Skip to content

UnresolvedAttribute Leaf Expression

[[name]] UnresolvedAttribute is a named spark-sql-Expression-Attribute.md[Attribute] leaf expression (i.e. it has a name) that represents a reference to an entity in a logical query plan.

UnresolvedAttribute is <> when:

  • AstBuilder is requested to sql/AstBuilder.md#visitDereference[visitDereference]

  • LogicalPlan is requested to spark-sql-LogicalPlan.md#resolve[resolve an attribute by name parts]

  • DescribeColumnCommand is DescribeColumnCommand.md#run[executed]

[[resolved]] UnresolvedAttribute can never be Expression.md#resolved[resolved] (and is replaced at <>).

[[analysis-phase]] [NOTE] ==== UnresolvedAttribute is resolved when Logical Analyzer is executed by the following logical resolution rules:

* ResolveSubquery

[[Unevaluable]][[eval]][[doGenCode]] Given UnresolvedAttribute can never be resolved it should not come as a surprise that it cannot be evaluated either (i.e. produce a value given an internal row). When requested to evaluate, UnresolvedAttribute simply reports a UnsupportedOperationException.

Cannot evaluate expression: [this]

[[creating-instance]] [[nameParts]] UnresolvedAttribute takes name parts when created.

[source, scala]

import org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute

scala> val t1 = UnresolvedAttribute("t1") t1: org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute = 't1

scala> val t2 = UnresolvedAttribute("db1.t2") t2: org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute = 'db1.t2

scala> println(s"Number of name parts: ${t2.nameParts.length}") Number of name parts: 2

scala> println(s"Name parts: ${t2.nameParts.mkString(",")}") Name parts: db1,t2


[[apply]] UnresolvedAttribute can be created with a fully-qualified name with dots to separate name parts.

[source, scala]

apply(name: String): UnresolvedAttribute

[TIP]

Use backticks (```) around names with dots (.`) to disable them as separators.

The following is a two-part attribute name with a.b and c name parts.

`a.b`.c

[[quoted]] UnresolvedAttribute can also be created without the dots with the special meaning.

[source, scala]

import org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute val attr1 = UnresolvedAttribute.quoted("a.b.c") scala> println(s"Number of name parts: ${attr1.nameParts.length}") Number of name parts: 1


[NOTE]

Catalyst DSL defines two Scala implicits to create an UnresolvedAttribute:

  • StringToAttributeConversionHelper is a Scala implicit class that converts $"colName" into an UnresolvedAttribute

  • symbolToUnresolvedAttribute is a Scala implicit method that converts 'colName into an UnresolvedAttribute

Both implicits are part of ExpressionConversions Scala trait of Catalyst DSL.

Import expressions object to get access to the expression conversions.

[source, scala]

// Use sbt console with Spark libraries defined (in build.sbt) import org.apache.spark.sql.catalyst.dsl.expressions._ scala> :type $"name" org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute

import org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute val nameAttr: UnresolvedAttribute = 'name

scala> :type nameAttr org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute


====

[NOTE]

A UnresolvedAttribute can be replaced by (resolved) a NamedExpression using an spark-sql-LogicalPlan.md#resolveQuoted[analyzed logical plan] (of the structured query the attribute is part of).

[source, scala]

val analyzedPlan = Seq((0, "zero")).toDF("id", "name").queryExecution.analyzed

import org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute val nameAttr = UnresolvedAttribute("name") val nameResolved = analyzedPlan.resolveQuoted( name = nameAttr.name, resolver = spark.sessionState.analyzer.resolver).getOrElse(nameAttr)

scala> println(nameResolved.numberedTreeString) 00 name#47: string

scala> :type nameResolved org.apache.spark.sql.catalyst.expressions.NamedExpression


====


Last update: 2021-02-18
Back to top