Upgrade Guides

Compatibility Policy

Slick requires Scala 2.10 or 2.11. (For Scala 2.9 please use ScalaQuery, the predecessor of Slick).

Slick version numbers consist of an epoch, a major and minor version, and possibly a qualifier (for milestone, RC and SNAPSHOT versions).

For release versions (i.e. versions without a qualifier), backward binary compatibility is guaranteed between releases with the same epoch and major version (e.g. you could use 2.1.2 as a drop-in relacement for 2.1.0 but not for 2.0.0). Slick Extensions requires at least the same minor version of Slick (e.g. Slick Extensions 2.1.2 can be used with Slick 2.1.2 but not with Slick 2.1.1). Binary compatibility is not preserved for slick-codegen, which is generally used at compile-time.

We do not guarantee source compatibility but we try to preserve it within the same major release. Upgrading to a new major release may require some changes to your sources. We generally deprecate old features and keep them around for a full major release cycle (i.e. features which become deprecated in 2.1.0 will not be removed before 2.2.0) but this is not possible for all kinds of changes.

Release candidates have the same compatibility guarantees as the final versions to which they lead. There are no compatibility guarantees whatsoever for milestones and snapshots.

Upgrade from 2.1 to 3.0

Package Structure

Slick has moved from package scala.slick to slick. A package object in scala.slick provides deprecated aliases for many common types and values.

Database I/O Actions

The simple and Implicits imports from drivers are deprecated and will be removed in Slick 3.1. You should use api instead, which will give you the same features, except for the old Invoker and Executor APIs for blocking execution of database calls. These have been replaced by a new monadic database I/O actions API. See Database I/O Actions for details of the new API.

Join Operators

The old outer join operators did not handle null values correctly, requiring complicated mappings in user code, especially when using nested outer joins or outer joins over mapped entities. This is no longer necessary with the new outer join operators that lift one (left or right outer join) or both sides (full outer join) of the join into an Option. This is made possible by the new nested Options and non-primitive Options support in Slick.

The old operators are deprecated but still available. Deprecation warnings will point you to the right replacement:

  • leftJoin -> joinLeft
  • rightJoin -> joinRight
  • outerJoin -> joinFull
  • innerJoin -> join

Passing an explicit JoinType to the generic join operator does not make sense anymore with the new join semantics and is therefore deprecated, too. join is now used exclusively for inner joins.

first

The old Invoker API used the first and firstOption methods to get the first element of a collection-valued query. The same operations for streaming Actions in the new API are called head and headOption respectively, consistent with the names used by the Scala Collections API.

Column Type

The type Column[T] has been subsumed into its supertype Rep[T]. For operations which are only available for individual columns, an implicit TypedType[T] evidence is required. The more flexible handling of Option columns requires Option and non-Option columns to be treated differently when creating an implicit Shape. In this case the evidence needs to be of type OptionTypedType[T] or BaseTypedType[T], respectively. If you want to abstract over both, it may be more convenient to pass the required Shape as an implicit parameter and let it be instantiated at the call site where the concrete type is known.

Column[T] is still available as a deprecated alias for Rep[T]. Due to the required implicit evidence, it cannot provide complete backwards compatibility in all cases.

Closing Databases

Since a Database instance can now have an associated connection pool and thread pool, it is important to call shutdown or close when you are done using it, so that these pools can be shut down properly. You should take care to do this when you migrate to the new action-based API. As long as you exclusively use the deprecated synchronous API, it is not strictly necessary.

Warning

Do not rely on the lazy initialization! Slick 3.1 will require Database objects to always be closed and may create connection and thread pool immediately.

Metadata API and Code Generator

The JDBC metadata API in package slick.jdbc.meta has been switched to the new API, producing Actions instead of Invokers. The code generator, which uses this API, has been completely rewritten for the asynchronous API. It still supports the same functionality and the same concepts but any customization of the code generator will have to be changed. See the code generator tests and the Schema Code Generation chapter for examples.

Inserting from Queries and Expressions

In Slick 2.0, soft inserts (where auto-incrementing columns are ignored) became the default for inserting raw values. Inserting from another query or a computed expression still uses force-insert semantics (i.e. trying to insert even into auto-incrementing columns, whether or not the database supports it). The new DBIO API properly reflects this by renaming insert(Query) to forceInsertQuery(Query) and insertExpr to forceInsertExpr.

