Skip to content

Commit

Permalink
Adds Inserting and Updating section
Browse files Browse the repository at this point in the history
  • Loading branch information
franciscodr committed Aug 7, 2016
1 parent 4f7d7a1 commit 0310c0b
Show file tree
Hide file tree
Showing 6 changed files with 368 additions and 13 deletions.
5 changes: 3 additions & 2 deletions src/main/scala/doobie/DoobieLibrary.scala
Expand Up @@ -11,12 +11,13 @@ object DoobieLibrary extends Library {
override def owner: String = "scala-exercises" override def owner: String = "scala-exercises"
override def repository: String = "exercises-doobie" override def repository: String = "exercises-doobie"


override def color = Some("#5B5988") override def color = Some("#E35E31")


override def sections: List[Section] = List( override def sections: List[Section] = List(
ConnectingToDatabaseSection, ConnectingToDatabaseSection,
SelectingDataSection, SelectingDataSection,
MultiColumnQueriesSection, MultiColumnQueriesSection,
ParameterizedQueriesSection ParameterizedQueriesSection,
UpdatesSection
) )
} }
6 changes: 3 additions & 3 deletions src/main/scala/doobie/DoobieUtils.scala
@@ -1,10 +1,10 @@
package doobie package doobie


import java.sql.Connection

import doobie.Model._ import doobie.Model._
import doobie.imports._
import doobie.free.{ drivermanager => FD } import doobie.free.{ drivermanager => FD }

import doobie.imports._
import java.sql.Connection


import scalaz.concurrent.Task import scalaz.concurrent.Task
import scalaz._, Scalaz._ import scalaz._, Scalaz._
Expand Down
238 changes: 238 additions & 0 deletions src/main/scala/doobie/UpdatesSection.scala
@@ -0,0 +1,238 @@
package doobie

import doobie.UpdatesSectionHelpers._
import doobie.imports._
import org.scalaexercises.definitions.Section
import org.scalatest.{FlatSpec, Matchers}

import scalaz._, Scalaz._

/**
* In this section we examine operations that modify data in the database, and ways to retrieve the
* results of these updates.
*
* ==Data Definition==
* It is uncommon to define database structures at runtime, but '''doobie''' handles it just fine
* and treats such operations like any other kind of update. And it happens to be useful here!.
*
* Let’s create a new table, which we will use for the exercises to follow. This looks a lot like
* our prior usage of the `sql` interpolator, but this time we’re using `update` rather than
* `query`. The `.run` method gives a `ConnectionIO[Int]` that yields the total number of rows
* modified
*
* {{{
* val drop: Update0 =
* sql"""
* DROP TABLE IF EXISTS person
* """.update
*
* val create: Update0 =
* sql"""
* CREATE TABLE person (
* id SERIAL,
* name VARCHAR NOT NULL UNIQUE,
* age SMALLINT
* )
* """.update
* }}}
*
* We can compose these and run them together.
* {{{
* (drop.run *> create.run).transact(xa).unsafePerformSync
* }}}
*
* ==Inserting==
* Inserting is straightforward and works just as with selects. Here we define a method that
* constructs an `Update0` that inserts a row into the `person` table.
*
* {{{
* def insert1(name: String, age: Option[Short]): Update0 =
* sql"insert into person (name, age) values ($name, $age)".update
* }}}
*
* @param name inserting_and_updating
*/
object UpdatesSection extends FlatSpec with Matchers with Section {

/**
* Let's insert a new row by using the recently defined `insert1` method.
*/
def insertOneRow(res0: Int) = {

val insertedRows =
insert1("John", Option(35))
.run
.transact(xa)
.run

insertedRows should be(res0)
}

/**
* On the contrary, if we want to insert several rows, there are different ways to do that. A
* first try could be to use a `for-comprehension` to compose all the single operations.
*/

def insertSeveralRowsWithForComprehension(res0: Int) = {

val rows = for {
row1 <- insert1("Alice", Option(12)).run
row2 <- insert1("Bob", None).run
row3 <- insert1("John", Option(17)).run
} yield row1 + row2 + row3

val insertedRows = rows
.transact(xa)
.run

insertedRows should be(res0)
}

/**
* If there is no dependency between the SQL operations, it could be better to use an applicative
* functor.
*/
def insertSeveralRowsWithApplicativeFunctor(res0: Int) = {

val insertedOnePerson = insert1("Alice", Option(12)).run

val insertedOtherPerson = insert1("Bob", None).run

val insertedRows = (insertedOnePerson |@| insertedOtherPerson)(_ + _ )
.transact(xa)
.run

insertedRows should be(res0)
}

/**
* If all the data to be inserted is represented by a `List`, other way could be by using the
* Scalaz `traverse` method.
*/
def insertSeveralRowsWithTraverse(res0: Int) = {

val people = List(
("Alice", Option(12)),
("Bob", None),
("John", Option(17)),
("Mary", Option(16)))

val insertedRows = people
.traverse(item => (insert1 _).tupled(item).run)
.transact(xa)
.run

insertedRows.sum should be(res0)
}

/**
* ==Updating==
* Updating follows the same pattern. For instance, we suppose that we want to modify the age of
* a person.
*/
def updateExistingRow(res0: Int, res1: Int, res2: Int) = {

val result = for {
insertedRows <- insert1("Alice", Option(12)).run
updatedRows <- sql"update person set age = 15 where name = 'Alice'".update.run
person <- sql"select id, name, age from person where name = 'Alice'".query[Person].unique
} yield (insertedRows, updatedRows, person)

val (insertedRows, updatedRows, person) = result
.transact(xa)
.run

insertedRows should be(res0)
updatedRows should be(res1)
person.age should be(Option(res2))
}

