I have recently been talking online about the need to improve the importance of my .NET code. I have successfully created high-performance solutions using purely F# and will continue to. There comes a point, though, when you are fighting the system you are building on so much that you end up looking for another tool.
I have long been a programming language geek and have enjoyed learning about other languages and the problems that motivated them. Early in my career, I prized productivity more than anything else, and to be honest, I still do. What makes a language productive is how well it fits your problem space. F# is a high-productivity language for many problem domains and will remain in my toolbox for a long time. I have reached the point where I need to push the performance of the product I am building, and the .NET platform doesn’t fit anymore.
While exploring other languages, I landed on Odin, a language developer by Bill Hall (aka Ginger Bill). Odin appears to be a language purpose-built for the domain that I am working in. As I have started diving into Odin code, I thought it would be valuable to provide a compare and contrast of the two languages. Spoiler: neither language is superior to the other. To judge a language fairly, you must consider the problem space it is trying to address. F# and Odin have different goals, which I want to highlight.
Fortunately, the .NET runtime and F# make it easy to call Odin from F# so a solution doesn’t have to be completely one or the other. The current project I am working on has an F# shell and an Odin core. We leverage both languages for what they are best at.
The Goal of F#
From my understanding, the goal of F# was to provide a high-productivity language in the ML family for .NET. If you want to know the history, check out the HOPL paper. It gives a detailed history of how the language came to be and the forces that shaped it. I think of F# as a language designed to get work done; it just happens to have a functional flare to it. It leans toward immutability and composing solutions by defining types and functions that operate on those types. In 99% of scenarios, you can achieve excellent performance with F#. It wants to simplify developers’ lives so it has safe and robust defaults. Being built on top of .NET it has Garbage Collection, which is a boon to developers who don’t want to care about managing memory (more on this later).
F# has a terse syntax that provides minimal noise and gets out of your way. It also has access to the enormous .NET ecosystem which can help bootstrap any project. I still believe F# is an excellent first language, and if I need to “just get something done,” F# is still my preferred tool. It has excellent tooling, and debugging is straightforward due to the language’s design dragging you down the pit of success.
The Goal of Odin
Odin started in July 2016, born out of Bill Hall’s frustration with the C and C++ languages. I recommend this video for a quick introduction to the language. Bill’s background is in Physics, and he has many years of programming experience. He’s currently employed by JangaFX, which develops real-time simulation engines for movies and video games. I won’t put words in Bill’s mouth, so the following is my synopsis of Odin.
I see Odin as one of the C-Next languages. It is very purposefully not intended as a successor to C++. I think this is an important distinction. Many of the other languages I have looked at appeared to be more interested in becoming the next C++. I would put Rust, Jai, Nim, and Vale in this category. Odin is conservative in its feature set; it just does it very well. This may not seem impressive, but there is something incredibly pleasing about using an extremely well-made tool.
Odin is intended as a language for high-performance computing. You can see that in its extensive support for memory allocators and controlling memory layout. Including the Struct of Array layout using #soa
and the using
keyword makes it easy to play with how your data is laid out and accessed.
Odin intends to make writing high-performance code joyful, and I can tell you it succeeds wildly after working in it for a couple of weeks.
Let’s now start comparing these languages and discussing their differences.
Defining Types
Records and Structs
Both F# and Odin make it easy to define types. The most significant difference is that F# defaults to creating objects on the heap, while Odin defaults to the stack. Let’s define a Chicken
type in both languages. The default for F# would be to define Chicken
as a Record, an immutable and heterogenous collection of fields allocated on the Heap.
F#
|
|
Odin doesn’t have the idea of a Record. Instead, the closest thing we have is a struct
which is allocated on the Stack if you define it locally but can also be allocated on the heap. Odin has extensive support for custom memory allocation which gives you fine grained control over where memory is laid out. We’ll come back to this when we talk about Garbage Collection in a future post. For now, here’s how to define a Chicken
in Odin.