Default String Types

The default type for String columns of unconstrained length in JdbcProfile has traditionally been VARCHAR(254). Some drivers (like H2Driver) already changed it into an unconstrained string type. Slick 3.0 now also uses VARCHAR on PostgreSQL and TEXT on MySQL. The former should be harmless but MySQL’s TEXT type is similar to CLOB and has some limitations (e.g. no default values and no index without a prefix length). You can use an explicit O.Length(254) column option to go back to the previous behavior or change the default in the application.conf key slick.driver.MySQL.defaultStringType.

JdbcDriver

The JdbcDriver object has been deprecated. You should always use the correct driver for your database system.

Upgrade from 2.0 to 2.1

Query type parameters

Query now takes 3 type parameters instead of two. 2.0’s Query[T,E] is equivalent to Slick 2.1’s Query[T,E,Seq]. The third parameter is the collection type to be returned when executing the query using .run, which always returned a Seq in Slick 2.0. This is the only place where it is used right now. In the future we will work on making queries correspond to the behavior of the corresponding Scala collection types, i.e. Query[_,_,Set] having the uniqueness property, Query[_,_,List] being order preserving, etc. The collecton type can be changed to C by calling .to[C] on a query.

To upgrade your code to 2.1 you can either rename the new Query type to something else in the import, i.e. import ....simple.{Query=>NewQuery,_} and then write a type alias type Query[T,E] = NewQuery[T,E,Seq]. Or you can add Seq as the third type argument in your code. This regex should work for most places: replace ([^a-zA-Z])Query\[([^\]]+), ?([^\]]+)\] with \1Query[\2, \3, Seq].

.list and .first

These methods had an empty argument list before the implicit argument list in 2.0. This has been dropped for uniformity. Calls like .list() need to be replaced with .list and .first() by .first.

.where

This method has been deprecated in favor of the Scala collections conformant .filter method.

.isNull and .isNotNull

These methods have been deprecated in favor of new Scala standard library conformant isEmpty and isDefined methods. They can now only be used on Option columns. Otherwise you get a type error. A quick workaround for using them on non-Option columns is casting them into Option columns using .?, e.g. someCol.?.isDefined. The reason that you have to do this points to using a wrong type for your column however, i.e. non-Option for a nullable column and should really be fixed in your Table definition.

Static Plain SQL Queries

The interface for using argument placeholders has been changed. Where in 2.0 you could write

StaticQuery.query[T,…]("select ...").list(someT)

you now have to write

StaticQuery.query[T,…]("select ...").apply(someT).list

Slick code generator / Slick model

The code generator has been moved into a separate artifact in order to evolve it faster than Slick core. it moved from package slick.model.codegen to package slick.codegen. Binary compatibility will not be guaranteed, as it is supposed to be used before compile time. Add

"com.typesafe.slick" %% "slick-codegen" % "3.0.0"

to the dependencies of your code generator sbt project.

Method SourceCodeGenerator#Table#compound has been replaced by two methods compoundValue and compoundType generating potentially differently shaped code for values and types of compound values.

Method getTables of the Slick drivers, which returns an Invoker for listing all default database tables has been deprecated in favor of new method defautTables, which returns the tables directly in order to allow Slick to exclude meta tables at this point.

Method slick.jdbc.meta.createModel(tables) has been moved into the drivers and can now be invoked using e.g. H2Driver.createModel(Some(tables))

The model generated by Slick now contains improved information like the database column type, length of string columns, default values for strings in MySQL. The code generator will embed the portable length into generated code and can optionally embed the non-portable database column type into generated code when overriding SlickCodeGenerator#Table#Column#dbType with true.

The new ModelBuilder can be extended to customize model creation from jdbc meta data, similar to how the code generator can be customized. This allows working around differences and bugs in jdbc drivers, when creating the model or making up for missing features in Slick, e.g supporting specific types of your dbms of choice.

Upgrade from 1.0 to 2.0

Slick 2.0 contains some improvements which are not source compatible with Slick 1.0. When migrating your application from 1.0 to 2.0, you will likely need to perform changes in the following areas.

Code Generation

Instead of writing your table descriptions or plain SQL mappers by hand, in 2.0 you can now automatically generate them from your database schema. The code-generator is flexible enough to customize it’s output to fit exactly what you need. More info on code generation.

