Guidelines to design a C# library well usable from F#

Question

I just want to point out that this is question is not the reverse of

Best approach for designing F# libraries for use from both F# and C#

Here I'm not asking how to design a functional library written C# to be used in both worlds.

I'd like to know good practices on what design choices embrace or avoid to get a reasonable compromise for make this library usable from F#.

Practices like (for example):

• keep object hierarchy as simple as possible

• avoid mutating state of objects, but return new ones

• etc...

Anyone that already done it, can share it's experience?

Side note

It's interesting note this OSS project, IronJS. Yes, it's written in F#, but the author expose two specialized host IronJS.Hosting.FSharp and IronJS.Hosting.CSharp.

Answer 1

Imagine one day you would like to rewrite your C# library in F# for better usability. Here are the paths you are likely to take:

I focus on the path "Imperative C# --> Functional C# --> Functional F# --> Idiomatic F#". The more functional your C# library is, the more usable your library is in F#. Functional style helps increase composability and is closer to idiomatic F# code. Along these lines, you can:

• Embrace immutability by default principle. If you don't know whether you need to update a field/property later, just mark it readonly first.
• Follow expression-based and declarative programming style. LINQ operations are good examples.
• Use immutable collections or mutable collections in an immutable way. With the introduction of C# immutable collections, it should be easier than ever before.

The picture above is taken from F# for fun and profit's Porting from C# to F# series. They are very helpful; knowing how C# concepts are expressed in F# will improve usability of your library.

It's hard to avoid C#'s object-oriented features. Remember that F# type inference doesn't work very well with these features. Along the line of keeping object hierarchy simple, you should reduce number of member overloads. A big number of member overloads will easily confuse F# type checker. Moreover, it doesn't hurt to distribute a thin F# wrapper with your C# library. Certain things you need to do are turning some methods into module functions and creatingActive Patterns to decompose object hierarchy.

Answer 2

Interop with existing .NET libraries was a major design goal of F#, so there aren't any constraints on the libraries to be consumed.

That said, because of F#'s stricter typing, there are some patterns that result in slightly clunkier code. The builder pattern is one.

var bldr = new StringBuilder();
bldr.Append("abc"); //ignoring return value

vs.

bldr.Append("abc") |> ignore //must be explicitly ignored

But this is easily worked around using an extension method or let-bound function. Bottom line: interop is one of F#'s strengths and greatest achievements.