Scala application structure

Question

I am learning Scala now and I want to write some silly little app like a console Twitter client, or whatever. The question is, how to structure application on disk and logic

Answer 1

I'm going to disagree with Jens, here, though not all that much.

Project Layout

My own suggestion is that you model your efforts on Maven's standard directory layout.

Previous versions of SBT (before SBT 0.9.x) would create it automatically for you:

dcs@ayanami:~$ mkdir myproject
dcs@ayanami:~$ cd myproject
dcs@ayanami:~/myproject$ sbt
Project does not exist, create new project? (y/N/s) y
Name: myproject
Organization: org.dcsobral
Version [1.0]: 
Scala version [2.7.7]: 2.8.1
sbt version [0.7.4]: 
Getting Scala 2.7.7 ...
:: retrieving :: org.scala-tools.sbt#boot-scala
    confs: [default]
    2 artifacts copied, 0 already retrieved (9911kB/134ms)
Getting org.scala-tools.sbt sbt_2.7.7 0.7.4 ...
:: retrieving :: org.scala-tools.sbt#boot-app
    confs: [default]
    15 artifacts copied, 0 already retrieved (4096kB/91ms)
[success] Successfully initialized directory structure.
Getting Scala 2.8.1 ...
:: retrieving :: org.scala-tools.sbt#boot-scala
    confs: [default]
    2 artifacts copied, 0 already retrieved (15118kB/160ms)
[info] Building project myproject 1.0 against Scala 2.8.1
[info]    using sbt.DefaultProject with sbt 0.7.4 and Scala 2.7.7
> quit
[info] 
[info] Total session time: 8 s, completed May 6, 2011 12:31:43 PM
[success] Build completed successfully.
dcs@ayanami:~/myproject$ find . -type d -print
.
./project
./project/boot
./project/boot/scala-2.7.7
./project/boot/scala-2.7.7/lib
./project/boot/scala-2.7.7/org.scala-tools.sbt
./project/boot/scala-2.7.7/org.scala-tools.sbt/sbt
./project/boot/scala-2.7.7/org.scala-tools.sbt/sbt/0.7.4
./project/boot/scala-2.7.7/org.scala-tools.sbt/sbt/0.7.4/compiler-interface-bin_2.7.7.final
./project/boot/scala-2.7.7/org.scala-tools.sbt/sbt/0.7.4/compiler-interface-src
./project/boot/scala-2.7.7/org.scala-tools.sbt/sbt/0.7.4/compiler-interface-bin_2.8.0.RC2
./project/boot/scala-2.7.7/org.scala-tools.sbt/sbt/0.7.4/xsbti
./project/boot/scala-2.8.1
./project/boot/scala-2.8.1/lib
./target
./lib
./src
./src/main
./src/main/resources
./src/main/scala
./src/test
./src/test/resources
./src/test/scala

So you'll put your source files inside myproject/src/main/scala, for the main program, or myproject/src/test/scala, for the tests.

Since that doesn't work anymore, there are some alternatives:

giter8 and sbt.g8

Install giter8, clone ymasory's sbt.g8 template and adapt it to your necessities, and use that. See below, for example, this use of unmodified ymasory's sbt.g8 template. I think this is one of the best alternatives to starting new projects when you have a good notion of what you want in all your projects.

$ g8 ymasory/sbt
project_license_url [http://www.gnu.org/licenses/gpl-3.0.txt]:
name [myproj]:
project_group_id [com.example]:
developer_email [john.doe@example.com]:
developer_full_name [John Doe]:
project_license_name [GPLv3]:
github_username [johndoe]:

Template applied in ./myproj

$ tree myproj
myproj
├── build.sbt
├── LICENSE
├── project
│   ├── build.properties
│   ├── build.scala
│   └── plugins.sbt
├── README.md
├── sbt
└── src
    └── main
        └── scala
            └── Main.scala

4 directories, 8 files

np plugin

Use softprops's np plugin for sbt. In the example below, the plugin is configured on ~/.sbt/plugins/build.sbt, and its settings on ~/.sbt/np.sbt, with standard sbt script. If you use paulp's sbt-extras, you'd need to install these things under the right Scala version subdirectory in ~/.sbt, as it uses separate configurations for each Scala version. In practice, this is the one I use most often.

$ mkdir myproj; cd myproj
$ sbt 'np name:myproj org:com.example'
[info] Loading global plugins from /home/dcsobral/.sbt/plugins
[warn] Multiple resolvers having different access mechanism configured with same name 'sbt-plugin-releases'. To avoid conflict, Remove duplicate project resolvers (`resolvers`) or rename publishing resolver (`publishTo`).
[info] Set current project to default-c642a2 (in build file:/home/dcsobral/myproj/)
[info] Generated build file
[info] Generated source directories
[success] Total time: 0 s, completed Apr 12, 2013 12:08:31 PM
$ tree
.
├── build.sbt
├── src
│   ├── main
│   │   ├── resources
│   │   └── scala
│   └── test
│       ├── resources
│       └── scala
└── target
    └── streams
        └── compile
            └── np
                └── $global
                    └── out

12 directories, 2 files

mkdir

You could simply create it with mkdir:

$ mkdir -p myproj/src/{main,test}/{resource,scala,java}
$ tree myproj
myproj
└── src
    ├── main
    │   ├── java
    │   ├── resource
    │   └── scala
    └── test
        ├── java
        ├── resource
        └── scala

9 directories, 0 files

Source Layout

Now, about the source layout. Jens recommends following Java style. Well, the Java directory layout is a requirement -- in Java. Scala does not have the same requirement, so you have the option of following it or not.

