[edgyswingset]

F# is a phenomenal language. Not that C# is bad (it's fantastic!), but the more I use F# the more I feel it's a superior language for most tasks. First-class .NET support is the real kicker, since you can do anything from DB interaction to writing MVC controllers. Integration with existing C# projects is seamless too, so it's really just a joy to work with.

[murphm8]

Can you recommend any resources to learn about F#?

I'm currently working on a big C# codebase and it would be interesting to see if F# would make some of it cleaner.

[edgyswingset]

Absolutely. One of the best starting places would be the Why Use F# series:

http://fsharpforfunandprofit.com/series/why-use-fsharp.html

It compares and contrasts C# and equivalent F# code for relatively simple, but common-in-the-real-world examples, while introducing some functional constructs. I'd also recommending reading the "Thinking Functionally" series.

After that I'd recommend skimming some of the topics on the F# wikibook:

http://en.wikibooks.org/wiki/F_Sharp_Programming

And then I'd begin with rewriting some components in your existing project while continuing reading through that book and other online resources.

Me and a coworker also rewrote ~600 line C# module into a working module in F#, along with some interop POC here: https://github.com/cartermp/CSharpToFSharp

It's the product of a little over 20 hours of development across two people new to the language (and thus has some warts...), so take it as a grain of sand. Uses MS Unit Testing framework for F# (available via NuGet).

[MichaelGG]

F# will almost definitively help any C# code. Even if you use F# as a "better C#" you'll see improvements. Just removing excessive type annotations is a nice step up in clarity.

F#'s benefit will come as a bunch of tiny improvements, "programming in the small" as they call it. For instance:

  let xs = [ while r.Read() do yield r.GetInt 0 ]

In C#, it's:

  var xs = new List<int>();
  while (r.Read()) { xs.Add(r.GetInt(0)); }

Or:

  let f x = 
    let x = try int s with _ -> -1
    x * 3

In C#, it's uglier. First because try... isn't an expression. Second, because you cannot shadow variables by rebinding them, so you always have to keep "old" vars around and in scope, and cannot reuse handy variable names:

  int f(string x) {  
   int x1;
   try { x1 = int.Parse(s); }
   catch { x1 = -1; }
   return x1 * 3;
  }

Or:

  let x = 
    use db = new DB()
    db.GetX()

In C#, you've got to declare x outside a block:

  SomeType x;
  using(var db = new DB()) {
    x = db.GetX()
  }

Which is more annoying than it might seem. How much nicer is it to be able to create a new scope at any point in a function, and return a value out of it cleanly?

These are by no means a complete or even important showcase of C#'s lacking. Just a few quick thoughts off the top of my head. In general, every time I'm writing C# code, I keep realising how things would be much more concise if I was in F#.

[CmonDev]

F# is a nice language, but not a very necessary one (especially given the lack of tooling).

1) Simply write a quick extension method (not ideal but not a reason to switch languages):

    var xs = r.YourExtensionMethod<int>(rr => rr.GetInt(0)).ToList();

2) Simply use the appropriate built-in method:

    int f(string x)
    {
        int number; /*Will be not necessary in C#6.*/
        return Int32.TryParse(value, out number) ? numeber * 3 : -1;
    }

3) It's a matter of taste. Not a bad feature, but not a killer one either.

[MichaelGG]

None of them are killer features. Nor is pattern matching, assorted comprehensions, array slicing, binary literals, bytestrings, tuples, active patterns, records, type inference, immutability, shadowing, nested functions, custom operators, typechecked printf, type providers, workflows[1], sum types, agents, etc. etc.

But it sure all adds up.

Switching for a codebase might not be a wise move for many reasons. But writing new code doesn't have those excuses.

Actually workflows are the closest to a "killer" feature but C# took the most popular, async, and hard-coded it in.

[CmonDev]

All those words surely do sound impressive when you list them all together just like that. But you can remove at least pattern matching and records, as they are coming to C#. At least some of them is a matter of taste (e.g. nested functions , global inference and custom operators - the latter two can reduce readability IMHO) etc. For me however the killer feature of C# is ReSharper and other tooling. There is no point to argue, F# is a great language, but C# is a great one as well, no need to write code samples to misuse the language and make it look bad. I considered using F# for physics computations (to leverage the units of measure) but unfortunately it provides no solution for marking external types with them (which means I cannot integrate it with e.g. MonoGame). I still have a hope for finding a good use case for F#.

[MichaelGG]

Pattern matching in C# would be a rather big departure and change. And then to match F#'s level of deconstruction, that'd be neat. C# can become F# if it wants, they just have to add in features. They've been hesitant to do so based on limitations of the compiler codebase, as well as fears that C# might get too difficult for mediocre programmers.

