Domain-specific Languages

When I started working with Scala, I did it all wrong. I learned Scala from oakies, so my team used Scala just as a syntax sugar for Java (like Kotlin or Xtend).

That’s a great way to strongly underuse Scala.

Scala isn’t Java. Scala binds to Java, but goes forth. Programming in Scala, you can access Java resources, but even the feeling smells different.

Both languages are object-oriented, but Java is an imperative language in the most strict sense, while Scala is impure functional, tending tightly to the pureness.

Another notable difference is that, even though both languages are general-purpose, Scala makes it possible (and easy) to create microlanguages for specific domains, called domain-specific languages, or just DSL – which’s very hard to do using Java.

Example

Just for instance, we’re gonna implement a very flat and incomplete set of SQL, in fact, just some SELECT resources. For the sake of simplicity, we aren’t taking care of sql injection or complex queries; The idea is enabling to represent a simple SELECT statement in a syntax as close as possible to SQL.

The target is:

Select("name", "surname") from "t_users" where Condition("surname") == "Doe" limit 10

It must serialise to:

SELECT name, surname FROM t_users WHERE surname = 'Doe' LIMIT 10

The condition

In order to build the Select, we need the From. To build the From, we need the Condition, so let’s start there.

However, in order to build the Condition, we need the Criteria, that represent “equals,” “not equals,” etc.

Let’s leave other criteria out and deal only with comparisons, “is null,” “not null.” The Criteria trait doesn’t need to be accessible outside of the Condition class, so it can be private:

private sealed trait Criteria {def format(key: String, value: String): String}

private object EQUALS     extends Criteria {def format(key:  String, value: String): String = s"$key = $value"}
private object GE         extends Criteria {def format(key:  String, value: String): String = s"$key >= $value"}
private object GT         extends Criteria {def format(key:  String, value: String): String = s"$key > $value"}
private object ISNULL     extends Criteria {def format(key:  String, value: String): String = s"$key ISNULL"}
private object LE         extends Criteria {def format(key:  String, value: String): String = s"$key <= $value"}
private object LT         extends Criteria {def format(key:  String, value: String): String = s"$key < $value"}
private object NOT_EQUALS extends Criteria {def format(key:  String, value: String): String = s"$key <> $value"}
private object NOTNULL    extends Criteria {def format(key:  String, value: String): String = s"$key NOTNULL"}

Now, the Condition must hold the field name, the comparing value, and the criteria. Since some conditions have no value, the value may be optional.

Let’s use a valueless criteria for default:

case class Condition[A](field: String, value: Option[A] = None) {

  private var criteria: Criteria = NOTNULL

  def ==(value: A): Condition[A] = {
    val res = Condition(field, Option(value))
    res.criteria = EQUALS
    res
  }

  def !=(value: A): Condition[A] = {
    val res = Condition(field, Option(value))
    res.criteria = NOT_EQUALS
    res
  }

  def >(value: A): Condition[A] = {
    val res = Condition(field, Option(value))
    res.criteria = GT
    res
  }

  def >=(value: A): Condition[A] = {
    val res = Condition(field, Option(value))
    res.criteria = GE
    res
  }

  def <(value: A): Condition[A] = {
    val res = Condition(field, Option(value))
    res.criteria = LT
    res
  }

  def <=(value: A): Condition[A] = {
    val res = Condition(field, Option(value))
    res.criteria = LE
    res
  }

  def isNull: Condition[A] = {
    val res = Condition(field, value)
    res.criteria = ISNULL
    res
  }

  def notNull: Condition[A] = {
    val res = Condition(field, value)
    res.criteria = NOTNULL
    res
  }

  override def toString: String = criteria format (field, value match {
    case None                => "NULL"
    case Some(value: String) => s"'$value'"
    case Some(value: Number) => value.toString
    case _                   => ???
  })
}

Going down to SELECT

The Select class is the simpliest of all, it just build a simple SELECT:

case class Select private(fields: String*) {

  def from(tables: String*): From = new From(this,tables: _*)

  override def toString: String = "SELECT %s" format fields.mkString(", ")
}

Creating the FROM

Now we got the condition, let’s deal with the From class. It must be able to deal with the SELECT syntax. We’re implementing only WHERE, GROUP BY,
ORDER BY, HAVING, LIMIT, and OFFSET.

Those parameters are represented by attributes:

• WHERE → conditions
• GROUP BY → _groupBy
• ORDER BY → _orderBy
• HAVING → _having
• LIMIT → _limit
• OFFSET → _offset

And the respective methods where, groupBy, orderBy, having, limit, and offset, each one returning a new instance of From.

For making it possible, we need to override the clone method too.

Most of the logic is gonna be in the toString method, responsible for build the SQL statement.

class From(val select: Select, val tables: String*) {

  private var conditions: Seq[Condition[_]] = Nil
  private var groupBy: Seq[String] = Nil
  private var orderBy: Seq[String] = Nil
  private var _having: Seq[Condition[_]] = Nil
  private var _offset: Int = 0
  private var _limit: Int = 0


  def where(conditions: Condition[_]*): From = {
    val res = clone
    res.conditions = conditions
    res
  }

  def groupBy(fields: String*): From = {
    val res = clone
    res.groupBy = fields
    res
  }

  def having(conditions: Condition[_]*): From = {
    val res = clone
    res._having = conditions
    res
  }

  def orderBy(fields: String*): From = {
    val res = clone
    res.orderBy = fields
    res
  }

  def offset(value: Int): From = {
    val res = clone
    res._offset = value
    res
  }

  def limit(value: Int): From = {
    val res = clone
    res._limit = value
    res
  }

  override def clone: From = {
    val res = new From(select, tables: _*)
    res.conditions = conditions
    res.groupBy = groupBy
    res.orderBy = orderBy
    res._offset = _offset
    res._limit = _limit
    res
  }

  override def toString: String = {
    val res = new StringBuilder
    res append select.toString
    res append " FROM "
    res append tables.mkString(", ")
    if (conditions.nonEmpty) {
      res append " WHERE "
      res append conditions.map {_.toString}.mkString(" AND ")
    }
    if (groupBy.nonEmpty) {
      res append " GROUP BY "
      if (groupBy.size == 1)
        res append groupBy.head.toString
      else {
        res append "("
        res append groupBy.map {_.toString}.mkString(", ")
        res append ")"
      }

      if (_having.nonEmpty) {
        res append " HAVING "
        res append _having.map {_.toString}.mkString(" AND ")
      }
    }
    if (orderBy.nonEmpty) {
      res append " ORDER BY "
      if (orderBy.size == 1)
        res append orderBy.head.toString
      else {
        res append "("
        res append orderBy.map {_.toString}.mkString(", ")
        res append ")"
      }
    }
    if (_limit > 0) {
      res append " LIMIT "
      res append _limit.toString
    }
    if (_offset > 0) {
      res append " OFFSET "
      res append _offset.toString
    }
    res.toString
  }
}

Conclusion

And it’s done! We’ve just created a very basic implementation of SELECT statement.

You can go on and implement all the SQL features just for fun, but that’s not this post’s target; I just wanna show you how Scala is powerful to design DSL.

---

Post from Kodumaro.