Odin
|
|
You see that it is nearly the same. To define Chicken
as a struct in F#, you need to add the [<Struct>]
attribute.
F#
|
|
Note that the F# Chicken
is immutable. You cannot change the values of the fields. If you want to do that, you need to add the mutable
keyword.
F#
|
|
This is unusual in F#. Generally, if you wanted an updated version of a Record, you would use the with
syntax to update the appropriate fields. Here’s an example of creating a new instance of Chicken
with a new name value of “Drumstick”.
F#
|
|
If you add the mutable
keyword to each of the fields of Chicken
you now have something equivalent to the Odin Chicken
.
Let’s compare the syntax for creating an instance of Chicken
in F# and Odin.
F#
|
|
Odin
|
|
You will see in this case that the F# is a little more verbose. This is because Odin allows you to create instances of struct
s by providing arguments in the order they are declared. If you wanted to be more explicit, you could use the field names as well:
Odin
|
|
If we use the default let
binding in F#, we still won’t have quite the same behavior as Odin. Though the fields of the F# Chicken
type are mutable, we aren’t allowed to change them without taking a mutable binding. Let’s do the Odin equivalent in F#.
F#
|
|
We now have something that behaves exactly like the Odin Chicken
. Now, is the Odin version of the F# version better? Before you even think about that question, you must ask, “Under what criteria are we judging these languages?" In a later post, I will dive more into this. For now, I want you to observe what F# makes easy and what Odin makes easy. That will inform the design intention behind these languages and, therefore, how we should judge them.
I showed you how to express the F# Chicken
with the same functionality as the Odin Chicken
. Notice that I did not do the inverse. From my limited knowledge, it’s impossible to express the idea of immutable types in Odin, which makes sense given the roots of the language. Let’s move on to Discriminated Unions.
Discriminated Unions
Both F# and Odin have algebraic type systems, meaning they have “AND” and “OR” types. An F# Record or Odin struct
are AND types. You can think of a Chicken
as being a “Name” and an “Age” and a “Size”. An OR type is one where you have a type which could be a Foo or a Bar or a Baz. Let’s say we wanted to define a type Value
which could be an int
, float
, or string
. In F#, we would use a Discriminated Union (DU) like so:
F#
|
|
An F# DU requires that each case of the DU have a label and a type associated with the label. The labels in this example are “A”, “B”, and “C”. The value associated with “A” is an int
, the value associated with “B” is a float
, and the value associated with the label “C” is a string
. Odin’s DUs are just called Unions and are similar to F# but drop the Label. Only types make up the cases of a Union in Odin. Here is what Value
looks like in Odin:
Odin
|
|
The lack of labels means that the following F# cannot be expressed in Odin.
F#
|
|
The OtherValue
type has three possible cases, but the underlying type is always an int
. This can’t be written in Odin since the Union type in Odin only comprises types. This may sound bad, but Odin has another trick: the distinct
types. Odin is described as a distinctly typed language, which means that just because two types share the exact same underlying representation, they should not be considered equivalent types. What makes this easy is having a distinct
keyword. distinct
allows you to define a new type with the same data representation as another type, but the compiler does not consider them interchangeable. Let’s see this in action.
We return to our F# example and say that the different cases correspond to IDs in our domain. Instead of defining our union as OtherValue
, we are going to say it is a BirdId
which could be holding the Id for three different kinds of birds: Chicken, Turkey, and Goose.
F#
|
|
How do we do something like this in Odin? We use the distinct
keyword.
Odin
|
|
Even though all of these IDs use an int
as the underlying value, we have told Odin that it should consider them distinct. I have found the combination of distinct
with union to solve any issues with not having labels for the different cases. In F#, you would likely use a single-case DU to wrap the int
values to differentiate them.
Next Time
In the next post, we’ll discuss functions and procedures and how F# and Odin differ. I hope by the end, you’ll have a deeper appreciation for both languages and understand how their different design goals formed their evolution. You’ll understand their strengths and weaknesses and see how being able to use both of them can lead to incredibly powerful solutions.