If you do follow it, assuming the base package is org.dcsobral.myproject, then source code for that package would be put inside myproject/src/main/scala/org/dcsobral/myproject/, and so on for sub-packages.

Two common ways of diverging from that standard are:

• Omitting the base package directory, and only creating subdirectories for the sub-packages.

  For instance, let's say I have the packages org.dcsobral.myproject.model, org.dcsobral.myproject.view and org.dcsobral.myproject.controller, then the directories would be myproject/src/main/scala/model, myproject/src/main/scala/view and myproject/src/main/scala/controller.

• Putting everything together. In this case, all source files would be inside myproject/src/main/scala. This is good enough for small projects. In fact, if you have no sub-projects, it is the same as above.

And this deals with directory layout.

File Names

Next, let's talk about files. In Java, the practice is separating each class in its own file, whose name will follow the name of the class. This is good enough in Scala too, but you have to pay attention to some exceptions.

First, Scala has object, which Java does not have. A class and object of the same name are considered companions, which has some practical implications, but only if they are in the same file. So, place companion classes and objects in the same file.

Second, Scala has a concept known as sealed class (or trait), which limits subclasses (or implementing objects) to those declared in the same file. This is mostly done to create algebraic data types with pattern matching with exhaustiveness check. For example:

sealed abstract class Tree
case class Node(left: Tree, right: Tree) extends Tree
case class Leaf(n: Int) extends Tree

scala> def isLeaf(t: Tree) = t match {
     |     case Leaf(n: Int) => println("Leaf "+n)
     | }
<console>:11: warning: match is not exhaustive!
missing combination           Node

       def isLeaf(t: Tree) = t match {
                             ^
isLeaf: (t: Tree)Unit

If Tree was not sealed, then anyone could extend it, making it impossible for the compiler to know whether the match was exhaustive or not. Anyway, sealed classes go together in the same file.

Another naming convention is to name the files containing a package object (for that package) package.scala.

Importing Stuff

The most basic rule is that stuff in the same package see each other. So, put everything in the same package, and you don't need to concern yourself with what sees what.

But Scala also have relative references and imports. This requires a bit of an explanation. Say I have the following declarations at the top of my file:

package org.dcsobral.myproject
package model

Everything following will be put in the package org.dcsobral.myproject.model. Also, not only everything inside that package will be visible, but everything inside org.dcsobral.myproject will be visible as well. If I just declared package org.dcsobral.myproject.model instead, then org.dcsobral.myproject would not be visible.

The rule is pretty simple, but it can confuse people a bit at first. The reason for this rule is relative imports. Consider now the following statement in that file:

import view._

This import may be relative -- all imports can be relative unless you prefix it with _root_.. It can refer to the following packages: org.dcsobral.myproject.model.view, org.dcsobral.myproject.view, scala.view and java.lang.view. It could also refer to an object named view inside scala.Predef. Or it could be an absolute import refering to a package named view.

If more than one such package exists, it will pick one according to some precedence rules. If you needed to import something else, you can turn the import into an absolute one.

This import makes everything inside the view package (wherever it is) visible in its scope. If it happens inside a class, and object or a def, then the visibility will be restricted to that. It imports everything because of the ._, which is a wildcard.

An alternative might look like this:

package org.dcsobral.myproject.model
import org.dcsobral.myproject.view
import org.dcsobral.myproject.controller

In that case, the packages view and controller would be visible, but you'd have to name them explicitly when using them:

def post(view: view.User): Node =

Or you could use further relative imports:

import view.User

The import statement also enable you to rename stuff, or import everything but something. Refer to relevant documentation about it for more details.

So, I hope this answer all your questions.

Answer 2

Scala supports and encourages the package structure of Java /JVM and pretty much the same recommendation apply:

• mirror the package structure in the directory structure. This isn't necessary in Scala, but it helps to find your way around
• use your inverse domain as a package prefix. For me that means everything starts with de.schauderhaft. Use something that makes sense for you, if you don't have you own domain
• only put top level classes in one file if they are small and closely related. Otherwise stick with one class/object per file. Exceptions: companion objects go in the same file as the class. Implementations of a sealed class go into the same file.
• if you app grows you might want to have something like layers and modules and mirror those in the package structure, so you might have a package structure like this: <domain>.<module>.<layer>.<optional subpackage>.
• don't have cyclic dependencies on a package, module or layer level