/**
* ==Retrieving info==
* When we insert we usually want the new row back, so let’s do that. First we’ll do it the hard
* way, by inserting, getting the last used key via `lastVal()`, then selecting the indicated
* row.
* {{{
* def insert2(name: String, age: Option[Short]): ConnectionIO[Person] =
* for {
* _ <- sql"insert into person (name, age) values ($name, $age)".update.run
* id <- sql"select lastval()".query[Long].unique
* p <- sql"select id, name, age from person where id = $id".query[Person].unique
* } yield p
* }}}
*
* This is irritating but it is supported by all databases (although the “get the last used id”
* function will vary by vendor).
*
* Some database (like H2) allow you to return [only] the inserted id, allowing the above
* operation to be reduced to two statements (see below for an explanation of
* `withUniqueGeneratedKeys`).
*/
def retrieveInfo(res0: String, res1: Int) = {

def insert2_H2(name: String, age: Option[Int]): ConnectionIO[Person] =
for {
id <- sql"insert into person (name, age) values ($name, $age)".update.withUniqueGeneratedKeys[Int]("id")
p <- sql"select id, name, age from person where id = $id".query[Person].unique
} yield p

val person = insert2_H2("Ramone", Option(42))
.transact(xa)
.run


person.name should be(res0)
person.age should be(Option(res1))
}

/**
* Other databases (including PostgreSQL) provide a way to do this in one shot by returning
* multiple specified columns from the inserted row.
*
* {{{
* def insert3(name: String, age: Option[Int]): ConnectionIO[Person] = {
* sql"insert into person (name, age) values ($name, $age)"
* .update.withUniqueGeneratedKeys("id", "name", "age")
* }
* }}}
*
* The `withUniqueGeneratedKeys` specifies that we expect exactly one row back (otherwise an
* exception will be thrown), and requires a list of columns to return.
*
* This mechanism also works for updates, for databases that support it. In the case of multiple
* row updates we use `withGeneratedKeys` and get a Process[ConnectionIO, Person] back.
*
* == Batch Updates ==
* '''doobie''' supports batch updating via the `updateMany` and `updateManyWithGeneratedKeys`
* operations on the `Update` data type.
*
* By using an `Update` directly we can apply many sets of arguments to the same statement, and
* execute it as a single batch operation.
*
* - `updateMany` will return the updated of affected rows
*
* - For databases that support it (such as PostgreSQL) we can use `updateManyWithGeneratedKeys`
* to return a stream of updated rows.
*/

def batchUpdates(res0: Int) = {
type PersonInfo = (String, Option[Short])

def insertMany(ps: List[PersonInfo]): ConnectionIO[Int] = {
val sql = "insert into person (name, age) values (?, ?)"
Update[PersonInfo](sql).updateMany(ps)
}

// Some rows to insert
val data = List[PersonInfo](
("Frank", Some(12)),
("Daddy", None))

val insertedRows = insertMany(data)
.transact(xa)
.run

insertedRows should be(res0)
}

}
42 changes: 42 additions & 0 deletions src/main/scala/doobie/UpdatesSectionHelpers.scala
@@ -0,0 +1,42 @@
package doobie

import doobie.imports._
import doobie.free.{ drivermanager => FD }

import java.sql.Connection

import scalaz.concurrent.Task
import scalaz._, Scalaz._

object UpdatesSectionHelpers {

case class Person(id: Long, name: String, age: Option[Short])

val xa = new Transactor[Task] {

val driver = "org.h2.Driver"
def url = "jdbc:h2:mem:"
val user = "sa"
val pass = ""

val connect: Task[Connection] =
Task.delay(Class.forName(driver)) *> FD.getConnection(url, user, pass).trans[Task]

val create: Update0 =
sql"""
CREATE TABLE person (
id IDENTITY,
name VARCHAR NOT NULL UNIQUE,
age INT)
""".update

override val before =
super.before <* create.run

}

def insert1(name: String, age: Option[Int]): Update0 =
sql"""
INSERT INTO person (name, age) VALUES ($name, $age)
""".update
}
17 changes: 9 additions & 8 deletions src/test/scala/doobie/SelectingDataSectionSpec.scala
Expand Up @@ -25,22 +25,23 @@ class SelectingDataSectionSpec extends Spec with Checkers {
) )
) )
} }

def `select optional country name` = { def `select optional country name` = {
val value: Option[String] = None val value: Option[String] = None
check( check(
Test.testSuccess( Test.testSuccess(
SelectingDataSection.selectOptionalCountryName _, SelectingDataSection.selectOptionalCountryName _,
value :: HNil value :: HNil
) )
) )
} }


def `select unique country name` = { def `select unique country name` = {
check( check(
Test.testSuccess( Test.testSuccess(
SelectingDataSection.selectUniqueCountryName _, SelectingDataSection.selectUniqueCountryName _,
"Spain" :: HNil "Spain" :: HNil
) )
) )
} }


Expand Down

0 comments on commit 0310c0b

Please sign in to comment.