Table Descriptions

In Slick 1.0 tables were defined by a single val or object (called the table object) and the * projection was limited to a flat tuple of columns that had to be constructed with the special ~ operator:

// --------------------- Slick 1.0 code -- does not compile in 2.0 ---------------------

object Suppliers extends Table[(Int, String, String)]("SUPPLIERS") {
  def id = column[Int]("SUP_ID", O.PrimaryKey)
  def name = column[String]("SUP_NAME")
  def street = column[String]("STREET")
  def * = id ~ name ~ street
}

In Slick 2.0 you need to define your table as a class that takes an extra Tag argument (the table row class) plus an instance of a TableQuery of that class (representing the actual database table). Tuples for the * projection can use the standard tuple syntax:

class Suppliers(tag: Tag) extends Table[(Int, String, String)](tag, "SUPPLIERS") {
  def id = column[Int]("SUP_ID", O.PrimaryKey)
  def name = column[String]("SUP_NAME")
  def street = column[String]("STREET")
  def * = (id, name, street)
}
val suppliers = TableQuery[Suppliers]

You can import TupleMethods._ to get support for the old ~ syntax. The simple TableQuery[T] syntax is a macro which expands to a proper TableQuery instance that calls the table’s constructor (new TableQuery(new T(_))). In Slick 1.0 it was common practice to place extra static methods associated with a table into that table’s object. You can do the same in 2.0 with a custom TableQuery object:

object suppliers extends TableQuery(new Suppliers(_)) {
  // put extra methods here, e.g.:
  val findByID = this.findBy(_.id)
}

Note that a TableQuery is a Query for the table. The implicit conversion from a table row object to a Query that could be applied in unexpected places is no longer needed or available. All the places where you had to use the raw table object in Slick 1.0 have been changed to use the table query instead, e.g. inserting (see below) or foreign key references.

The method for creating simple finders has been renamed from createFinderBy to findBy. It is defined as an extension method for TableQuery, so you have to prefix the call with this. (see code snippet above).

Mapped Tables

In 1.0 the <> method for bidirectional mappings was overloaded for different arities so you could directly pass a case class’s apply method to it:

// --------------------- Slick 1.0 code -- does not compile in 2.0 ---------------------

def * = id ~ name ~ street <> (Supplier _, Supplier.unapply)

This is no longer supported in 2.0. One of the reasons is that the overloading led to complicated error messages. You now have to use a function with an appropriate tuple type. If you map to a case class you can simply use .tupled on its companion object:

def * = (id, name, street) <> (Supplier.tupled, Supplier.unapply)

Note that .tupled is only available for proper Scala functions. In 1.0 it was sufficient to have a method like apply that could be converted to a function on demand (<> (Supplier.apply _, Supplier.unapply)).

When using a case class, the companion object extends the correct function type by default, but only if you do not define the object yourself. In that case you should provide the right supertype manually, e.g.:

case class Supplier(id: Int, name: String, street: String)

object Supplier // overriding the default companion object
  extends ((Int, String, String) => Supplier) { // manually extending the correct function type
  //...
}

Alternatively, you can have the Scala compiler first do the lifting to a function and then call .tupled:

def * = (id, name, street) <> ((Supplier.apply _).tupled, Supplier.unapply)

Profile Hierarchy

Slick 1.0 provided two profiles, BasicProfile and ExtendedProfile. These two have been unified in 2.0 as JdbcProfile. Slick now provides more abstract profiles, in particular RelationalProfile which does not have all the features of JdbcProfile but is supported by the new HeapDriver and DistributedDriver. When porting code from Slick 1.0, you generally want to switch to JdbcProfile when abstracting over drivers. In particular, pay attention to the fact that BasicProfile in 2.0 is very different from BasicProfile in 1.0.

Inserting

In Slick 1.0 you used to construct a projection for inserting from the table object:

// --------------------- Slick 1.0 code -- does not compile in 2.0 ---------------------

(Suppliers.name ~ Suppliers.street) insert ("foo", "bar")

Since there is no raw table object any more in 2.0 you have to use a projection from the table query:

suppliers.map(s => (s.name, s.street)) += ("foo", "bar")

Note the use of the new += operator for API compatibility with Scala collections. The old name insert is still available as an alias.

