List, LargeList, and FixedSizeList.
+table ListLiteralData {
+  data:[LiteralData] (required);
+}
+
+/// For Struct
+table StructLiteralData {
+  data:[LiteralData] (required);
+}
+
+/// For Union
+table UnionLiteralData {
+  /// The type code (referencing the Union type) needed to reconstruct the
+  /// correct literal value.
+  type_code:int;  // required
+
+  value:LiteralData (required);
+}
+
+/// Literal serializes a scalar (constant) value in an array expression.
+table Literal {
+  type:Type (required);
+
+  /// The data needed to reconstruct the literal value.
+  data:LiteralData (required);
+}
+
+/// A sequence of literal values all having the same type.
+table LiteralVector {
+  type:Type (required);
+  data:[LiteralData] (required);
+}
+
+/// A name (key) and literal value, to use for map-like options fields.
+table NamedLiteral {
+  name:string;
+  value:Literal;
+}
+
+/// ----------------------------------------------------------------------
+/// One-dimensional operations (array/scalar input and output) and ArrayExpr,
+/// which is an operation plus a name and output type.
+
+/// A reference to an antecedent table schema in an expression tree
+table TableReference {
+  ///
+  name:string (required);
+}
+
+/// A reference to an antecedent column from a table schema in an expression
+/// tree.
+table ColumnReference {
+  name:string (required);

Review comment:
       i think a similar question can be asked for `name` in TableReference as well.




-- 
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: github-unsubscribe@arrow.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org