DEV Community

Brandon Weaver
Brandon Weaver

Posted on • Updated on

New in Ruby 3.2 - Data.define

Victor Shepelev (Zverok) has just landed an extremely useful feature in Ruby, Data.define. You can find the merge here:

https://github.com/ruby/ruby/pull/6353

...and the Ruby discussion here:

https://bugs.ruby-lang.org/issues/16122

So what is it and what does it do? Well that's what we're going to take a look into.

What it Does

Succinctly put Data.define creates an immutable Struct-like type which can be initialized with either positional or keyword arguments:

Point = Data.define(:x, :y)
origin = Point.new(0, 0)
north_of_origin = Point.new(x: 0, y: 10)
Enter fullscreen mode Exit fullscreen mode

Note: This does not inherit from Struct

...and because the API supports pattern matching you can very much do things like this still:

case origin
in Point[x: 0 => x, y] # rightward assignment, pattern matching
  Point.new(x:, y: y + 5) # shorthand hash syntax
else
  origin
end
Enter fullscreen mode Exit fullscreen mode

It can also take a block for creating additional methods if you'd like, much like Struct:

Point = Data.define(:x, :y) do
  def +(other) = new(self.x + other.x, self.y + other.y)
end
Enter fullscreen mode Exit fullscreen mode

Then Why Not Struct?

You can absolutely use Struct still. Given that why should you use Data instead?

Because it's stricter, and the values it produces are immutable. With the advent of more functional patterns in Ruby this can be a very useful thing.

Consider data passing in Ractors which requires immutable state, this becomes incredibly useful for value objects and message passing for what may inevitably be our next generation of Puma and other web servers.

Why not dry-rb?

Interestingly the dry-rb folks are very pragmatic about things like this. In fact some of their core folks are already talking about how to wrap this new feature into their immutable structs to build on top of it, rather than in parallel:

https://twitter.com/solnic29a/status/1577683251086376963

They did the same with pattern matching, and it's great to see how we build on top of what the language formally adopts.

What Do You Think of It?

Personally? I like it. I plan to use it a lot in the REPL whenever I need a quick data type I can match against without pulling out a full Struct for it as often times I don't intend to mutate it and the keyword_init flag can be a bit pesky to remember.

It reminds me a lot of case classes from Scala:

case class Book(isbn: String)

val frankenstein = Book("978-0486282114")
Enter fullscreen mode Exit fullscreen mode

...and of data classes from Kotlin:

data class User(val name: String = "", val age: Int = 0)
Enter fullscreen mode Exit fullscreen mode

So this is not a novel concept, but a useful one for Ruby to adopt.

Closing Thoughts

While the new language features have certainly slowed, in accordance with Matz's focus on tooling, we're still seeing some interesting movement in the language core. Of course there's a lot of fascinating tooling too, and I'm thrilled to see those coming to fruition, and perhaps soon I'll write about them as well.

In the mean time thanks to Zverok for seeing this through, and everyone who participated!

Top comments (9)

Collapse
 
zw963 profile image
Billy

This Data.define stolen from Crystal. The key different with Struct is:

Struct is just a conveniently way to create a class, it is reference, create on heap.
Data.define is a value, create on stack.

Collapse
 
janko profile image
Janko Marohnić • Edited

I love it, I was excited to see this alternative implementation of Struct. However, you noted in this article that the Data class inherits from Struct, whereas I understood from the PR that it inherits from Object. How did you find that out? The docs also seem to indicate that Data is an alternative implementation:

See also Struct, which is a similar concept, but has more container-alike API, allowing to change contents of the object and enumerate it.

Collapse
 
tjad profile image
Tjad Clark

Yeah, and doing a call to ancestors shows that it doesn't inherit from Struct

Image description

Collapse
 
baweaver profile image
Brandon Weaver

Article amended to reflect that. In the past it was going to loosely inherit it so I must have missed that fork

Thread Thread
 
tjad profile image
Tjad Clark

In the source code, it is written in the struct.c file - there is some relation

Collapse
 
tjad profile image
Tjad Clark

Based on this post, I was playing around with Data, thinking that it would be a good fit for Ractor. The idea that Data is immutable is sound, however the reality is that it may be misleading, as to fully gain advantage of Ractor communication, the object needs to be fully immutable (or deeply immutable). So by default Data instances will still be copied to each instance of Ractor, unless Ractor.make_shareable was called to deep freeze the data instance.

Collapse
 
bdewater profile image
Bart de Water

It would be useful to have a with method to clone an instance with new data (like Sorbet's T::Struct): bugs.ruby-lang.org/issues/19000

Collapse
 
martinos profile image
Martin Chabot

The only thing that bugs me is that is that there's no way to set default values.

Collapse
 
niku profile image
niku • Edited

You can override initialize method to set default values.

Point = Data.define(:x, :y) do
  def initialize(x: 0, y: 1)
    super
  end
end

p1 = Point.new
p p1 # => <data Point x=0, y=1>

p2 = Point.new(x: 3, y: 4)
p p2 # => <data Point x=3, y=4>
Enter fullscreen mode Exit fullscreen mode

docs.ruby-lang.org/en/3.2/Data.htm...

Note that Measure#initialize always receives keyword arguments, and that mandatory arguments are checked in initialize, not in new. This can be important for redefining initialize in order to convert arguments or provide defaults:

hope it helps.