Skip to content

Command-line arguments reader to a class with respective fields

License

Notifications You must be signed in to change notification settings

krzys9876/command-line-reader

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

28 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Scala command-line reader

Convert arguments to fields within a simple class

This short project is an exercise of scala implicit type conversions with a bit of java reflection.

If you're using your code in a way that the key configurations are passed as program arguments (this is what I do with my spark applications) you've probably already came across libraries that hide all the complexity of this (not that simple) task.

I usually try to understand the mechanics of the task before I decide if I should use an external library or write some code on my own. This is a result of a study of scala implicits (which I find extremely useful) and java reflection, which I already knew from other projects.

The idea is to allow a developer to create the simplest possible class which contains fields corresponding to every argument. These fields should be easily definable (e.g. required/optional) and accessible (without too much boilerplate). Both scala files are close to 150 lines, which I guess qualifies them as simple. There would be even less without parsing additional types or printing instructions

Example

Consider an example argument list:

--input-file="/tmp/some_folder/input.bin" --algorithm=gz --iterations=10 --output-file="/tmp/some_other_folder/output.bin"

You may define a class like this:

class SampleArgs(args:Array[String]) extends ArgsAsClass(args) {
  val inputFile:Argument[String]=Argument.required
  val algorithm:Argument[String]=Argument.required
  val iterations:Argument[Int]=Argument.optional(5)
  val outputFile:Argument[String]=Argument.required
  val verbose:Argument[Boolean]=Argument.optional(false)

  parse()
}

Invocation of the method parse() at the end of the class body is actually the only boilerplate (apart from type declaration). It fills all the required values just after instantiation of the class. See side note below.

Instantiate the class at the very top of your main class:

object Main extends App {
  val arguments=new SampleArgs(args)
  ...
}

NOTE: this will throw MissingArgumentException if a required argument is missing!

NOTE2: if a user provides --help as arguments (this is a reserved name!), this will throw PrintHelpAndExit exception with message containing usage instructions. You might leave it as it is, but if you prefer not to print a standard stack trace you should catch it and print instructions in a more friendly manner. Yes, this could be done by packing parse method result in e.g. Either, so feel free to modify it if you find exceptions less elegant.

Now you can access values by:

// assignment to a typed val - scala will do implicit type conversion for you:
val inFile:String=arguments.inputFile

// assignment to a val using explicit type conversion:
val numOfIterations=arguments.iterations.value
// or using apply():
val numOfIterations=arguments.iterations()


// assignent to a optional val with implicit type conversion
val outFile:Option[String]=arguments.outputFile
// or explicitly
val outFile:String=arguments.outputFile.optValue.getOrElse("/tmp/any_folder/out.bin")

This is safe (i.e. no error will be thrown) but you have to deal with options yourself.

It was a bit complicated to accept Boolean parameters without values. This will not work if you mix boolean parameters with positional ones, particularly when a boolean parameter is provided without value and right before named ones. This will confuse parser and cause incorrect behaviour.

I tend not to mix parameter types (named/positional) if possible.

Side note 1 on immutability

You may have noticed that I use some private vars to keep a name, position and actual argument. The problem is this: I need to set contents of a field on the basis of its actual name. In order to use reflection the field must be instantiated, i.e. I cannot set vals afterwards.

Note that all this happens during, or rather just after, instantiation of the class. It means that there is no access to vars at runtime since they are set only once. This makes them a bit like lazy vals.

Side note 2 on testability

As you use runtime arguments, you may also want to be using similar construct for testing. You may create separate test class with arguments defined not as parser for array of strings but even more verbose.

Say you define a trait with application-wide arguments:

trait SampleArgsBase {
  val inputFile:Argument[String]
  val algorithm:Argument[String]
  val iterations:Argument[Int]
  val outputFile:Argument[String]
  val verbose:Argument[Boolean]
}

Note: there's no parse method.

In your code you pass around a trait not the concrete type. You may make it implicit as well as other dependencies that you have to inject.

At runtime, you may use the above class:

class SampleArgsRuntime(args:Array[String]) extends ArgsAsClass(args) with SampleArgsBase {
    //same body as above

    parse()
}

Note: You have to use parse() here.

For testing, you may define a separate class in a different way:

class SampleArgsTest extends SampleArgsBase {
  val inputFile:Argument[String]=Argument.static("/test/files/in.bin")
  val algorithm:Argument[String]=Argument.static("fast")
  val iterations:Argument[Int]=Argument.static(2)
  val outputFile:Argument[String]=Argument.static("/test/files/in.bin")
  val verbose:Argument[Boolean]=Argument.ignored
}

This is just a different convention. By using static you explicitly set configuration values and more importantly by using ignored you show that this parameter is not used during testing. Since there is nothing to be parsed, you don't need the parse method.

I find this more verbose.

Side note 3 on adding new types

Depending on personal preferences you may find useful arguments of very specific types, e.g. Double from scientific notation. To add new type all you have to do is:

  1. Define a new parsing method in RawArgument class, e.g. asYourType. It should convert the textual value into option of the type you're adding.
  2. Add another implicit object to RawArgumentConverter, which overrides toValue method. It should invoke asYourType method (you need flatMap here since you must map option to option).

This would look like:

case class RawArgument(key:Either[String,Int], value:String) {
    ...
  def asDouble:Option[Double]=
    Double.toDoubleOption
}

object RawArgumentConverter {
    ...
  implicit object RawArgumentToDouble extends RawArgumentConverter[Double] {
    override def toValue(rawArgument: Option[RawArgument]): Option[Double] = rawArgument.flatMap(_.asDouble)
  }
}

You could argue that value conversion and exposing implicit object could be combined RawArgumentConverter. Still I prefer to separate these two responsibilities, even if it generates some more boilerplate.

Now you can define an argument as double:

class SampleArgs2(args:Array[String]) extends ArgsAsClass(args) {
  val doubleValue:Argument[Double]=Argument.required
  parse()
}

About

Command-line arguments reader to a class with respective fields

Topics

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages