5 tips to build long-lasting Scala OSS Kazuhiro Sera @seratch Scale By The Bay 2018 

@seratch • My name is Kaz (Kazuhiro Sera) • Happy with Scala for 8 years • Came from Tokyo, Japan ! • Come and join ScalaMatsuri in June next year! • My 2nd talk at Scal(ae) By The Bay • Love open source software projects • Started ScalikeJDBC, Skinny Framework, AWScala, etc • Took over the maintenance of json4s, Scalate 

What is ScalikeJDBC? • My most long-lasting OSS (7 year history) • A relational database client library in Scala • Provide a “Scala-like” (Scala idiomatic) way • Practical: easy-to-use APIs + less dependencies • Brief history: • Started as a simple lib similar to Twitter’s querulous • Has more functionalities now • Scala 2.10: String Interpolation, Macros • SQL objects, one-to-x extraction, type-safe DSL 

Basic usage of ScalikeJDBC • Only 3 steps to issue a database query: • 1) Build a SQL object • 2) Define a function to extract values from ResultSet • 3) Perform SQL execution with an implicitly passed database session (which can be transaction-wired) 1)> val q: SQL[Nothing, NoExtractor] = sql"select id, name from members limit 2" 2)> val toTuple: (WrappedResultSet) => (Int, Option[String]) = | (rs) => (rs.get[Int]("id"), rs.get[Option[String]]("name")) 3)> val rows: Seq[(Int, Option[String])] = q.map(toTuple).list.apply() [SQL Execution] select id, name from members limit 2; (0 ms) rows: Seq[(Int, Option[String])] = List((1,Some(Alice)), (2,Some(Bob))) 

Other libraries? • According to the number of stars, • 1) Slick 2,124 • 2) Quill 1,311 • 3) Doobie 1,283 • 4) ScalikeJDBC 948 • ScalikeJDBC is ranked fourth (Not bad!) • Try in-memory database example on scalikejdbc.org • Any feedback? Reach out to me today 

ScalikeJDBC turned 7 yrs old! 

Happened during the time • Scala binary versions: 5 • 2.9.1 ~ 2.9.2 ~ 2.10 ~ 2.11 ~ 2.12 ~ (2.13.0-M5 ~) • (Prior to 2.10, every single patch version was bin-incompatible) • JDK major versions: 5 • 7 ~ 8(LTS) ~ 9 ~ 10 ~ 11(LTS) ~ • JDBC API spec minor updates: 3 • 4.1 (JDK7) ~ 4.2 (JDK8) ~ 4.3 (JDK9) • Added 4.2 APIs when dropping JDK7 support • sbt migration: 0.13 to 1 • For the sbt plugin generating source code from database 

5 tips to build long-lasting OSS • 1) Find your lifework project • 2) Be careful to add Scala dependencies • 3) Stick with binary compatibility • 4) Provide cross builds • 5) Have effective CI builds 

1) Find your lifework project • Not specific to Scala OSS • It’s all about the sustainability of voluntary works • Two things to know • Focus on lifework projects • Many projects terminated due to author’s social events • e.g. Changing job, having a family, school graduation • Can continue it regardless of those big changes? • Difficulty to find successors • Inconvenient truth; even if you have many users… • Ideally, continue working on your project somehow 

2) Be careful to add Scala deps • It’s unfortunate that I have to say this but it’s critical • Two things to know • "You’re in the same boat with many others” • Major upgrade: have to wait for all releases; it causes significant delay of your release • Cross builds: can’t do if just one of them doesn’t • Suggestion: Choose Java over Scala inside • Not cool… but I have to say it’s a realistic trade-off • Pros: Can make upgrade/cross builds much easier • Cons: Handle null values inside 

2’) Work with null in Scala <<< Working with java.util.List >>> scala> import scala.collection.JavaConverters._ import scala.collection.JavaConverters._ scala> val j = java.util.Arrays.asList("A", "B", "C") j: java.util.List[String] = [A, B, C] scala> val s = j.asScala s: scala.collection.mutable.Buffer[String] = Buffer(A, B, C) scala> null.asInstanceOf[java.util.List[String]].asScala // use Option! res0: scala.collection.mutable.Buffer[String] = null <<< Working with primitives >>> scala> val j: java.lang.Integer = null j: Integer = null scala> val s: Int = j // auto boxing s: Int = 0 scala> val s = Option(j).map(Int.unbox) s: Option[Int] = None java.util.List can be null. null.asScala returns null. A null value can be unexpectedly converted to the type’s default value. (0 : scala.Int’s default value) 

3) Stick with bin-compatibility • Bin-compatible? • Normal situation; not see any runtime linkage errors • If you break bin-compatibility: • Painful runtime errors (NoSuchMethodError etc) • Scala compiler cannot detect runtime issues beforehand • Two things to do • Define policies about bin-compatibility guarantee • Once you released the first stable version • Surely guarantee binary-compatibility • Solution: Use MiMa 

3’) Use MiMa to ensure that • Use MiMa (Migration Manager for Scala) • Can detect syntactic incompatibilities • Developed and maintained by Lightbend devs // project/plugins.sbt addSbtPlugin("com.typesafe" % "sbt-mima-plugin" % "0.3.0") // build.sbt project.settings( MimaPlugin.mimaDefaultSettings ++ Seq( mimaPreviousArtifacts := { Set("0.1.0", "0.1.1").map { v => organization.value % s"${name.value}_${scalaBinaryVersion.value}" % v } }, test in Test := { mimaReportBinaryIssues.value (test in Test).value } ) ) 

4) Provide cross builds • Cross builds = publish against multiple Scala bin versions • Reality: We cannot always use the latest Scala • Apache Spark took 2 years to support Scala 2.12 • Remember the tip 2) “be careful to add Scala deps” • Two things to do • Support latest stable major & its previous major • To be specific, Scala 2.12 and 2.11 at this moment • Don’t have many files under src/main/scala-X.Y • Testing a plenty of code there is hard+time-consuming 

5) Have effective CI builds • Why do we need it? • To reduce review cost of pull requests • To merge pull requests with confidence • You need reliable evidences (regressions, bin-combat, code style, etc) • Two things to do • Enable it anyway! • You can use TravisCI/CircleCI for free • Automate everything apart from code reviews • Check bin-compatibility, consistency of code style, cross builds, multiple JDKs, etc • Configure matrix builds to cover all patterns 

5’) TravisCI Build Example language: scala sudo: false scala: - 2.11.12 - 2.12.7 - 2.13.0-M5 jdk: - oraclejdk8 - openjdk11 matrix: include: - scala: 2.12.7 jdk: oraclejdk9 script: sbt ++$TRAVIS_SCALA_VERSION scalafmtCheck test # code formatter check + test • Scala 2.11 + JDK 8, 11 = 2 builds • Scala 2.12 + JDK 8, 9, 11 = 3 builds • Scala 2.13.0-M5 + JDK 8, 11 = 2 builds Additional 1 matrix build 3 Scala versions * 2 JDKs = 6 patterns 7 patterns in total 

Start with the sbt template! • sbt project template (aka g8 template) • Including all techniques in this talk • Should be a good starting point for you mkdir my-awesome-library cd my-awesome-library sbt new seratch/long-lasting-scala.g8 sbt test 

Recap • You found out about ScalikeJDBC • Contact me at any time during the conf • If you like it, increment the stars (1K soon!) • 5 tips to build long-lasting Scala OSS • 1) Find what interests you for years • 2) Be careful about adding Scala dependencies • 3) Stick with binary compatibility • 4) Provide cross builds • 5) Have effective CI builds