Slick 2.0 will now automatically exclude AutoInc fields by default when inserting data. In 1.0 it was common to have a separate projection for inserts in order to exclude these fields manually:

// --------------------- Slick 1.0 code -- does not compile in 2.0 ---------------------

case class Supplier(id: Int, name: String, street: String)

object Suppliers extends Table[Supplier]("SUPPLIERS") {
  def id = column[Int]("SUP_ID", O.PrimaryKey, O.AutoInc)
  def name = column[String]("SUP_NAME")
  def street = column[String]("STREET")
  // Map a Supplier case class:
  def * = id ~ name ~ street <> (Supplier.tupled, Supplier.unapply)
  // Special mapping without the 'id' field:
  def forInsert = name ~ street <> (
    { case (name, street) => Supplier(-1, name, street) },
    { sup => (sup.name, sup.street) }
  )
}

Suppliers.forInsert.insert(mySupplier)

This is no longer necessary in 2.0. You can simply insert using the default projection and Slick will skip the auto-incrementing id column:

case class Supplier(id: Int, name: String, street: String)

class Suppliers(tag: Tag) extends Table[Supplier](tag, "SUPPLIERS") {
  def id = column[Int]("SUP_ID", O.PrimaryKey, O.AutoInc)
  def name = column[String]("SUP_NAME")
  def street = column[String]("STREET")
  def * = (id, name, street) <> (Supplier.tupled, Supplier.unapply)
}
val suppliers = TableQuery[Suppliers]

suppliers += mySupplier

If you really want to insert into an AutoInc field, you can use the new methods forceInsert and forceInsertAll.

Pre-compiled Updates

Slick now supports pre-compilation of updates in the same manner like selects, see Compiled Queries.

Database and Session Handling

In Slick 1.0, the common JDBC-based Database and Session types, as well as the Database factory object, could be found in the package slick.session. Since Slick 2.0 is no longer restricted to JDBC-based databases, this package has been replaced by the new DatabaseComponent (a.k.a. backend) hierarchy. If you work at the JdbcProfile abstraction level, you will always use a JdbcBackend from which you can import the types that were previously found in slick.session. Note that importing simple._ from a driver will automatically bring these types into scope.

Dynamically and Statically Scoped Sessions

Slick 2.0 still supports both, thread-local dynamic sessions and statically scoped sessions, but the syntax has changed to make the recommended way of using statically scoped sessions more concise. The old threadLocalSession is now called dynamicSession and the overloads of the associated session handling methods withSession and withTransaction have been renamed to withDynSession and withDynTransaction respectively. If you used this pattern in Slick 1.0:

// --------------------- Slick 1.0 code -- does not compile in 2.0 ---------------------

import scala.slick.session.Database.threadLocalSession

myDB withSession {
  // use the implicit threadLocalSession here
}

You have to change it for Slick 2.0 to:

import slick.jdbc.JdbcBackend.Database.dynamicSession

myDB withDynSession {
  // use the implicit dynamicSession here
}

On the other hand, due to the overloaded methods, Slick 1.0 required an explicit type annotation when using the statically scoped session:

myDB withSession { implicit session: Session =>
  // use the implicit session here
}

This is no longer necessary in 2.0:

myDB withSession { implicit session =>
  // use the implicit session here
}

Again, the recommended practice is NOT to use dynamic sessions. If you are uncertain if you need them the answer is most probably no. Static sessions are safer.

Mapped Column Types

Slick 1.0’s MappedTypeMapper has been renamed to MappedColumnType. Its basic form (using MappedColumnType.base) is now available at the RelationalProfile level (with more advanced uses still requiring JdbcProfile). The idiomatic use in Slick 1.0 was:

// --------------------- Slick 1.0 code -- does not compile in 2.0 ---------------------

case class MyID(value: Int)

implicit val myIDTypeMapper =
  MappedTypeMapper.base[MyID, Int](_.value, new MyID(_))

This has changed to:

case class MyID(value: Int)

implicit val myIDColumnType =
  MappedColumnType.base[MyID, Int](_.value, new MyID(_))

If you need to map a simple wrapper type (as shown in this example), you can now do that in an easier way by extending MappedTo:

case class MyID(value: Int) extends MappedTo[Int]

// No extra implicit required any more