bkietz commented on a change in pull request #10934: URL: https://github.com/apache/arrow/pull/10934#discussion_r688741589
########## File path: format/ComputeIR.fbs ########## @@ -0,0 +1,211 @@ +// 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. + +include "Schema.fbs"; +include "Message.fbs"; + +namespace org.apache.arrow.flatbuf.computeir; + +/// Wrapper for blobs of arbitrary bytes +table Blob { + bytes: [ubyte]; +} + +/// An expression is one of +/// - a Literal datum +/// - a reference to a Field from a Relation +/// - a call to a named function +/// On evaluation, an Expression will have either array or scalar shape. +union Expression { + Literal, FieldRef, Call +} + +table Literal { + /// Number of slots. If this is absent, the literal should be + /// considered scalar. + length: long; + + /// The type of this literal. + type: Type (required); + + /// Buffers containing `length` elements of arrow-formatted data. + /// If `length` is absent (this Literal is scalar), these buffers + /// are sized to accommodate a single element of arrow-formatted data. + /// XXX this can be optimized for trivial scalars later + buffers: [Buffer]; +} + +table FieldRef { + /// A sequence of field names to allow referencing potentially nested fields + path: [string]; + + /// For Expressions which might reference fields in multiple Relations, + /// this index may be provided to indicate which Relation's fields + /// `path` points into. For example in the case of a join, + /// 0 refers to the left relation and 1 to the right relation. + relation_index: int; + + /// The type of data in the referenced Field. + type: Type; +} + +table Call { + /// The name of the function whose invocation this Call represents. + function_name: string (required); + + /// Parameters for `function_name`; content/format may be unique to each + /// value of `function_name`. + options: Blob; + + /// The arguments passed to `function_name`. + arguments: [Expression] (required); + + /// The type of data which invoking `function_name` will return. + type: Type; +} + +/// A relation is a set of rows with consitent schema. +table Relation { + /// The namespaced name of this Relation. + /// + /// Names with no namespace are reserved for pure relational + /// algebraic operations, which currently include: + /// "filter" + /// "project" + /// "aggregate" + /// "join" + /// "order_by" + /// "limit" + /// "literal" + /// "interactive_output" + relation_name: string (required); + + /// Parameters for `relation_name`; content/format may be unique to each + /// value of `relation_name`. + options: Blob; + + /// The arguments passed to `relation_name`. + arguments: [Relation] (required); + + /// The schema of rows in this Relation + schema: Schema; +} + +/// The contents of Relation.options will be FilterOptions +/// if Relation.name = "filter" +table FilterOptions { + /// The expression which will be evaluated against input rows + /// to determine whether they should be excluded from the + /// "filter" relation's output. + filter_expression: Expression (required); +} + +/// The contents of Relation.options will be ProjectOptions +/// if Relation.name = "project" +table ProjectOptions { + /// Expressions which will be evaluated to produce to + /// the rows of the "project" relation's output. + expressions: [Expression] (required); +} + +/// The contents of Relation.options will be AggregateOptions +/// if Relation.name = "aggregate" +table AggregateOptions { + /// Expressions which will be evaluated to produce to + /// the rows of the "aggregate" relation's output. + aggregations: [Expression] (required); + /// Keys by which `aggregations` will be grouped. + keys: [Expression]; +} + +enum JoinKind : int { + INNER, + LEFT, + RIGHT, + FULL, + SEMI, + ANTI, +} + +/// The contents of Relation.options will be JoinOptions +/// if Relation.name = "join" +table JoinOptions { + /// The expression which will be evaluated against rows from each + /// input to determine whether they should be included in the + /// "join" relation's output. + on_expression: Expression (required); + join_kind: JoinKind; +} + +/// Whether lesser values should precede greater or vice versa. +enum Ordering : uint8 { + ASCENDING, + DESCENDING, +} + +/// Whether nulls should precede or follow other values. +enum NullOrdering : uint8 { + FIRST, + LAST +} + +table SortKey { + value: Expression (required); + ordering: Ordering = ASCENDING; + null_ordering: NullOrdering = LAST; +} + +/// The contents of Relation.options will be OrderByOptions +/// if Relation.name = "order_by" +table OrderByOptions { + /// Define sort order for rows of output. + /// Keys with higher precedence are ordered ahead of other keys. + keys: [SortKey] (required); +} + +/// The contents of Relation.options will be LimitOptions +/// if Relation.name = "limit" +table LimitOptions { + /// Set the maximum number of rows of output. + count: long; +} + +/// The contents of Relation.options will be LiteralOptions +/// if Relation.name = "literal" +table LiteralOptions { + /// Batches of rows in this literal. + batches: [RecordBatch] (required); + /// A dictionary batch for this literal. + dictionary_batch: DictionaryBatch; +} + +/// The contents of Relation.options will be InteractiveOutputOptions +/// if Relation.name = "interactive_output" +table InteractiveOutputOptions { + /// In an interactive context there is an implicit pipe open between + /// the producer and the consumer along which output may be sent. + /// `id` is an opaque identifier for such outputs; for example it could be used to + /// identify which Jupyter output cell should receive the output. + id: long; Review comment: ```suggestion options: Blob; ``` -- 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]