I'm not sure how new List<string> { "a", "b" } is more readable than new List<_> {"a", "b"), but hey, sure, if you want to argue type inference is bad, go for it. F# also lacks loop constructs like break/continue.

I don't feel I've misrepresented C# at all to make it look bad. I've written a lot of C# code and a fair amount of F# code. Line by line, char-by-char, expression-by-expression, F# is simply much less code. Those examples are just things off the top of my head, from real codebases.

C#'s alright, because the competition (like Java) is laughable. So in that sense, it's "great". In absolute terms, it doesn't measure up (and this isn't a secret, bit-by-bit C# adopts features F# proved out.)

Units-of-measurements are implemented via erasure, yes. How would you represent float<m> externally to make it available to common types, but without losing performance? The compiler consuming them would need to be aware of it. And at runtime, you certainly don't want extra overhead, and I don't think the CLR has any efficient way of exposing primitives with additional type info. It's an unfortunate tradeoff.

[slambam]

I've got a safaribooks online account, and found "Real World Functional Programming with examples in F# and C#" the most useful of the f# books I've read. The author has a blog, http://tomasp.net/ But I got something out of all of the books.

[pjmlp]

The book is available for free from Microsoft.

http://msdn.microsoft.com/en-us/library/vstudio/hh314518%28v...

[CmonDev]

Don't do it if you are used to good tools like ReSharper. Wait for Roslyn and it's syntax sugar instead.

[bratsche]

Roslyn doesn't compile F# currently.

[CmonDev]

Exactly, don't use it - no proper automated refactoring. Not from MS not from third parties. And there are no plans.

[Rapzid]

Are you or is anyone aware of how the F# support for monodevelop(Xamarin Studio) and mono is coming along? I've had good experiences making console and service projects in Xamarin(monodevelop) and would like to give F# a go.. But I just don't see chaining myself to the VisualStudio/Windows environment.

[akoeplinger]

F# is supported in Xamarin Studio: http://developer.xamarin.com/guides/cross-platform/fsharp/fs...

[Touche]

I disagree; C# is bad by comparison. I cannot think of a single reason to use C# over F#, can you? F# is essentially C# but with better defaults and much more easily extensible.

[Hakeashar]

- Writing a WPF app in F# is a real pain (no partial classes, no nice MVVM framework support like Caliburn.Micro). C# is way better there.

- Accessing the DB is easier in C# (just select a nice micro-ORM), type providers do not always cut it.

- When using any kind of library that relies on anonymous types (e.g. Dapper for database access).

- Tooling is still not complete, not being able to create directories in F# project (!) from IDE level (and even if you edit .fsproj, it can be a bit buggy) is really annoying in both web and desktop apps.

- If you want to use some of object-oriented features e.g. co/contravariance in interfaces, protected access modifier.

- (admittedly, that does not happen often) no unsafe context/keyword

That being said, I agree that F# has a very sane set of defaults and I really like using F# for library work.

[MichaelGG]

F# has unsafe capabilities. Look at NativePtr and friends. And since you can make operators, you can end up writing fairly concise code.

The rest of the real arguments are essentially "library support". Which is a good argument, but not really about the language itself. There's no reason that WPF, DB access, etc. can't be just as fine in F#.

(C#'s limitations, like no tuples, led high-profile projects like ASP.NET MVC to do dumb hacks like "pass in an object and we'll assume each field is a parameter". If C# had been more capable in the first place, they wouldn't have resorted to such hacks.)

[Touche]

> Writing a WPF app in F# is a real pain (no partial classes, no nice MVVM framework support like Caliburn.Micro). C# is way better there.

You shouldn't be writing WPF apps at all.

[seanmcdirmid]

There is no replacement for WPF on desktop. I'm pretty sure at this point that WPF is still more popular than WinRT.

[CmonDev]

HTML5 sucks really hard after XAML :(.

[seanmcdirmid]

I don't use XAML in my WPF code.

[sandyarmstrong]

> You shouldn't be writing WPF apps at all.

Says who? Are we supposed to stop writing Windows desktop apps in C# just because the frameworks are in maintenance mode?

[Touche]

Yes. Write web apps instead.

[CmonDev]

All of the web's legacy languages are lame and not going anywhere - HTML, CSS, JS. At least WPF gives you some choice - C#, F#, VB. All without crap-trans-piling. The REST is not bad though.

[math]

I trialed F# a while back. Something I certainly noticed was not having as much help from the IDE (no resharper). The lambda notation also annoyed me (by comparison to C#). Yeah, I know it's no big deal and I understand the trade-offs being made.. but it still annoyed me. Pattern matching is one of the draw cards - but I see that is coming in C# soon.

