Github user marmbrus commented on a diff in the pull request:
https://github.com/apache/spark/pull/1759#discussion_r15790708
--- Diff: sql/core/src/main/scala/org/apache/spark/sql/TypedSql.scala ---
@@ -0,0 +1,202 @@
+package org.apache.spark.sql
+
+import org.apache.spark.sql.catalyst.analysis._
+import org.apache.spark.sql.catalyst.expressions.{Expression, ScalaUdf,
AttributeReference}
+import org.apache.spark.sql.catalyst.plans.logical.LocalRelation
+import org.apache.spark.sql.catalyst.types._
+
+import scala.language.experimental.macros
+import scala.language.existentials
+
+import records._
+import Macros.RecordMacros
+
+import org.apache.spark.annotation.Experimental
+import org.apache.spark.rdd.RDD
+import org.apache.spark.sql.catalyst.{SqlParser, ScalaReflection}
+
+/**
+ * A collection of Scala macros for working with SQL in a type-safe way.
+ */
+private[sql] object SQLMacros {
+ import scala.reflect.macros._
+
+ def sqlImpl(c: Context)(args: c.Expr[Any]*) =
+ new Macros[c.type](c).sql(args)
+
+ case class Schema(dataType: DataType, nullable: Boolean)
+
+ class Macros[C <: Context](val c: C) extends ScalaReflection {
+ val universe: c.universe.type = c.universe
+
+ import c.universe._
+
+ val rowTpe = tq"_root_.org.apache.spark.sql.catalyst.expressions.Row"
+
+ val rMacros = new RecordMacros[c.type](c)
+
+ trait InterpolatedItem {
+ def placeholderName: String
+ def registerCode: Tree
+ def localRegister(catalog: Catalog, registry: FunctionRegistry)
+ }
+
+ case class InterpolatedUDF(index: Int, expr: c.Expr[Any], returnType:
DataType)
+ extends InterpolatedItem{
+
+ val placeholderName = s"func$index"
+
+ def registerCode = q"""registerFunction($placeholderName, $expr)"""
+
+ def localRegister(catalog: Catalog, registry: FunctionRegistry) = {
+ registry.registerFunction(
+ placeholderName, (_: Seq[Expression]) => ScalaUdf(null,
returnType, Nil))
+ }
+ }
+
+ case class InterpolatedTable(index: Int, expr: c.Expr[Any], schema:
StructType)
+ extends InterpolatedItem{
+
+ val placeholderName = s"table$index"
+
+ def registerCode = q"""$expr.registerTempTable($placeholderName)"""
+
+ def localRegister(catalog: Catalog, registry: FunctionRegistry) = {
+ catalog.registerTable(None, placeholderName,
LocalRelation(schema.toAttributes :_*))
+ }
+ }
+
+ case class RecSchema(name: String, index: Int, cType: DataType, tpe:
Type)
+
+ def sql(args: Seq[c.Expr[Any]]) = {
+
+ val q"""
+ $interpName(
+ scala.StringContext.apply(..$rawParts))""" = c.prefix.tree
+
+ //rawParts.map(_.toString).foreach(println)
+
+ val parts =
+ rawParts.map(
+ _.toString.stripPrefix("\"")
+ .replaceAll("\\\\", "")
+ .stripSuffix("\""))
+
+ val interpolatedArguments = args.zipWithIndex.map { case (arg, i) =>
+ // println(arg + " " + arg.actualType)
+ arg.actualType match {
+ case TypeRef(_, _, Seq(schemaType)) =>
+ InterpolatedTable(i, arg,
schemaFor(schemaType).dataType.asInstanceOf[StructType])
+ case TypeRef(_, _, Seq(inputType, outputType)) =>
+ InterpolatedUDF(i, arg, schemaFor(outputType).dataType)
+ }
+ }
+
+ val query = parts(0) + args.indices.map { i =>
+ interpolatedArguments(i).placeholderName + parts(i + 1)
+ }.mkString("")
+
+ val parser = new SqlParser()
+ val logicalPlan = parser(query)
+ val catalog = new SimpleCatalog(true)
+ val functionRegistry = new SimpleFunctionRegistry
+ val analyzer = new Analyzer(catalog, functionRegistry, true)
+
+ interpolatedArguments.foreach(_.localRegister(catalog,
functionRegistry))
+ val analyzedPlan = analyzer(logicalPlan)
+
+ val fields = analyzedPlan.output.map(attr => (attr.name,
attr.dataType))
+ val record = genRecord(q"row", fields)
+
+ val tree = q"""
+ ..${interpolatedArguments.map(_.registerCode)}
+ val result = sql($query)
+ result.map(row => $record)
+ """
+
+ // println(tree)
+ c.Expr(tree)
+ }
+
+ // TODO: Handle nullable fields
--- End diff --
Your point about changing return type based on complex static analysis is
well taken and that is my hesitation as well. That said...
My though was that you could do something that is similar to a type
ascription by adding a `WHERE a IS NOT NULL` or similar to the query whenever
you don't want to deal with the option type. This forces the programmer to
explicitly denote a handling for null values (filter them out).
Regarding joins, for inner joins you won't change the nullability of any
output attributes so it'll still relate to the database schema. For outer
joins I think we *should* be forcing the programmer to explicitly deal with the
fact that they are introducing nullability though their choice of join.
Another possibility here would be to have two interpolators, one with
boxing costs but simple semantics and one with explicit Options or primitives
based on the SQL analysis.
---
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]