[edgyswingset]

Did you give F# Power Tools a go? It gives you a lot of what is missing out of the box. It's still not as great as C# IDE support, but for most things it gets the job done.

[MichaelGG]

While this may come across as sour grapes, I find generally I need less IDE help in F#, since I'm not typing as much boilerplate. For instance, almost all type annotations are inferred (20:1 in my own empirical study), so there's a ton of code I simply don't need to type.

In general, I find writing in C#, I'm going to need 50-200% more code for "business logic" type code that doesn't particularly benefit from F#. That is, using F# as a better C#. And this is after C# hacked in async - before that, there's no comparison if you need async style code.

If C# added pattern matching, tuples, active patterns, type inference, everything-as-an-expression, nesting, more comprehensions, etc. etc., then yes, C# would be competitive. And with all the resources C# gets, the IDE would be far better than F#, sure.

[CmonDev]

Sadly F# had inherited all the OCAML luggage, including the arcane academic ML syntax.

[nbevans]

This. F# is superior in every way. The constant stream of articles about C# 6.0 are getting ridiculous when .NET already has a language that is a natural successor.

C# is for legacy and large corporate red-tape-encumbered enterprise projects only, as far as I'm concerned.

[seanmcdirmid]

C# is a stronger OO language than F#, if you prefer OOP over FP. It also has a lot of functional features that allow you to dip into FP when that is convenient. Not as great as Scala, but definitely a huge step up from Java.

[MichaelGG]

What in particular do you mean by this? "Protected" accessibility, automatic mutual recursion for everything, nested classes -- and a few other edges? Is that enough to say C#'s a "stronger OO" language? Perhaps strictly speaking, as in, it has those few features more. But in practise?

But then what of C#, which lacks object expressions? Which forces you to declare a type, often an interface, just to implement a member or two?

[seanmcdirmid]

Small potatoes. If I have an OO design, I'm not going to be able to use modules for that, or even structural typing. If you do a lot of class programming, F# doesn't make sense.

[kvb]

I don't think that's fair at all; F#'s integration with the .NET type system is one of its best features. I find that creating class types (and instances) in F# is usually easier than doing so in C#. And with object expressions, you can even program in an arguably purer object oriented style, declaring only interfaces but never concrete class types.

[MichaelGG]

I'm not sure how this is relevant. F# handles creating classes just fine. It's come a long way from the OCaml crosscompiler that you might be remembering.

[pjmlp]

Besides all the issues pointed out by Hakeashar.

In the enterprise C# is the way to go. You won't find many Fortune 500 companies, with offshoring projects, going F# instead of C#.

For many 9 - 5 developers, C# is already too complex, many of each are ex-VB developers.

F# lacks the Visual Studio tooling support for the typical enterprise projects workflow and frameworks.

[72deluxe]

If C# is already too complex for 9 to 5 developers, how does it earn more than a C++ developer?! The mind boggles.

[pjmlp]

In which company?!

Usually it is C++ > Java > C# > VB.NET > .... in the Fortune 500 world set I work on.

Not to mention the C++ salary rates in High Performance Finance and High Performance Computing in general. I doubt any C# developer would get those rates.

[72deluxe]

Here in the UK I see jobs being advertised in Birmingham for big money for C# but C++ is always lower.

It is a pity. I did do some C# work but preferred writing C++ hence I do it as a day job now (and a sideline for fun too!).

Even Java jobs are offered for more than C++.

[pjmlp]

Thanks for the clarification, what about London (city)?

Clearly not all places are equal. :)

I did C++ for many years, was on the C++ side since day one on the C vs C++ wars, since 1993.

Nowadays mostly Java, C# on the job. C++, among others, mostly on hobby projects.

Given my Turbo Pascal initial background I still prefer more memory safe systems languages though. Although modern C++ surely feels good.

[edgyswingset]

Sometimes a procedural approach is better for certain (in my opinion, limited) problems. I'd rather use a procedural language for those cases.

Also, the .NET libraries we end up using were, to my knowledge, written in C#. Sometimes interacting with them feels awkward if done functionally.

[MichaelGG]

Fortunately, F# does OO and procedural pretty well, so you shouldn't feel pressure to use a functional style if it doesn't fit.

I noticed that was something certain folks tended to do, they'd insist on creating an expression as a series of folds and maps, even if it had a very simple expression otherwise (as a list comprehension or simple for..in loop). Cast those cares aside, and write what is most useful for your use case.

Sometimes though, you do want break/goto/continue, although in most cases I think such code is better off in C or Rust.

[DrJokepu]

For me, writing procedural OO code in F# always felt awkward and inconvenient. This is a problem because most of the .NET runtime was designed with procedural OO code in mind.